Introduction

1. Toward high-resolution prediction and design of transmembrane helical protein structuresP. Barth, J. Schonbrun and D. Baker PNAS Sep 28 2007 Tim Nugent – BUGF - 18/10/07

2. Introduction • While membrane proteins comprise ~30% of a proteome, they are severely under-represented in structural databases (e.g. ~1% PDB) due to the problems associated with crystallisation. • Given the biological significance of TM proteins, an understanding of their structures is an important target for theoretical prediction methods. • This paper presents an all-atom physical model that describes intra-protein and protein-solvent interaction in the membrane environment. • The model is evaluated by: • Side chain packing • Energy gaps between native and non-native structures • Distortions in TM helices (helical kinks) • Test set of 18 high-resolution crystal structures.

3. Overview of the Physical Model • Van der Waals packing, hydrogen bonding, solvation and electrostatics interactions are thought to be the main forces that stabilise TM proteins. • However, the magnitude and relative importance of these forces has never been clearly established and no current method has been shown to quantitatively capture the energetics involved. • This model attempts to address this problem: it describes interactions between protein residues at atomic detail, whereas the water, hydrophobic core and lipid head groups are treated using continuum solvents models. • Membrane described by 3 continuum phases: 2 isotropic (water and hydrophobic core of lipid bilayer) and 1 anisotropic (lipid head group region of the membrane). • Hydrogen bonds are treated explicitly, including weak CH-O h-bonds and bifurcated h-bonds. • The energy function is an extension of the ROSETTA full-atom potential developed for water soluble proteins structure prediction.

4. Force Field • Lennard-Jones potential for modelling VDW forces. • Backbone torsional term for local structural propensities. • Knowledge-based pair interaction term for side chain interactions. • 20 reference energies that control overall amino acid composition. • Implicit atomic solvation term based on Lazaridis and Karplus model. • Orientation-dependent h-bond term. • H-bond and solvation energies modified for the anisotropic membrane environment. • CH-O and bifurcated H-bonds modelled explicitly • Thought to play an important role in helical distortions and stabilising polar residues in membrane proteins.

5. Side-chain Conformation Recovery Test • The ability of the model to predict side chain conformations was assessed by simultaneous re-packing side chains on fixed backbones derived from 18 high-resolution crystal structures. • Ch1 and Ch2 dihedral angles correct for 73% of buried positions. • Recovery is higher in regions embedded in the hydrophobic core, consistent with a higher degree of packing in those regions compared to the lipid head group.

6. Amino Acid Recovery Test • The energy function was tested in redesign experiments in which sequence space is searched for the combination of amino acids that minimises the structure's free energy. • The method selected the native amino acid at >45% of buried and ~35% of all positions. • These values compare well with those obtained for water soluble proteins. • For both lipid and water soluble proteins, recovery is higher for non-polar than polar residues, consistent with the latter being often selected by nature for function rather than stability. • The gradient in amino acid polarity between core and head group regions agrees with those observed in native proteins.

7. Native TMH Docking Test • Monte Carlo simulations using fixed backbones but flexible side chains were used to dock single TMHs on to their templates to generate diverse sets of near-native and non-native conformations. • Native and near-native were significantly lower in energy than non-native (Z score). • Validates the energy function. Effect of the membrane hydrogen bond potential on the discrimination between native and nonnative decoys in native helix random docking calculations of Glycophorin A. Decoys scored with the ROSETTA hydrogen bond potential with (Left) or without (Right) the membrane-specific bifurcated/weak hydrogen bond potential are represented. The lowest-energy decoys are encircled in black.

8. Analysis of the Contribution of Individual Terms in the Model • Solvation • Solvation term dominated by free-energy cost of transferring polar groups and free energy gain of transferring non-polar groups from water into the bilayer. • Removal of the solvation potential from the function led to a 25% decrease in total sequence recovery, and also loss of the polarity gradient. • VDW and H-bonding • Short-range VDW and h-bond interactions contribute to the discrimination between TMH interfaces that have smaller structural discrepancies to the native structures • Addition of CH-O and bifurcated h-bond terms increased the energy gaps between native and non-native docked conformations by 31% on average. • This highlights their role in stabilising tightly packed interfaces, e.g. Involving glycine zippers in glycerol channels and glycophorin A, or interfaces accompanying small polar residues. • Incorporating CH-O and bifurcated h-bond terms improved recovery of polar residues (especially Thr and Ser) by 13%. Overall sequence recovery improved by 2%.

9. VDW and H-bonding

10. Sequence-Based Modelling of Distorted TMHs • Most TMH distortions occur at Pro residues or at positions where kinks are initiated by Pro residues and later stabilised by tertiary interactions during evolution. • The kink usually propagates 4 residues N-terminal to the residue responsible for the bend. • They hypothesised that the native conformation of the distorted helix could be identified by modelling the chain away from the bend-induced hinges by ensembles of pairs of ideal helix fragments with a range of orientations. • Ideal helix fragments docked to native protein templates - select pairs with the lowest non-local interaction energy with other TMHs. • Control experiments where ideal helices for the entire TMH length were docked on protein templates confirmed that native tertiary interactions could not be recapitulated with ideal backbone geometries. • Native conformations could also not be identified by individual helix fragments away from the bend.

11. Sequence-Based Modelling of Distorted TMHs

12. De novo prediction of interface-bound peptide structures • Domains lying parallel to the bilayer are recurrent in membrane-embedded polypeptides. • Structure prediction performed on the fd-coat protein – NMR structure consisting of 1 TMH and one interfacial helix. • Course-grained ROSETTA structure was refined and relaxed with the energy function. • Lowest energy structure had an rmsd of 2.4 Å. • Model has characteristic hydrophobic and polar residues snorkelling in and out of the membrane. (A) fd-coat protein. (Left) Backbone superposition of the experimental structure determined by solid-state NMR (blue) and the lowest-energy decoy generated by ROSETTA (pink) starting from an extended chain. The rmsd over 30 C atoms is 2.4 Å. (Right) All-atom representation of the lowest-energy decoy generated by ROSETTA. The boundaries predicted by ROSETTA between the hydrophobic core and the interface regions of the membrane are represented with a black solid line.

13. Prediction of TMH oligomeric interfaces by docking • Many functions of TMP are driven by oligomerisation. • They docked randomly orientated glycophorin A molecules. • Symmetry of the native homodimer wasn't enforced • Docked structure has rmsd of 0.65 Å compared to the native dimer. (B) Glycophorin A. Isolated monomers were docked with the ROSETTA protein–protein docking protocol and the all-atom membrane force field. The superposition between the native (blue) and the lowest-energy predicted structure (pink) is represented. The rmsd over 45 C atoms is 0.65 Å.

14. De novo prediction of polytopic membrane structures • To be observed experimentally, native structures must be significantly lower in free energy than non-native conformations. • Course-grained ROSETTA structures were refined and relaxed with the energy function. • The energies of the ab initio models were compared to native structures relaxed by the same protocols. • Lowest energy models have near atomic resolution - rmsd of 2.1 Å over 111 atoms and 2.4 Å over 139 atoms for the 4-helix bundles BRD4 and VATP respectively. (C and D) Ab initio structure prediction of polytopic membrane proteins. Native polytopic membrane protein conformations define a narrow energy basin in the all-atom conformational energy landscape. When near-native topologies are generated at the coarse-grained level, all-atom relaxed decoys define a funnel toward the native basin and the lowest-energy predicted structures have near-atomic resolution structures. In the energy versus rmsd plots, nonnative (red points) (generated from sequence by the ROSETTA coarse-grained structure prediction mode) and native conformations (green points) were relaxed by sampling the conformational degrees of freedom of all backbone and side-chain atoms. Cartoons show superposition between the native (blue) and the lowest-energy predicted structure (pink). Boxed areas show regions where close to native side-chain packing arrangements were obtained. The rmsd values are 2.1 Å over 111 C atoms for BRD4 (C) and 2.4 Å over 139 C atoms for VATP (D).

15. Conclusions • Model validated by a variety of in silico tests. • 73% of side-chain rotamers and 35% of all native residues recovered in a diverse range of TMPs. • Other important properties of TMP examined, e.g. Helix kinks and interfacial helices. • Good recovery suggests that the solvent model captures the main solvation properties of the membrane. • VDW and h-bond interactions are essential for stability and structural specificity of TMH bundles. • Weak and bifurcated h-bonds important – demonstrated by increase in Z scores (measure of structural determination in structure prediction tests). • Modelling TMH kinks using ideal helix fragments reduces the complexity of the search in conformational space. • Near-atomic resolution ab initio structures produced for 3 TMPs (45-145 residues).

16. Conclusions • Model captures essential physical properties that govern solvation and stability of membrane proteins. • More sophisticated model needed to predict the effects of membrane deformations and specific amino acid/lipid interactions that are involved in the regulation of membrane protein structures and functions. • A more accurate treatment of electrostatics accounting for induced polarisation effects may be necessary to model functional properties involving networks of buried residues or water/ion-solvated regions in channels and transporters.