Proteins at Solid Interfaces

Introduction

Proteins occur at solid interfaces across nature; examples include biomineralization and anti-freeze proteins. Understanding such examples in nature and exploiting this understanding in the medical and commercial contexts is of increasing interest. Some examples of particular relevance here are (see figure below) control of cell deposition in tissue scaffolds, sensors for detecting viruses, cells, biomolecules and heavy metal particulates, and self-assembly of nanoscale building blocks like nanotubes and nanoparticles to form novel nanostructured materials and systems (e.g. electronic devices).

We are using various molecular simulation methods to elucidate the fundamentals of the protein-solid surface interactions, and to encapsulate this within a design tool for the in silico rational de novo design of surface binding peptides for applications such as those mentioned above. We describe some of the work undertaken to date as part of our efforts to address these objectives.

 

Effective EAs for ab initio prediction of 3D protein structure

Assuming the thermodynamic hypothesis holds (i.e. the three-dimensional (3D) structure of a protein corresponds to the global free energy minimum), the ab initio prediction of the 3D structure of a protein requires only the protein sequence, details of its environment (e.g. water, surfaces), a physics-based method for evaluating the energy of interaction between the various elements of the systems (e.g. intra-protein, protein-surface, protein-water, water-water interactions), and a method for identifying the global free energy minimum. Evolutionary algorithms (EAs) were applied to the latter task in the 1990s with limited success - this was surprising and so we looked at why. As the figure below shows, we found that EA design is critical to the success of an EA in finding the global minimum and the efficiency with which it does so. We have found that real-encoded, elitist steady-state EAs with crossover using uniform selection and mutation is best in general, although there is some evidence that aspects of the best design could vary with protein type and potential model - this suggests that it would be desirable to be able to change aspects of the design in response to performance (i.e. adaptive design), which we plan to implement in the near future.

The figures below left and middle show that EA performance is also critically dependent on the EA control parameters - most particularly the mutation and crossover rates - and details of the potential model used. The figure below right also shows that the optimal control parameters vary with the level of desired accuracy of the structure - once again, it is clear that it would be desirable to be able to change the control parameters in response to performance (i.e. adaptive control parameters), which we also plan to implement in the near future. We have applied our EA-based ab initio structure prediction approach to a range of peptides - the figure below shows two examples where the predicted (red) and actual structures from the PDB database (green) correspond very well. These were generated using the Amber potential model.

 

Switching in polyalanine at a solid surface

In our first study of a protein on a solid surface, we observed that polyalanine switches between distinct conformations as the surface energy is increased (right) rather than undergoing a gradual change in conformation. It appears that this switching behaviour arises out of the ability of the polyalanine molecule to sustain intramolecular hydrogen bonding and, we hypothesize, its symmetry. This switching, which is also accompanied by significant length changes, could be exploited technologically (e.g. molecular switches), and may be of relevance to the activity and function of natural proteins when near solid surfaces.

 

Efficient prediction of solvation energies
One of the major challenges in protein structure prediction is inclusion of solvents such as water. Explicit representation of water molecules around a protein is very expensive and is certainly not feasible in the design context towards which we are working. Implicit models, on the other hand, are unlikely to be satisfactory when solid interfaces are involved because of the possibility of structuring of the water between the protein and the solid surface. We have found that combining the semi-explicit Langevin dipole (LD) method with the Amber classical potential model can yield results as good as explicit approaches in a tiny fraction - less than 1% - of the time. As the figure below left shows, we have also shown that, unlike implicit methods, the LD-Amber approach is capable of yielding inhomogeneous electric fields that are analogous to that produced by the much more expensive explicit approaches. The example shown here is for the case where water forms so-called 'hydrogen bond bridges' between widely separated parts of the penta-peptide met-enkephalin (shown by the arrows in the figure on the right; the plane in this figure corresponds to that of the electric field shown in figure on the left) - this is extremely important as such phenomena play an important role in determining the 3D structure of proteins.
Greater details of this work may be found in the article of Mijajlovic and Biggs (2007a)

 

EA-based ab initio prediction of 3D protein structure in water
We have recently used our EA to predict the structure of met-enkephalin in water both with neutral end caps (below left) and without them, which leads to oppositely charged ends (below right). The actual structure of met-enkephalin in water is not well known because of its small size and the presence of two glycine residues in the middle. The predicted structures are, however, reasonable, showing the important role that hydrogen bond bridging (indicated by the reddish area in the electric fields) can play, as well as how water can shield the charge-charge interactions between the ends of the uncapped peptide. Further analysis of our results is currently underway and will be reported on in the near future.

 

EA-based ab initio prediction of 3D protein structure at solid-water interfaces

We have very recently used our EA to predict the structure of uncapped met-enkephalin at a graphite-water interface - images from these simulations are being prepared and will be mounted here in the very near future.