Date of Award
Ph.D. in Pharmaceutical Sciences
Robert J. Doerksen
John M. Rimoldi
Human protein kinases belong to a large and diverse enzyme family that contains more than 500 members. Deregulation of protein kinases is associated with many disorders, and this is why protein kinases are attractive targets for drug discovery. Due to the high conservation of the ATP binding pocket among this family, designing specific and/or selective inhibitors against certain member(s) is challenging. Several studies have been conducted on protein kinases to validate them as suitable drug targets. Although there are numerous target-validated protein kinases, the efforts to develop small molecule inhibitors have so far led to only a limited number of therapeutic agents and drug candidates.
In our studies, we tried to understand the basic structural features of protein kinases using available computational tools. There are wide structural variations between different states of the same protein kinase that affect the enzyme specificity and inhibition. Many protein kinases do not yet have an available X-ray crystal structure and have not yet been validated to be drug targets. For these reasons, we developed a new homology modeling approach to facilitate modeling non-crystallized protein kinases and protein kinase states. Our homology modeling approach was able to model proteins having long amino acid sequences and multiple protein domains with reliable model quality and a manageable amount of computational time. Then, we checked the applicability of different docking algorithms (the routinely used computational methodology in virtual screening) in protein kinase studies.
After performing the basic study of kinase structure modeling, we focused our research on cyclin dependent kinase 2 (CDK2) and glycogen synthase kinase-3β (GSK-3β). We prepared a non-redundant database from 303 available CDK2 PDB structures. We removed all structural anomalies and proceeded to use the CDK2 database in studying CDK2 structure in its different states, upon ATP, ligand and cyclin binding. We clustered the database based on our findings, and the CDK2 clusters were used to generate protein ligand interaction fingerprints (PLIF). We generated a PLIF-based pharmacophore model which is highly selective for CDK2 ligands. A virtual screening workflow was developed making use of the PLIF-based pharmacophore model, ligand fitting into the CDK2 active site and selective CDK2 shape scoring.
We studied the structural basis for selective inhibition of CDK2 and GSK-3β. We compared the amino acid sequence, the 3D features, the binding pockets, contact maps, structural geometry, and Sphoxel maps. From this study we found 1) the ligand structural features that are required for the selective inhibition of CDK2 and GSK-3β, and 2) the amino acid residues which are essential for ligand binding and selective inhibition. We used the findings of this study to design a virtual screening workflow to search for selective inhibitors for CDK2 and GSK-3β.
Because protein–protein interactions are essential in the function of protein kinases, and in particular of CDK2, we used protein–protein docking knowledge and binding energy calculations to examine CDK2 and cyclin binding. We applied this study to the voltage dependent calcium channel 1 (VDAC1) binding to Bax. We were able to provide important data relevant to future experimental researchers such as on the possibility of Bax to cross biological membranes and the most relevant amino acid residues in VDAC1 that interact with Bax.
Elokely, Khaled M., "Protein kinases: Structure modeling, inhibition, and protein-protein interactions" (2013). Electronic Theses and Dissertations. 1443.
Emphasis: Medicinal Chemistry