| Specificity is key to the function of biologically active molecules, including messengers, hormones, and drugs. Perplexingly, many small-molecule enzyme inhibitors show poor specificity. To investigate the origins of inhibitor promiscuity, 110 compounds were studied. These compounds included hits from virtual and high-throughput screening, leads for drug design, and clinically-used drugs. Fifty of these compounds (35 hits, 8 leads, and 7 drugs) were found to inhibit diverse model enzymes, including beta-lactamase, chymotrypsin, dihydrofolate reductase, malate dehydrogenase, and beta-galactosidase. These promiscuous inhibitors acted in a time-dependent manner that was sensitive to enzyme concentration, ionic strength, and albumin. By light scattering and electron microscopy, promiscuous compounds were observed to form spherical particles of 30–1000 nm diameter. Particles were absent from solutions of the 60 non-promiscuous compounds (10 hits, 7 leads, and 43 drugs). Based on these observations, it was hypothesized that the promiscuous compounds formed aggregates in solution, and the aggregates inhibited the model enzymes. To explore the mechanism of aggregate-based inhibition, biophysical and microscopy studies were performed. Co-sedimentation experiments suggested a direct interaction between enzyme and aggregrates; by electron microscopy, enzyme molecules were observed to associate with the surface of aggregates. This interaction was prevented by Triton X-100, which reversed enzyme inhibition by promiscuous compounds but not by specific inhibitors. These studies suggest that inhibition by aggregate-forming promiscuous compounds results from the reversible adsorption of enzyme onto the surface of aggregates, although absorption of enzyme into the aggregate interior cannot be excluded. This dissertation also provides methods for the identification of aggregate-forming promiscuous compounds, which appear to be widespread.; In its final chapter, this dissertation returns to the problem of modeling specificity in molecular docking. To address the influence of receptor representation on virtual screening, a crystallographic ligand-bound (holo), crystallographic ligand-free (apo), and a homology modeled representation of each of ten different enzyme systems were used to screen a database of 96,000 small molecules. Each structure was evaluated for its ability to discriminate between ligands and decoys. In seven of ten systems, the holo structure provided the best discrimination; the apo representation was more successful if the holo form was overdetermined. |
