Laboratory of Molecular Biophysics
Laboratory Journal 2002
Mark S. P. Sansom

Return to: Contents.

Mark S. P. Sansom

Structural bioinformatics of membrane proteins

Prof. Mark S.P. Sansom; Dr Phil Biggin; Dr Hyunji Kim; Dr Joanne Bright; Dr Carmen Domene; Dr Marc Baaden; Dr Alessandro Grottesi; Richard Law; Charlotte Capener; José Faraldo-Gómez; George Patargias; Oliver Beckstein; Yalini Arinaminpathy; Pete Bond; Jonathan Cuthbertson (with Dr. Declan Doyle); Sundeep Deol; Jeff Campbell; Katherine Cox; Jennifer Johnston; Robert d'Rozario; Christoph Meier; John Holyoake; Andy Pang; Dr Andrew Horsfield (UCL); Dr Reinhart Reithmeir (U Toronto).

1. Membrane proteins: biological importance

The overall theme of work in this group is to employ a range of computational techniques (structural analysis, modelling, simulations) to membrane proteins. Membrane proteins play keys role in cell biology e.g. as ion channels, drug receptors, and solute transporters. Indeed, it has been estimated that ca. 30% of genes code for membrane proteins, and that ca. 50% of potential new drug targets are membrane proteins

Structural bioinformatics: why?
Despite the central importance of membrane proteins, the number of high resolution structures (from X-ray diffraction and more recently from NMR) remains small, amounting to only ca. 45 membrane proteins. Furthermore, those membrane proteins for which structures have been determined are mainly bacterial. Very few structures have been determined for mammalian (let alone human) membrane proteins. Of course, given sufficient time and the ongoing advances in experimental techniques, the number of structures will rise. However, given the paucity of structural information we wish to use computational methods to extrapolate from the available experimental data. There are two ways in which computational methods are valuable. Firstly, they can be used (via molecular dynamics simulations) to probe the relationship between static/average structures and the resultant dynamics of the protein, which in turn plays an important part in membrane protein function. The second approach, which has been extensively employed for globular proteins, is to analyse known membrane protein structures in order to reveal underlying principle of membrane protein structure and stability, thus (ultimately) aiding structure prediction.

2. Molecular simulations: from structure to dynamics

Figure 1 Outer membrane protein in a lipid bilayer ...more

Molecular simulations provide us with a powerful tool for (membrane) protein structure analysis. The most common form of such simulations, molecular dynamics (MD), corresponds to a Newtonian simulation of membrane protein dynamics using an empirical forcefield. Thus, whilst an X-ray study provides an average structure of a membrane protein at ca. 100 K in crystal, MD simulations provide information on protein dynamics at 300 K in a water plus membrane environment (Figure 1). It is not sufficient simply to be able to perform such simulations. The challenge is to relate structural dynamics to biological function

Three key directions
We are extending our simulation studies of membrane proteins in three directions. Firstly, by increasing the depth of our simulations we wish to increase our understanding of e.g. the physics of ion permeation through channels. Secondly, by increasing the range of simulations we aim to perform comparative studies of the dynamics of membrane proteins. Thirdly, we wish to increase the complexity of simulations e.g. to examine structural dynamics in multi-component transport systems.

3. Ion channels & pore-like transporters

Simulation and modelling studies are being used to explore structural dynamics in a wide range of ion channels and related pore-like transporters [1, 2]. The biological roles of ion channels are quite diverse. Ion channels are found in all cells but are of especial importance in neurones. The opening/closing of channels with different ion selectivities give rise to changes in voltage across cell membranes. Such changes are important in generation and propagation of action potentials and in synaptic neurotransmission.
A number of our studies have focussed on potassium channels. These studies are based on the structures of two bacterial K channels: KcsA and, more recently, MthK. The overall aim of these studies is to progress from structure via simulation to an understanding of the role of structural dynamics in channel physiology. This presents a number of computational challenges, not the least of which is the wide range of timescales spanned from potein backbone and sidechain dynamics (on the ns timescale) to ion permeation (10 to 100 ns) to channel gating (µs to ms)

Permeation mechanisms.
We have continued to apply MD simulations to understanding ion permeation through KcsA [3-5]. In particular, recent studies have focussed on channel behaviour vs. ionic species, comparing the behaviour of K⁺, Rb⁺ vs. Cs⁺ ions in the selectivity filter (Domene & Sansom, in press). The results suggest that as one progresses from K⁺ to Rb⁺ to Cs⁺, decreasing permeability correlates with increasing distortion of the filter to accommodate the ion. This set of simulations has also been used to examine the degree of filter flexibility during permeation, and the robustness of simulation results to changes in simulation protocol. These studies also emphasise the need for improved modelling of close ion/protein interactions, for example to include some consideration of electronic polarisation. To obtain accurate energetics for understanding ion channel selectivity we may need to move to large scale hydrid (i.e. QM/MM) simulations. This will need high performance computing (HPC - see below) resources.

Figure 2. Water density in a nanopore vs. pore radius ...more

Figure 3. Simulation of Tc1 Toxin from Tityus cambridgei: trace radius proportional to the atomic displacements ..more

Water in nanopores
In addition to the complexities of permeation through real channels, we have continued to explore the underlying physics using a simple model of a nanopore. Previous studies [6] had revealed that changes in nanopore dimensions and polarity could result in pore 'gating' to water permeation. More extended (50 ns) simulations have revealed unexpectedly complex behaviour of water in nanopores (Figure 2; Beckstein & Sansom, in press). In particular 'pulses' of water enter/exit such pores on a multi-nanosecond timescale. This has enabled a detailed analysis of the thermodynamics and kinetics of water entry and exit. The dynamic behaviour of water with simple peptide models of ion channels has also been explored in extended (c. 15 ns) simulations. In these studies of channels formed by the M2

peptide [7] the dynamics of the pore and of the water within the pore appear to be coupled.

K Channel diversity
From a biological perspective, we have employed the structures of KcsA and of MthK to model K channel gating. KcsA provides a template for a closed state of the K channel core, and MthK provides a template for an open state. Steered MD simulations have been used to explore the closed to open transition in KcsA [8]. KcsA has also been used as a template for modelling and simulation studies of more distantly related channels, including inward rectifier (Kir) and voltage-gated (Kv) potassium channels, and the channel domains of ionotropic glutamate receptors from both bacteria (GluR0) and mammals (GluR2; see below) [9]. The latter are rather more distant homologues of KcsA.
We have continued our modelling and simulation studies of human Kir6.2 (a channel involved in regulating insulin release) based the structure of KcsA [10]. This provides a case study of the use of computational methods to extrapolate from bacterial to human K channels. We have shown that such studies can help to rationalise the effect of Kir6.2 mutations on the permeation behaviour of the channel [11]. Based on our experience with Kir6.2 we are developing a systematic approach to modelling and simulation of many K channel families. These include Kv channels, which are targets for many channel-blocking toxins. Comparative MD simulations of four scorpion toxins have been performed (Figure 3) [12], as a prelude to docking studies that we hope will facilitate refinement of initial Kv models.

Pore-like transporters
We have extended our simulation studies to passive pore-like transporters of the aquaporin family, namely Aqp1 (mammalian) and GlpF (bacterial). Studies of Aqp1 have established that good homology models of such transporters can be of comparable value to models based on medium resolution electron microscopy data, and that aspects of the water transport mechanism of Aqp1 are relatively robust to the exact model used in simulations. Studies of GlpF have focussed on the use of steered MD simulations to compare the permeation mechanism of urea, of glycerol and of water.
A major challenge to modelling and simulation studies is presented by the active pore-like transporters of the ABC transporter family. A combined modelling and simulation approach has been used to extend the low resolution model of a bacterial ABC tranporter (MsbA) and to develop models of P-glycoprotein, a human multi-drug resistance protein. Long (30 ns) MD simulations have been used to explore the nature of nucleotide-induced conformational changes in the nucleotide-binding domains (NBDs) of ABC transporters.

4. Receptors & comparative simulations

Conformational changes induced by ligand-binding play an important role in e.g. activation of receptor-gated ion channels. Following preliminary simulation studies of ligand-induced changes in receptor dynamics for the extracellular domain of rat GluR2 receptors [13], we have been exploring a number of techniques to establish the significance of these results. Better sampling of conformational changes induced by ligands can b provided by both long simulations and by multiple simulations. We have utilised both of these approaches, along with 'driven' dynamics methods to access longer timescales. Another approach to establishing the significance of observed conformational changes is to improve simulation sampling by exploiting evolution, i.e. by recognising that novel biology often emerges from comparisons. To this end we have been performing comparative simulations of mammalian GluR receptor vs. structurally related bacterial periplasmic binding proteins. Initial results of comparing simulations of rat GluR2 and bacterial GlnBP are promising. Both proteins reveal ligand-induced domain closure in X-ray studies. Simulation studies have shown that ligand binding reduces inter-domain motions. Such changes in mobility may be related to the gating mechanism of GluRs. The comparison of GluR2 and GlnBP suggests patterns of mobility may be conserved over a protein fold family. Further simulation studies of other homologues (GluR0, LAOBP) are underway to test this proposal.
In addition to MD simulations of ligand-induced conformational changes, we are using homology modelling and related techniques to model the structure of vertebrate neurotransmitter receptors. For example, the snail acetylcholine binding protein may be used as a template structure for modelling the ligand-binding domain of the acetylcholine receptor, enabling combined experimental and computational studies of drug/receptor interactions.

5. Bacterial outer membrane proteins

The outer membranes of Gram negative bacteria contain a large number of membrane proteins based upon an anti-parallel

-barrel architecture. Numerous structures of bacterial outer membrane proteins (OMPs) are known - both from X-ray diffraction and more recently from NMR studies. These proteins are of some interest as potential antibiotic and vaccine targets. They also provide us with the opportunity to run extended simulations on a whole family of (distantly) related proteins, i.e. in principle we are able to develop a database of OMP simulations. Current simulations include: OmpA (a narrow pore); OmpX and OpcA (recognition proteins); OmpT & OMPLA (both outer membrane enzymes); and FhuA and FepA (complex Fe³⁺ transporters). We will continue to extend these simulations to encompass all OMPs of known structure. This is a first step towards large scale simulations of a virtual outer membrane. In turn, we hope such a simulation will provide proof-of-principle of the use of molecular simulations in systems biology.

Figure 4. Simulation of OmpA in a Crystal
...more

OmpA: dynamics vs. environment
OmpA provides us with a simple test case to simulate how the dynamics of a membrane protein alter as a function of changes in environment. We have explored the conformational dynamics of OmpA in: (i) a detergent micelle (as used in NMR studies); (ii) a lipid bilayer (an approximation to in vivo); and (iii) a crystal lattice (as used in X-ray studies). These studies have employed multiple, comparative simulations on a 10 to 30 ns timescale. The importance of studying dynamics vs. environment is emphasised by the observation that small changes in conformational flexibility can open the central pore of the OmpA molecule [14]. Significantly, OmpA in a detergent micelle shows backbone fluctuations of magnitude ca.1.5x those in a phospholipid bilayer (Bond & Sansom, in press).
We have also used OmpA as a test system to evaluate the accuracy of MD simulations of membrane proteins, via a comparison of simulated vs. experimental B-values. This was based on 30 ns simulations of a unit cell of an OmpA crystal (4 monomers/asymmetric unit). The simulated dynamics compare well with the experimental dynamics (Figure 4).

FhuA: a complex Fe³⁺ transporter
At the opposite end of the OMP spectrum, we have performed a number of simulations of FhuA, an outer membrane active transport protein, in order to better understand its transport mechanisms. This protein transports Fe³⁺ in the form of ferrichrome - a tight complex of the ion with a modified cyclic peptide. We have performed 10 ns simulations of FhuA in a DMPC bilayer with and without bound ferrichrome (Faraldo-Gomez & Sansom, in press). Comparison of water trajectories in the two simulations supports a 'plug removal' model of transport in which interactions with the inner membrane protein TonB induce a major conformational change in FhuA before the ferrichrome can move through a presumed pore formed by the FhuA in the outer membrane. Simulations of FhuA and of a number of other OMPs are also being used to explore issues of sampling of conformational changes in molecular simulations.

6. Analysis of membrane protein structures

Detailed analysis of membrane protein structures should provide insights into membrane protein structure and stability. This is turn will enable development of improved methods for prediction of membrane protein structure. Membrane protein structure prediction is likely to play an increasingly important role, given the large discrepancy between the number of membrane protein structures known (<50) and the number of genes coding for membrane proteins (ca. 10,000 in the human genome).

Transmembrane helices: analysis & prediction
Sequence-based methods for predicting the number and approximate location of transmembrane (TM) helices within membrane proteins are ca. 85% accurate. A number of such methods are known. Based on a database of ca. 300 TM helices for which high resolution structures are known, it is possible to evaluate and compare the accuracy of TM prediction methods. In particular, we suspect that analysis of prediction failures will provide insights into how one might improve such methods. Two classes of TM helices which present problems are: (i) long 'stalk' helices that project beyond the bilayer; and (ii) short membrane embedded helices that fail to span the bilayer (e.g. the P helix of KcsA).
The database of TM helices is also being used as the basis of detailed analyses of their structures and of interactions between adjacent TM helices. By analysis of TM helix/helix interactions we hope to formulate packing rules and/or potentials that can be used in prediction of integral membrane protein folds.

Figure 5. Proline induced hinge ...more

Prolines in TM Helices
Prolines are favoured in the TM helices of integral membrane proteins, in contrast with globular proteins where prolines are rarely found in the middle of

-helices. It has been suggested that intra-helix prolines may have possible roles in channel and receptor activation [15]. A structural analysis of ca. 200 TM helices in X-ray structures of membrane proteins suggested that proline could introduce an aniostropic molecular hinge (Figure 5) into the midst of a TM helix [16]. This has been explored further in an extensive series of MD simulations of model TM helices (containing a wide range of proline-motifs [17]) and of proline-containing TM helix fragments from more complex membrane proteins (e.g. the S6 helix of Kv channels [18] and the TM1 helix of the AE1 anion exchanger). These studies further strengthen the proposal of a dynamic functional role for proline-induced hinges. We are currently extending such studies to explore proline-hinged TM helices as possible switches for bionanotechnological applications. Structural studies of TM helices are being extended to TM helices containing non-proline distortions.

Lipid/protein interactions
Structural studies are also being used, in conjunction with simulations, to explore in more detail the nature of membrane protein/lipid and protein/detergent interactions. A database has been created to contain and enable interrogation of the 50 or so example of lipid/protein or lipid/detergent interactions represented in crystal structures. Analysis of the nature of these interactions is being compared with analysis of the time dependence of such interactions in membrane protein/bilayer simulations. Of particular interest are the interactions of lipid headgroups in the interfacial region with aromatic and basic sidechains on the surfaces of membrane proteins.

7. Computing developments

A number of computational issues arise in relation to biomolecular simulations in a post-genomic environment. In particular, the increase in number of protein structures being solved as a consequence of structural genomics and related developments in high-throughput structural biology pose some major challenges for simulation studies, both in terms of management and interrogation of simulation data, and in term of developing a high throughput simulation pipeline.
One approach to management of an increasing volume of simulation data is to exploit advances in GRID computing (see http://e-science.ox.ac.uk). Together with colleagues at the Universities of Southampton, Nottingham, London (Birkbeck College), Birmingham, York, and at CLRC (RAL) we are developing BioSim GRID, a prototype GRID database of biomolecular simulation data. The underlying model is one of distributed 'raw' simulation data, curated by the various simulation groups participating in the consortium. Using GRID approaches, we will develop two levels of metadata: 1^st level metadata describes the simulation data, whereas ^2nd level metadata will describe the results of generic analyses of the simulations. This approach will necessitate developing methods for federated queries on the simulation data. As may be anticipated, this project unites the skills of simulation scientists and computer scientists.
To meet the challenge of a much larger number of simulations, we need improvements in capacity computation. Much of this will exploit linux clusters, and we have expanded facilities both within LMB and at the Oxford Supercomputing Centre (http://www.osc.ox.ac.uk). Large scale simulations (e.g. virtual outer membrane - see above) will require access to capability computation facilities. In this respect, the new UK supercomputer HPCx (http://www.hpcx.ac.uk) offers exciting possibilities. By integrating such facilities within a GRID environment it should be possible to provide the infrastructure to support an overall vision whereby biomolecular simulation becomes an integral component of structural genomics within the UK.

References.

[1] Sansom, M.S.P., Shrivastava, I.H., Ranatunga, K.M. and Smith, G.R. (2000) Simulations of ion channels - watching ions and water move. Trends Biochem Sci 25: 368-374.
[2] Tieleman, D.P., Biggin, P.C., Smith, G.R. and Sansom, M.S.P. (2001) Simulation approaches to ion channel structure-function relationships. Quart Rev Biophys 34: 473-561.
[3] Shrivastava, I.H., Tieleman, D.P., Biggin, P.C. and Sansom, M.S.P. (2002) K+ vs. Na+ ions in a K channel selectivity filter: a simulation study. Biophys J 83: 633-645.
[4] Shrivastava, I.H. and Sansom, M.S.P. (2002) Molecular dynamics simulations and KcsA channel gating. Eur Biophys J 31: 207-216.
[5] Sansom, M.S.P., Shrivastava, I.H., Bright, J.N., Tate, J., Capener, C.E. and Biggin, P.C. (2002) Potassium channels: structures, models, simulations. Biochim Biophys Acta 1565: 294-307.
[6] Beckstein, O., Biggin, P.C. and Sansom, M.S.P. (2001) A hydrophobic gating mechanism for nanopores. J Phys Chem B 105: 12902-12905.
[7] Law, R.J., Tieleman, D.P. and Sansom, M.S.P. (2003) Pores formed by the nicotinic receptor M2d peptide: a molecular dynamics simulation study. Biophys J 84: 14-27.
[8] Biggin, P.C. and Sansom, M.S.P. (2002) Open-state models of a potassium channel. Biophys J 83: 1867-1876.
[9] Capener, C.E., Kim, H.J., Arinaminpathy, Y. and Sansom, M.S.P. (2002) Ion channels: structural bioinformatics and modelling. Human Molec Genet 11: 2425-2433.
[10] Capener, C.E. and Sansom, M.S.P. (2002) MD Simulations of a K channel model - sensitivity to changes in ions, waters and membrane environment. J Phys Chem B 106: 4543-4551.
[11] Capener, C.E., Proks, P., Ashcroft, F.M. and Sansom, M.S.P. (2003) Filter flexibility in a mammalian K channel: models and simulations of Kir6.2 mutants. Biophys J (in press).
[12] Grottesi, A. and Sansom, M.S.P. (2003) Molecular dynamics simulations of a K⁺-channel blocker: Tc1 toxin from Tityus cambridgei. FEBS Lett 535: 29-33.
[13] Arinaminpathy, T., Sansom, M.S.P. and Biggin, P.C. (2002) Molecular dynamics simulations of the ligand binding domain of the ionotropic glutamate receptor, GluR2. Biophys J 82: 676-683.
[14] Bond, P., Faraldo-Goméz, J. and Sansom, M.S.P. (2002) OmpA - A pore or not a pore? Simulation and modelling studies. Biophys J 83: 763-775.
[15] Sansom, M.S.P. and Weinstein, H. (2000) Hinges, swivels & switches: the role of prolines in signalling via transmembrane a-helices. Trends Pharm Sci 21: 445-451.
[16] Cordes, F.S., Bright, J.N. and Sansom, M.S.P. (2002) Proline-induced distortions of transmembrane helices. J Mol Biol 323: 951-960.
[17] Bright, J.N. and Sansom, M.S.P. (2003) The flexing/twirling helix: exploring the flexibility about molecular hinges formed by proline and glycine motifs in transmembrane helices. J Phys Chem B 107: 627-636.
[18] Bright, J.N., Shrivastava, I.H., Cordes, F.S. and Sansom, M.S.P. (2002) Conformational dynamics of helix S6 from Shaker potassium channel: simulation studies. Biopolymers 64: 303-313.