Modellierung von Metalloenzymen: 3D-QSAR-Untersuchungen an Carboanhydrase-Isoenzymen und virtuelles Screening nach Peptiddeformylase-Inhibitoren

The modern drug development process is usually divided into the phase of lead discovery and lead optimization. This thesis provides contributions to both aspects. The first part of this work deals with the establishment and validation of computer models for prediction of affinity and selectivity, thus it supports the area of lead optimization. Selectivity aspects are a major concern in drug development since they influence the risk of undesired side effects. QSAR approaches have been used for prediction of affinity and selectivity parameters using carbonic anhydrases (CAs) as model system. CAs are zinc containing hydrolases which catalyze the reversible hydration of carbon dioxide to bicarbonate and a proton. They are thus involved into various (patho)physiological processes and constitute interesting drug targets. The binding pockets of the numerous isozymes are rather similar in terms of physicochemical properties making the development of selective inhibitors quite difficult. The investigations were especially focused on 3D-QSAR models. Statistically highly significant and robust models could be established in order to predict affinity and selectivity of sulfonamide inhibitors toward CA I, II, and IV. The investigations revealed that the minor differences in the alignments in the three binding pockets produce only small differences in the statistical outcome. The best models were obtained using the CA II pocket as basis of the alignment, irrespective of the variable to be modeled. Probably the large number of X-ray structures available for CA II is responsible for this effect making the alignment more reliable. The isocontour maps obtained enable an interpretation of the models in terms of physicochemical properties important for affinity/selectivity. The comparison to qualitative protein-based contour maps underlines the complementary character of both approaches: The ligand based QSAR methods on the one hand implicitly reflect the structure of the binding site partially, on the other hand their outcome depends on the peculiarities of the training set. Protein-based approaches, however, also provide information about regions of the binding pocket which are not occupied by any of the training set ligands. Another objective was the utilization of QSAR models for database screening. This would allow one to identify the most interesting (i.e. potent/selective)compounds of a database as candidate for synthesis in the context of lead optimization. First, a protocol for automation of ligand alignment had to be developed and validated for 3D-QSAR. The ligand-based alignment procedure applied turned out to yield results comparable to the manual protein-based alignment. The 3D models performed superior to 2D fragment-based or 1D property-based models. As a practical example of application a virtual compound library composed by several thousand entries was designed and ranked using the most predictive models. The second part of the thesis covered a virtual screening for novel inhibitors of peptide deformylases (PDFs) hence being a task of lead discovery. PDFs are (mostly) ferrous ion containing enzymes cleaving off the formyl group from proteins synthesized in mitochondria, plastids, and eubacteria. 3D-Pharmacophore models have been established and validated using X-ray structures of potent PDF-inhibitors. These pharmacophores were able to identify structurally diverse inhibitors known from literature (sufficient sensitivity) and to strongly reduce the database to be screened (sufficient specificity). The pharmacophore models were used for screening databases of commercially available druglike compounds. By additional application of docking and scoring eleven out of approximately two million compounds have been identified and purchased. These molecules will be subjected to biological testing. First testing results show that at least two of the selected compounds are potent inhibitors against PDF1B from E. coli exhibiting IC50-values of 60 nM and 190 nM, respectively. This can be considered as a proof of the quality and suitability of the employed models and the screening protocol. PDF inhibitors could be used as herbicides, antibiotics, and antimalarial therapeutics.