Visualization of [A] actual and [B] modeled three-dimensional structure of SSH-2 [2nt2] from the improved workflow. [C] is the overlay of the actual and modeled structures. Note the very close agreement of the two structures. The colors have been changed [grey=actual, orange=modeled] clarity.

GRID-BASED 3-D PROTEIN MODEL,
BUILDING FOR SMALL MOLECULE DOCKING

Daniel Li and Brian Tsui, PRIME 2011

In previous years, PRIME students Charles Xue (2010) and Matthew K. Mui (2009) have identified possible chemical inhibitors of the slingshot-2 (SSH-2) protein using high-throughput docking. This protein is part of the large dual specificity phosphatase (DSP) family, which consists of seven subfamilies that exhibit high homology with a characteristic HX5R(S/T) catalytic site that is able to dephosphorylate phosphoserine/threonine or tyrosine residues. SSH-2 is a regulatory enzyme that affects the actin depolymerizing protein cofilin, which in turn controls various cellular processes and makes it an invaluable protein target for pharmaceutical studies. However, high homology within the DSP family requires thorough screening of each DSP member to help determine SSH-2 inhibitor specificity. Due to the time and complexity required to do this, only about one-third of the DSP family have experimentally-determined protein structures, therefore, it was vital to create accurate 3-D protein structure models in order to obtain a complete database for virtual docking of the remaining DSP family members.

Utilizing a grid-enabled 3-D modeling program, MODELLER, we created a streamlined workflow that can rapidly produce accurate 3-D models of DSPs that do not have known three-dimensional structures. These models were then applied to further docking studies in order to complete screening of the entire DSP family.

The modeling workflow consisted of four major steps:

1. A protein blast was applied in order to find template sequences with similar alignment to the protein of interest.
2. Top candidates were selected and arranged into groups based on phylogenetic similarity and applied to a target-template sequence and structural alignment via MODELLER. The program accounted for global and local atom pair distances, as well as solvent accessibility at the residues.
3. The top alignment was generated and selected through MODELLER and a set of 600 protein structure models was created. Each model was built through distance and angle restraints provided by the alignment with the template structures. The model structures were evaluated through GA341, a composite score that helps distinguish between good and bad protein folds, as well as an energy score evaluation via DOPE and molpdf scores. Additional models were generated if the results did not meet a predetermined cutoff (0.7 out of 1).
4. Finally, the best 3-D structure models were optimized via loop refinement (MODELLER) and energy minimization (Chimera). MolProbity analysis was then applied to verify that minimal intra-atomic clash and realistic protein folds were present in our modeled 3-D structure and the protein models were ready for subsequent docking applications.

Through this workflow, 43,800 total protein structure models were produced in just four weeks time, completing the database of DSP structures. Docking experiments are now underway to test for SSH-2 inhibitor specificity.

PARTICIPATING RESEARCHERS: Osaka University: Susumu Date, Kohei Ichikawa; UCSD: Jason H. Haga

QUICK LINKS