Biologische Hochschulschriften (Goethe-Universität)
Filtern
Erscheinungsjahr
- 2009 (2) (entfernen)
Dokumenttyp
- Dissertation (2)
Sprache
- Englisch (2)
Volltext vorhanden
- ja (2)
Gehört zur Bibliographie
- nein (2)
Schlagworte
- Molekulardesign (2) (entfernen)
Institut
Specific functions of biological systems often require conformational transitions of macromolecules. Thus, being able to describe and predict conformational changes of biological macromolecules is not only important for understanding their impact on biological function, but will also have implications for the modelling of (macro)molecular complex formation and in structure-based drug design approaches. The “conformational selection model” provides the foundation for computational investigations of conformational fluctuations of the unbound protein state. These fluctuations may reveal conformational states adopted by the bound proteins. The aim of this work is to incorporate directional information in a geometry-based approach, in order to sample biologically relevant conformational space extensively. Interestingly, coarse-grained normal mode (CGNM) approaches, e.g., the elastic network model (ENM) and rigid cluster normal mode analysis (RCNMA), have emerged recently and provide directions of intrinsic motions in terms of harmonic modes (also called normal modes). In my previous work and in other studies it has been shown that conformational changes upon ligand binding occur along a few low-energy modes of unbound proteins and can be efficiently calculated by CGNM approaches. In order to explore the validity and the applicability of CGNM approaches, a large-scale comparison of essential dynamics (ED) modes from molecular dynamics (MD) simulations and normal modes from CGNM was performed over a dataset of 335 proteins. Despite high coarse-graining, low frequency normal modes from CGNM correlate very well with ED modes in terms of directions of motions (average maximal overlap is 0.65) and relative amplitudes of motions (average maximal overlap is 0.73). In order to exploit the potential of CGNM approaches, I have developed a three-step approach for efficient exploration of intrinsic motions of proteins. The first two steps are based on recent developments in rigidity and elastic network theory. Initially, static properties of the protein are determined by decomposing the protein into rigid clusters using the graph-theoretical approach FIRST at an all-atom representation of the protein. In a second step, dynamic properties of the molecule are revealed by the rotations-translations of blocks approach (RTB) using an elastic network model representation of the coarse-grained protein. In the final step, the recently introduced idea of constrained geometric simulations of diffusive motions in proteins is extended for efficient sampling of conformational space. Here, the low-energy (frequency) normal modes provided by the RCNMA approach are used to guide the backbone motions. The NMSim approach was validated on hen egg white lysozyme by comparing it to previously mentioned simulation methods in terms of residue fluctuations, conformational space explorations, essential dynamics, sampling of side-chain rotamers, and structural quality. Residue fluctuations in NMSim generated ensemble is found to be in good agreement with MD fluctuations with a correlation coefficient of around 0.79. A comparison of different geometry-based simulation approaches shows that FRODA is restricted in sampling the backbone conformational space. CONCOORD is restricted in sampling the side-chain conformational space. NMSim sufficiently samples both the backbone and the side-chain conformations taking experimental structures and conformations from the state of the art MD simulation as reference. The NMSim approach is also applied to a dataset of proteins where conformational changes have been observed experimentally, either in domain or functionally important loop regions. The NMSim simulations starting from the unbound structures are able to reach conformations similar to ligand bound conformations (RMSD < 2.4 Å) in 4 out of 5 cases of domain moving proteins. In these four cases, good correlation coefficients (R > 0.7) between the RMS fluctuations derived from NMSim generated structures and two experimental structures are observed. Furthermore, intrinsic fluctuations in NMSim simulation correlate with the region of loop conformational changes observed upon ligand binding in 2 out of 3 cases. The NMSim generated pathway of conformational change from the unbound structure to the ligand bound structure of adenylate kinase is validated by a comparison to experimental structures reflecting different states of the pathway as proposed by previous studies. Interestingly, the generated pathway confirms that the LID domain closure precedes the closing of the NMPbind domain, even if no target conformation is provided in NMSim. Hence, the results in this study show that, incorporating directional information in the geometry-based approach NMSim improves the sampling of biologically relevant conformational space and provides a computationally efficient alternative to state of the art MD simulations.
Through the use of information about the biological target structure, the optimization of potential drugs can be improved. In this work I have developed a procedure that uses the quantitative change in the chemical perturbations (CSP) in the protein from NMR experiments for driving protein-ligand docking. The approach is based on a hybrid scoring function (QCSPScore) which combines traditional DrugScore potentials, which describe the interaction between protein and ligand, with Kendall’s rank correlation coefficient, which evaluates docking poses in terms of their agreement with experimental CSP. Prediction of the CSP for a specific ligand pose is done efficiently with an empirical model, taking into account only ring current effects. QCSPScore has been implemented in the AutoDock software package. Compared to previous methods, this approach shows that the use of rank correlation coefficient is robust to outliers. In addition, the prediction of native-like complex geometries improved because the CSP are already being used during the docking process, and not only in a post-filtering setting for generated docking poses. Since the experimental information is guaranteed to be quantitatively used, CSP effectively contribute to align the ligand in the binding pocket. The first step in the development of QCSPScore was the analysis of 70 protein-ligand complexes for which reference CSP were computed. The success rate in the docking increased from 71% without involvement of CSP to 100% if CSP were considered at the highest weighting scheme. In a second step QCSPScore was used in re-docking three test cases, for which reference experimental CSP data was available. Without CSP, i.e. in the use of conventional DrugScore potentials, none of the three test cases could be successfully re-docked. The integration of CSP with the same weighting factor as described above resulted in all three cases successfully re-docked. For two of the three complexes, native-like solutions were only produced if CSP were considered.Conformational changes in the binding pockets of up to 2 Å RMSD did not affect the success of the docking. QCSPScore will be particularly interesting in difficult protein-ligand complexes. They are in particular those cases in which the shape of the binding pocket does not provide sufficient steric restraints such as in flat protein-protein interfaces and in the virtual screening of small chemical fragments.