Are hydrophobic protein surfaces like big or small hydrophobes?

It seems to me that a paper on protein denaturation by Michele Vendruscolo at Cambridge, Stefano Gianni at the University of Rome La Sapienza, and their colleagues will repay careful study (C. Camilloni et al., Sci. Rep. 6, 28285; 2016 – paper here). The researchers use MD simulations to model observed NMR shifts during hot and cold denaturation, and thereby to gain insight into transition-state structures, changes in hydration and thermodynamic parameters. This enables them to characterize in some detail the differences between hot and cold denaturation: the former has more secondary structure, being more influenced by hydrophobic interactions. In effect this points to the existence of two alternative folding mechanisms from denatured states. What’s more, water molecules at the protein surface have the same number of hydrogen bonds on average as those in the bulk. This is what theory predicts for small hydrophobes (<1 nm or so), whereas for larger extended hydrophobic surfaces some hydrogen bonding is thought to be inevitably lost. In other words, it seems that the hydrophobicity of proteins must, on account of their complex surface topography and chemical heterogeneity, be considered to be more akin to that of small rather than large hydrophobes.

A detailed look at how protein surface topography and chemistry affects local water organization is described by Tom Kurtzman of Lehman College and coworkers (K. Haider et al., JPCB 120, 8743; 2016 – paper here). They look at the ligand-binding clefts of six structurally diverse proteins and identify circumstances where the constraints on local water structure compromise the enthalpy. These include deep and narrow cavities and ones with weak water-solute interactions, but also sometimes sites with charged residues in which there can be frustration of adjacent water-water interactions. Clearly this kind of information should be useful for designing ligands that bind competitively to a site, but it remains unclear whether one can yet identify generic design features or whether each case must be considered on its own terms. The broader question is perhaps whether the hydration characteristics must be considered to have been selectively fine-tuned or simply a necessary compromise occasioned by conflicting demands on the binding site. That’s to say, are the hydration environments each individually ‘functional’ or at least partly an epiphenomenon?

One of the clearest cases I can remember seeing for the coupling of hydration structure with protein motions and function is supplied by Tomotaka Oroguchi and Masayoshi Nakasako at Keio University (Sci. Rep. 6, 26302; 2016 – paper here). They have looked at the hexameric multi-domain protein glutamate dehydrogenase (GDH) using MD simulations and AFM. The opening and closing of a hydrophobic pocket HS1 are accompanied by wetting and drying of the pocket, while binding and unbinding of water molecules in a hydrophilic crevice HS2 accompany changes in its length. These two changes in hydration are coupled, creating a kind of hydration-driven mechanism for large-scale conformational change in GDH.


Drying and wetting of the hydrophobic pocket HS1 accompanying the opening and closing of the cleft.


Conformational changes in the GDH protein due to coupled hydration changes at the sites HS1 and HS2.

The coupling of hydration change to large-scale protein dynamics is also the subject of an experimental study by Keisuke Tominaga and colleagues at Kobe University using dielectric spectroscopy at THz frequencies (N. Yamamoto et al., JPCB 120, 4743; 2016 – paper here). They look at lysozyme in the solid state under different hydration conditions, and see two relaxational modes. They attribute the faster of them, with a ~20 ps relaxation time, to coupled water-protein motion: the mode is primarily due to hydration water dynamics, but the hydration water “drags” with it the hydrophilic groups at the protein surface. The slower (~100 ps) mode might be due to motions of the amino-acid side-chains induced by hydration. Looking at the temperature dependence of the spectra, the authors also see a signature of the familiar dynamical transition around 200 K.

The same issue is explored by Dongping Zhong and colleages at Ohio State University using femtosecond spectroscopy of tryptophan relaxation (Y. Qin et al., PNAS 113, 8424; 2016 – paper here). They too see coupling between hydration water and protein side-chain dynamics which slows down the water reorientational relaxation at the interface relative to the bulk. (They study DNA polymerase IV.) Mutational studies and MD simulations imply that causation here goes in the direction of the protein side-chain fluctuations being slaved to the cooperative dynamics of the hydration water.

A closer look at the dynamics of hydration water is provided by Peter Bolhuis of the University of Amsterdam and colleagues, who compare MD simulations with femtosecond IR spectroscopy of the hydration of bovine α-lactalbumin, both in its native and a misfolded state (Z. F. Brotzakis et al., JPCB 120, 4756; 2016 – paper here). The water relaxation times here are typically of the order of tenths of to several picoseconds; ‘slow’ waters have relaxation times > 7 ps, some as much as 20 ps. These waters tend to be located in concavities on the protein surface and make fewer hydrogen bonds with surrounding waters than do molecules in the bulk. Moreover, waters near hydrophobic groups tend to be slower on average than those near hydrophilic groups. But although misfolding exposes more of the hydrophobic surface, it also means that these hydrophobic regions are less concave, and so the water dynamics is somewhat faster on average and there are fewer of the “ultraslow” sites.


Water reorientational decay times seen in simulations of native (left) and misfolded (right) bovine α-lactalbumin.

A somewhat comparable exercise is conducted for B-DNA by James Hynes and Damien Laage at the ENS Paris and colleagues (E. Duboué-Dijon et al., JACS 138, 7610; 2016 – paper here). And there are some commonalities: while the hydration water is generally rather slower to reorient than in the bulk, the waters confined in the narrow minor groove are much more significantly retarded (relaxation times 30-85 ps). Moreover, there is considerable heterogeneity, and some of this comes from coupling of the macromolecular fluctuations with the water dynamics, especially in the minor groove. In other words, there does not seem in this case to be slaving of biomolecular dynamics to those of the solvent, but more or less the reverse.


Water reorientational times on the minor and major grooves of the B-DNA dodecamer (CGCCAATTCGCG)2

Lorna Dougan at Leeds and colleagues have found evidence of a low-density form of water at low temperatures (285-238K), which might be related to the putative phase transition separating low- and high-density liquids in the metastable regime (J. J. Towey et al., JPCB 120, 4439; 2016 – paper here). They keep the water liquid by mixing it with the cryoprotectant glycerol. Neutron scattering and simulation show that at low temperatures the mixture segregates at the nanoscale, and the water nanophase has greater tetrahedral ordering than the bulk.

Predicting protein structure from sequence data often draws on information on homologous structures or fragments from the Protein Data Base. But such homologies cannot always be spotted, or might not be present in the database, or might not be reliable. Peter Wolynes and colleagues at Rice have developed a scheme for predicting structures ab initio, without bioinformatics input, using what they call the atomistic, associative memory, water mediated structure and energy model (AAWSEM) (M. Chen et al., JPCB 120, 8557; 2016 – paper here). This uses coarse-grained simulations at the whole-protein level while drawing on atomistic simulation of fragments – and crucially, incorporates water-mediated interactions in the folding process. It’s a smart approach to the folding problem that draws on the biological reality – the fact that protein folding is funneled to make it evolutionarily robust to small variations in sequence – rather than brute-force number-crunching.

Water mediation is thought to be important too for the aggregation of amyloid fibrils. Samrat Mukhopadhyay and colleagues at the Indian Institute of Science Education and Research in Mohali have used time-resolved fluorescence measurements on the human prion protein (PrP) to investigate how (V. Dalal et al., ChemPhysChem 17, 2804; 2016 – paper here). They find that water hydrating the amyloid-competent oligomers has mobility retarded by three orders of magnitude relative to the bulk, perhaps because of entrapment in the collapsed polypeptide chains. They say that this water might create a hydrogen-bonded network that stabilizes the partly unfolded, molten oligomer conformation and acts as a scaffolding for the assembly of oligomers into fibrils.


Proposed role of ordered water molecules in the misfolding and amyloid formation of PrP – and in protein misfolding diseases more generally.

Antiviral drugs against influenza B could work by blocking the proton-conducting channel BM2, but no such have yet been devised. Mei Hong at MIT and colleagues have used NMR to investigate the mechanism of proton transport in BM2 and the role of hydration, and to elucidate the differences with AM2 from influenza A (J. K. Williams et al., JACS 138, 8143; 2016 – paper here). The His19 residue in BM2 remains unprotonated to lower pH than the corresponding His 37 in AM2, but increasing channel hydration in acidic conditions seems to enhance proton transport to His 19 from water molecules.

Why trehalose acts as a cryoprotectant of protein structure still isn’t fully understood. Jan Swenson and coworkers at Chalmers University of Technology in Göteborg try to develop a comprehensive picture by looking at how trehalose affects the protein glass transition, denaturation temperature, and solution viscosity (C. Olsson et al., JPCB 120, 4723; 2016 – paper here). They study the myoglobin-trehalose-water system using DSC and viscometry. In short, their results seem to exclude the picture in which trehalose displaces water in the solvation shell; on the contrary, they suggest that the protein retains one or two layers of water within a stabilizing water-trehalose matrix. This would be consistent with an apparent lack of coupling between the trehalose-water matrix dynamics and the stability of the native protein.


Schematic of the interactions between water, trehalose and protein.

That picture of a lack of direct interaction between trehalose and proteins – the disaccharide is in fact preferentially excluded from the protein hydration layer – is also the general context for an experimental study by Christina Othon of Wesleyan University in Connecticut and colleagues of trehalose bioprotection (N. Shukla et al., JPCB 120, 9477; 2016 – paper here). Using ultrafast fluorescence spectroscopy for two fluorescent probes, they see a slowdown of water reorganizational dynamics at relatively low trehalose concentrations (0.1-0.25 M, well below the vitrification threshold). At these concentrations, there is around 7 water layers between osmolyte molecules. These results therefore support an indirect mechanism for cryoprotection. Sucrose has much the same effect, but less markedly, the researchers say.

The interaction between two hydrophobic particles in water is generally attractive: this is simply the (water-mediated) hydrophobic effect. But Alenka Luzar and coworkers at Virginia Commonwealth University show that this interaction can become repulsive (B. S. Jabes et al., JPC Lett 7, 3158; 2016 – paper here). Such repulsion has been seen before in simulations of fullerenes and carbon nanotubes in water, and has sometimes been attributed to specific structural changes in the water. But Alenka and her colleagues show that it can be explained purely as a geometric effect of the thermodynamic cost of formation of a liquid-vacuum interface bridging the hydrophobic particles (in these calculations, pure and propyl-terminated graphitic nanoparticles) when drying occurs in the intervening space. This process can be modeled with a straightforward, bulk-like Young-type calculation of the surface free energies.

Nanoconfinement effects on water structure and properties are investigated by Vrushali Hande and Suman Chakrabarty of the National Chemical Laboratory in Pune through simulations of water inside reverse micelles and water-in-oil nanodroplets (Phys. Chem. Chem. Phys. 18, 21767; 2016 – paper here). For the reverse micelles the interface is (negatively) charged, and the deviations from bulk-like behaviour are longer-ranged for orientational order than they are for translational ordering. These effects are far less pronounced for nanodroplets in oil, where the interface is hydrophobic, indicating that electrostatic influences on the hydrogen bonding are more pronounced than spatial confinement per se.

Also on nanoconfinement: quite why water has an enhanced mobility in carbon nanotubes remains a matter of some debate. Using IR spectroscopy, Pascale Roy at the Synchrotron Soleil in Gif-sur-Yvette and colleagues suggest that it may be due to unusually “loose” hydrogen-bond networks among water molecules inside the nanotubes (S. D. Bernadina et al., JACS 138, 10437; 2016 – paper here). They look at nanotubes with diameters of 0.7-2.1 nm, in which the water varies from single-file chains to multilayers, and find a spectroscopic signature of “loosely bonded water” in all cases – in the latter seeming to correspond to waters in the outer layers with dangling OH bonds pointing towards the nanotube walls.

The distance dependence of the hydrophobic force between two hydrophobic walls is investigated in MD simulations by Biman Bagchi and colleagues at the Indian Institute of Science in Bangalore (preprint arxiv.1608.04107). They find a bi-exponential force law, with correlation lengths of 2 nm and 0.5 nm, and a crossover close to 1.5 nm. This behaviour is mimicked by the tetrahedral order parameter, but I’m not entirely clear what the authors’ mechanistic explanation is.

Of course, the issue with many studies of this kind is that your results might only be as good as your model. Angelos Michaelidies and colleagues at UCL offer an overview of the extent to which density-functional theory supplies a good description of water, from small clusters to the bulk (M. J. Gillan et al., JCP 144, 130901; 2016 – paper here). In particular they consider how well different functional forms of exchange-correlation terms perform, and what role many-body terms play. Looks like essential reading for anyone using DFT to model aqueous systems.

Many-body effects are also central to a study by Shelby Straight and Francesco Paesani at UCSD of influences of water’s dipole moment on the hydrogen-bond network of pure water (JPCB 120, 8539; 2016 – paper here). They use simulations to predict the infrared spectra of HOD in H2O, and in particular the shape of the OD stretch. They find that the calculated spectral diffusion of this vibrational frequency depends rather strongly on exactly how one truncates a many-body expansion of the water dipole.

How effectively can hydration be described with a coarse-grained model? Bill Jorgensen and colleagues consider the performance of one attempt to balance accuracy and speed that mixes all-atom and coarse-grained descriptions – the so-called AAX-CGS model, in which all-atom solutes are solvated with coarse-grained water (X. C. Yan et al., JPCB 120, 8102; 2016 – paper here). The approach works well for hydrophobic and halogenated alkane solutes, less so for those that are more polar or engage in hydrogen bonding (amines, alcohols). But the efficiency of the calculations beats that of all-atom simulations by about an order of magnitude or more.

Why are ion hydration free energies asymmetric with respect to ion charge? Rick Remsing and John Weeks investigate that question using an analytical model for calculating hydration free energies that involves gradually “turning on” the ion-solvent Coulomb interaction (JPCB 120, 6238; 2016 – paper here). This enables them to see why the Born solvation model fails to capture the asymmetry: in short, it works well enough for slowly varying Gaussian charge distributions but not for the abrupt, delta-function-like distributions in ion cores. Only in the latter case is the asymmetry in response to ion charge recovered.