Strukturvorhersage von Proteinkomplexen unter Verwendung weniger experimenteller Daten.
Final Report Abstract
The goal of this project was to develop methods to determine protein structures using lowresolution or sparse experimental data. Such methods are urgently needed as studies of large proteins or protein assemblies often yield only low-resolution data which makes structure determination and refinement a great challenge. In addition to X-ray crystallography, cryo-electron microscopy (cryoEM) has emerged as a powerful tool to reveal structures of large biomolecular systems. Our work focussed on structure refinement using data from both of these techniques. At low resolution the major challenge for refinement is to avoid over-fitting the data which is more likely to occur than at high resolution, since the number of parameters, i.e., the atomic coordinates, is much larger than the number of independent experimental measurements. Our strategy was to add information from structure prediction to the refinement process and thereby to alleviate the over-fitting problem. For this purpose we developed the Deformable Elastic Network (DEN) approach, which allows to balance experimental data and predicted structural information, as e.g., obtained by homology modeling. The deformability of the elastic network is controled by a single parameter, gamma, whose optimum value can be determined by cross-validation. The DEN method has been implemented into the CNS software package and systematically tested using diffraction data of penicillopepsin at various resolutions with different starting models. The results consistently show dramatic improvement of the quality of the refined structures at resolutions below 4 Å with respect to standard simulated annealing refinement. For example at a resolution of 4.5 Å the free R value dropped on average by about 10 percentage points. For structure refinement using density maps from cryoEM experiments we developed a new computer program, DireX, which combines the DEN method with fast geometry-based conformational sampling and real-space refinement. The program is available at https://simtk.org/home/direx/ In collaboration with the Center for Protein Folding Machinery (a NIH Nanomedicine Development Center) we work on modeling and refinement of protein structures using cryoEM data. The goal of this center is to understand the mechanisms of molecular chaperones; molecular machines that assist other proteins to fold correctly. The focus is on group II chaperonins (TRiC/CCT and Mm-cpn), which are barrel-like hexameric double-ring structures with a large cavity, inside which the unfolded peptide chain is supposed to be folded. The cavity is opened and closed upon a major conformational rearrangement of the subunits. We were able to characterize this conformational transition and to build detailed atomic models of the open and closed states. Our models suggest a mechanism of how nucleotide binding and hydrolysis could drive the opening motion. This mechanism implies, e.g, a dramatic change of the inter-ring contact interface; a fact that is currently being validated by our collaborators using mutagenesis experiments. As structural biology strives to study ever larger molecular systems to eventually reach out towards cellular scales, combining several sources of structural information with low resolution or sparse experimental data will be a key element of structural studies in the future. The tools we developed in this project, as we think, add an important piece to this puzzle and will be of general interest to the structural biology community. In fact, a number of collaborations have already been started as a consequence of the succes of this project.
Publications
- Combining Efficient Conformational Sampling with a Deformable Elastic Network Model Facilitates Structure Refinement at Low Resolution, Structure 15, 1–12, December 2007
G. F. Schröder, A. T. Brunger, and M. Levitt
