I. Development of Novel NMR Methods to Screen and Characterize HAV 3C Protease Inhibitors.
Hepatitis A is a member of the picornaviruses which include many human pathogens such as poliovirus, rhinovirus (the cause of the common cold), and encephalymycaridiovirus. Interestingly all picornaviral proteins are produced as a single, large polyprotein precursor. An endogenous protein, known as the viral 3C proteinase, is responsible for processing the precursor into individual proteins necessary for viral function and survival. The fact that 3C proteases are essential for the viral life cycle means that they would be excellent targets for structure-based drug design efforts. The hepatitis 3C viral protease, in particular, is a stable, monomeric protein consisting of 216 amino acids. An efficient, high-level, bacterial expression system has been developed which yields 10¹s of milligrams (per litre) of protein on minimal media. Furthermore, the 3C proteinase structure has recently been solved by X-ray crystallography to a resolution of 1.9 Å. We have recently shown that this protein is also quite amenable to NMR analysis, particularly if the concentration is kept relatively low (<0.5 mM). In this project we intend to use the HAV 3C protease as model ³receptor² with which to develop: (a) NMR-based techniques for rapid screening of putative inhibitors and (b) NMR-based techniques for the rational, structure-based design of improved protease inhibitors.
In pursuing these two goals, it will be necessary to first, complete the 15N and 1H assignments of this protein. We are currently using a novel combination of HM(S)QC-J, HSQC-TOCSY and X-ray data to complete the task. This method uses chemical shift information in combination with accurate J-coupling data and crystal structure dihedral angels to assist with the sequential assignment. We expect to complete the assignment of HAV 3C within the next 3 months. After the assignment has been completed, we plan to use techniques similar to those recently developed by Dr. Stephen Fesik for rapid drug candidate screening (SAR by NMR). In this process, aliquots of the 15N labelled target protein are co-dissolved with different sets (10-20) of unlabelled lead compounds. By rapidly (10 minutes) collecting 15N HSQC spectra of the protein/inhibitor mixture and detecting spectral perturbations relative to the ³free² protein spectra, it is possible, by simple inspection, to detect which region(s) of the protein are binding the inhibitor(s). With such a simple technique it is possible to screen 100¹s of compounds in a single day while at the same time determining the precise binding location for most of the higher affinity compounds.
Using this structure-based screen, in combination with our new methods for "back-calculating" 15N and 1HN chemical shifts (see below) we will be able to associate chemical shift changes in the HSQC spectra with exact 3D structure changes in the protein itself. This 3D structural information will be used to automatically model putative ligands into the active site and to characterize their interactions using molecular modelling software. This information will be used to refine or design better inhibitors and to further map the active site of this unique protease.
II. Cracking the Chemical Shift Code.
Over the past 10 years, Nuclear Magnetic Resonance (NMR) Spectroscopy has emerged as one of the most powerful tools in structural biology. By using of newly developed pulse sequences, advanced spectrometer designs and innovative data processing techniques, NMR spectroscopists have been able to determine the atomic structure of more than 200 peptides and proteins. However, the process by which proteins are assigned and their structures determined is still slow (often requiring more than one year per structure), tedious and error-prone. Many of these difficulties arise from the mind- numbing task of identifying, calibrating and measuring thousands of NOE's (Nuclear Overhauser Effects) from crowded and sometimes difficult-to-interpret NOESY spectra.
Despite predictions to the contrary, it has recently been shown that relatively detailed structural and spatial information in peptides and proteins can be obtained directly from chemical shift data (Williamson, 1990; Wishart et al., 1991). In particular, it is now possible to locate and identify secondary structures such as helices, beta-strands, helix N-caps and beta-turns with better than 90% accuracy using only raw chemical shifts. Given these encouraging results for secondary structure determination, we believe that tertiary structure determination could be performed just as well using chemical shifts. Indeed, we have preliminary evidence that suggests that 1H, 13C and 15N chemical shifts exclusive of any NOE information, can be used to accurately determine the three- dimensional structure of peptides and proteins.
In "cracking the chemical shift code" we have undertaken a multidisciplinary approach that combines experimental, computational and statistical approaches to assist in developing a robust, fully automated methodology. These approaches are directed at overcoming two key problems: (1) the lack of both data and methods to rapidly calculate chemical shifts from protein coordinate data; and (2) the lack of appropriate algorithms to extract and incorporate chemical shift information into protein folding simulations. To address these problems, we have started to: (i) experimentally determine "nearest neighbor effects" on the chemical shifts of all 20 amino acids using a specially designed set of random coil hexapeptides; (ii) determine the structure-dependent chemical shifts of all 20 amino acids on small conformationally constrained b-sheet and a-helical peptides; (iii) quantify the effects on the variation of peptide dihedral angles (f, y and c) on chemical shifts through statistical analysis of some 30 solved peptide and protein structures; (iv) integrate the above information into a series of Neural Networks, Genetic Algorithms and simplified energy minimizers to "fold" test proteins into putative structures; and (v) evaluate the technique on several small, well-characterized proteins (gramicidin S, T4 thioredoxin, ubiquitin and BPTI).
To date we have succeeded in developing a computer program (called PHIPSI2CS) that accurately and rapidly calculates backbone and sidechain 15N, 1H and 13C chemical shifts from 3D coordinate data. Tests indicate tha this program is the most accurate chemical shift prediction method so far developed. We have also developed a program (called PROTEIN BUILDER) that builds 3D structures from a combination of J-coupling and chemical shift data. This program was used to construct the 3D structures (independent of NOE information) of five cyclic gramicidin S peptides (published in the April issue of Nature Structural Biology). This is the first successful demonstration of NMR-based 3D structure generation without the use of explicit NOE information.
BACK to the main home page