Conformational changes in proteins: rates and mechanisms from discrete path sampling

Proteins are chains constructed from twenty different amino acids, which must fold up into a biologically active form after they are synthesised in cells. They play essential structural and catalytic roles in all living organisms. All the information required for a protein to function must ultimately be encoded in nucleic acids such as DNA. Recent efforts, such as the human genome project, are providing an increasing number of nucleic acid sequences, from which the amino acid sequences of every protein in a given organism can be derived. Once the amino acid sequence of a protein is known we can seek to determine the three-dimensional structure that it folds into, and the function that it performs in the organism. However, it has become increasingly apparent that understanding the biomolecular function of many proteins will require us to consider more than one structure. In particular, there is a growing body of experimental evidence suggesting that well-defined structural changes, usually called conformational changes, accompany protein function. Hence, to gain insight into how such proteins work, we must discover the pathways involved, and how quickly they are traversed. The potential benefits of this research are of great importance in fields such as drug design. For example, the usual strategy in efforts to create anti-cancer agents has generally been to target rapidly dividing cells. Recently, however, new drugs have been designed that work by binding to specific conformations of particular proteins, which prevents them from binding the molecules that they usually interact with, thus blocking their function. This line of research appears to be very promising, and depends upon knowledge of alternative protein conformations. Computer simulation of conformational changes could potentially play an important role in the above effort by supplying mechanistic information that is hard to obtain experimentally. In principle, we could adopt a model for the interatomic forces in the protein, and then solve Newton's equations of motion on a computer to advance the state of the protein in time using a series of small steps. The main problem with this approach is that a huge number of steps are needed, and even with recent advances in computer technology, it is not yet possible to follow the protein dynamics on the required micro- to millisecond time scale in this way. However, alternative simulation methods are available, which can bridge the time scale gap by making additional approximations. The present proposal is to employ a recently developed approach of this sort to characterise the conformational changes involved in four proteins of particular interest. In each case, the end-point conformations are known, or have been suggested, from experimental data, and computer simulations could be used to calculate the intervening paths. These simulations would also provide rate constants for the corresponding mechanisms, and could be used to analyse the effect of mutating strategically located amino acids into alternative forms. Four particular examples would be treated in this study to cover a range of important protein function, including catalysis, biosynthesis and signalling. For example, adenylate kinase (Adk) proteins perform a vital biological function, but differ slightly in their precise amino acid sequence between different organisms. The Adk protein from an organism adapted to live at a temperature of 80 centigrade exhibits reduced activity at 20 centigrade when compared with Adk from 'normal' bacteria. Recent experiments suggest that this difference is caused by slower conformational dynamics. Computer simulation would be used to test this hypothesis, and to derive details of the pathways involved to suggest future directions for experiment.

This proposal aims to characterise the pathways involved in conformational changes associated with the function of four key proteins. In each case the conformations involved are known, or suggested, from experiment, and protein dynamics appears to play a crucial role in the corresponding catalytic or signalling process. Computer simulation, using the recently developed discrete path sampling (DPS) approach to bridge the time scale gap, will be used to calculate pathways at an atomic level of detail, and to provide rates for comparison with experiment. Various experiments, especially NMR spin relaxation work, highlight the role of conformational change in protein function, and the need to consider such systems as dynamic objects. Understanding the corresponding pathways is likely to play a vital role in rational drug design in the future, and computer simulation should be at the forefront of such efforts. Unfortunately, conventional simulations cannot bridge the required time scales of micro- to milliseconds. Alternatives, such as elastic network models, steered molecular dynamics, and normal mode analysis, involve such radical approximations that realistic dynamical information is probably lost. However, the DPS method is capable of providing dynamics on the experimental time scale, using established unimolecular rate theory for the transition rates between individual local minima on the underlying potential energy surface (PES). Phenomenological rate constants are derived within this framework from ensembles of discrete paths, where each discrete path consists of a series of local minima and the transition states that connect them on the PES. Once appropriate pathway ensembles have been calculated they will be analysed to identify possible targets for rational drug design. As the recent development of imatinib has shown, understanding the pathways involved in protein function can provide the key to achieving specific inhibitor binding. Discovering the complete pathway in atomic detail will provide new opportunities for such design efforts. It should also be possible to determine the effect of mutations from DPS simulations, and make predictions for comparison with future experiments. The four systems proposed for this initial study cover a range of important catalytic and signalling functions. Dihydrofolate reductase will be considered first, to discover the pathways between the 'closed', 'open' and 'occluded' forms, and to compare the effect of single and double mutations for which experimental data is available. These results should provide an explanation for the correlated dynamics observed between spatially distant residues. Nitrogen regulatory protein C will be considered next. Here we will seek to answer the key question raised by experiment, of how the wild-type signalling protein reaches the active state. The calculated pathways will provide a detailed test of the proposed allosteric activation mechanism, where phosphorylation shifts the position of a preexisting equilibrium. The pathways and rates will then be compared for two adenylate kinases: one from a hyperthermophile, and the other from a mesophile. Experiments suggest that the rate-determining step in catalysis is the opening of the substrate 'lids', which appears to be significantly slower in the hyperthermophile protein. Pathways will be calculated for the two proteins in the presence of substrate to explain the observed lid-opening and closing rates. The pathways will then be analysed to suggest mutations that might have interesting effects on the enzymatic activity. The fourth protein to be considered will be human cyclophilin A, which is a peptidyl-prolyl isomerase. Here experiments suggest that isomerisation of a cis substrate to the trans form is strongly coupled to specific conformational changes in the enzyme. This transformation will be analysed in detail using DPS calculations, and the calculated rates compared with experiment.