Postdoctoral Research Initiatives

Virtual Screening

Jon Robertus, PhD - Department of Chemistry & Biochemistry, The University of Texas at Austin
Chandrajit Bajaj, PhD - Department of Computer Sciences, The University of Texas at Austin
David Gorenstein, PhD - Department of Biochemistry & Molecular Biology, The University of Texas Medical Branch

Virtual screening is the process of docking three-dimensional models of drug-like compounds into three dimensional models of potential drug receptors, usually proteins (Jorgensen, 2004; Shoichet, 2004). Such computational methods are far faster, and cheaper, than physically testing tens of thousands of potential drugs in chemical or cell-based assays, which has been a standard in the pharmaceutical industry for generations. For example, a study was carried out searching for inhibitors of tyrosine phosphatase-1b using both real and virtual screening methods (Doman et al, 2002). Some 400,000 compounds from the corporate library were tested in a high throughput screen (HTS); 85 (0.02%) showed inhibitory, IC50, values of 100mM or better, with a top score of 4 mM. In parallel, a commercial data base of 230,000 compounds was used in a virtual docking screen; it suggested 365 compounds were worthy of testing, and of those, 127 (35%) had IC50 values of 100 mM or less, with a best score of 1.7 mM. This key study suggested that the inhibitor hit rate for virtual screening was 1700 times as efficient as for a random HTS screen, at least in this case. The enhancement of virtual screening hit rates over random testing is referred to as the “enrichment” parameter.

Virtual screening is an important component of the TI-3D program for drug design. Such screening is computationally intense and TI-3D has acquired a powerful computer cluster to facilitate the method. They recently purchased an HP server with 32 dual-core AMD processors. Each of the 64 independent cores has 2 GB of RAM, and data are stored in a state of the art 1.3 TB RAID5 backup system. In addition, TI-3D has purchased a perpetual license to use the GOLD virtual screening software, one of the most highly regarded and flexible systems currently available (Jones et al, 1997; Verdonk et al, 2003). The screening facility also has academic licenses for several other software packages including eHiTS (Zsoldos et al, 2006) and SurFlex (Jain, 2003).

Screening Flavivirus targets
The promise of virtual screening for inhibitor design has attracted a number of collaborative efforts. One of the most important is the joint effort between University of Texas groups at Austin and at the Medical Branch in Galveston. The overall goal of that project is to identify compounds that can inhibit the binding of flaviviruses to host cells.

Flaviviruses are the cause of a number of human diseases, including dengue fever, yellow fever, West Nile disease, and various hemorrhagic fevers and encephalitic diseases. The viruses have icosahedral symmetry; a surface protein called E has been identified as the receptor for human cells. The E protein is fairly large, about 500 residues in length, and is composed of four domains. Domain 3, D3, has been identified as the actual receptor (Hung et al, 2004; Chu et al, 2005). It has a classic antibody fold, and the portions of the ED3 which bind the target cell are generally homologous to the CR regions of an antibody. There is evidence that ED3 binds at the 5-fold axis of the virus surface, which suggests that a pentamer of ED3 units may be the target cell receptor. If this site could be blocked by a specific, tight binding small molecule, it would likely act as an antiviral agent.

Dr. Gorenstein and colleagues (UTMB) used NMR to solve the three-dimensional structure of ED3 from West Nile virus (WNV), dengue fever virus (DFV) and from Omsk hemorrhagic fever virus (OHFV) (Volk et al, 2006). The monomer is shown as a ribbon drawing in Figure 1. The loops near the top of the molecule are the regions that interact with cell surface receptors; this is the region that will serve as the initial template for virtual screening. The UTMB team have created a model of the virus surface based on fitting this atomic structure to cryoEM images of the virus. The arrangement of the ED3 domains around the 5-fold and 3-fold virus operators will be an additional screening target. Inhibitors that contain higher symmetry and interact across these natural interfaces will bind particularly strongly to the virus capsid.

Figure 1: A ribbon drawing of OHFV ED3 protein. The inhibitor binding surface is along the top loops.

The virtual screening will be carried out in Austin by the Robertus group. Several chemical libraries will be screened, including the ChemBridge diversity set, SigmaAldrich, and Maybridge; these are available through the NIH-supported ZINC site (Irwin, 2005). Each library contains about 50,000 compounds. There is evidence to suggest that available software does a very good job of predicting inhibitor docking modes (Kellenberger et al, 2004; Kontoyianni et al, 2004; Perola et al, 2004) but that accurate ranking of binding is more problematic (Warren et al, 2006). One approach is to use several programs to screen each library and to look for compounds ranked highly by more than one. Our experience with a model system (ricin) showed that this method can greatly enrich the number of inhibitor hits from a list of candidates.

The physical screening of compounds will be carried out at UTMB, using either NMR binding assays, or high throughput plaque assays.