Computational tools for the high-throughput identification of protein-targeted drugs and probes

Abstract

This thesis is comprised of three projects that are driven by a common theme, which is the use of computational tools in aiding molecular probe and drug design. In the first project, the feasibility of using molecular docking and scoring to estimate binding affinity for small molecules labelled covalently with fluorophores was tested using several proof-of-concept experiments. The high-throughput nature of computational screening applications such as Hierarchical Virtual Ligand Screening (HierVLS) necessitate that, in order to screen these labelled compounds, there must be an automated way to generate the associated structures virtually from large databases of base compounds and fluorophores. A script was developed in MOE software using scientific vector language (SVL) that could identify key reactive functional groups in both reactive fluorophores and target base compounds, and create the appropriate labelled structures for screening. The final fluorescence-labelled database numbers 14,862 compounds, each tagged with the ATTO680 fluorophore. In a subsequent project, the fluorescence-tagged library was screened against carbonic anhydrase 9 (CAIX), a protein implicated as a biomarker in several cancer types. This screening was accompanied by the screening of a validation set of known CAIX ligands and appropriately chosen decoys. The best scoring protocol according to our analyses was that which used principal components analysis. Ten of the top scoring candidates are suggested for future testing as probe candidates. CAIX binding sites were compared with equivalent residues in the sequences of 24 other CA isoforms to identify sites that might confer CAIX specificity, and the top scoring ligands were ranked according to this scheme. Lastly, experimental characterization was performed on three previously identified potential ligands for a cancer-related receptor tyrosine kinase, EphB4. Two in vitro assay formats were used: a homogenous time-resolved fluorescence assay and an enzyme-coupled spectrophotometric assay. One candidate, DNP-L-Arg, was the only one of the three with some experimental evidence of affecting kinase activity. The first assays suggested that DNP-L-Arg may have an activating effect on EphB4. The plausibility of this effect discussed with respect to mechanisms found in the literature, and using predicted and experimental structures for docked ligands. The coupled assay format did not conclusively confirm this effect. The research presented underscores the ability for computational tools to be incorporated into a variety of different areas within the fields of biochemistry and drug design. Future complementary experimental work will be crucial both in evaluating and refining the suggested probe candidates and in further validating and improving the computational techniques used.