Technical Highlight - January 2012
Short description: Assessment of the biannual CASP experiment emphasizes the need for improved methods to model unique structural features.
The ultimate goal of protein structure prediction is to accurately model the three-dimensional fold of a protein from the primary sequence. This problem is far from trivial, and progress in the field is gauged by the biannual Critical Assessment of techniques for protein Structure Prediction (CASP) experiment. Participants aim to predict structures for protein targets that are about to be solved by traditional experimental methods (X-ray crystallography or NMR). The experimentally-determined structures are revealed at the end of the experiment.
Schwede and colleagues analyzed the results of the CASP9 experiment, which include contributions from the major structural genomics consortia. While over 80% of the 129 targets selected were from three consortia (PSI JCSG, MCSG and NESG), several individual groups also submitted targets from traditional hypothesis-driven projects. Comparison of the global features of predicted and experimental structures allowed an objective ranking, while closer examination of the experimental structures revealed determinants of functional specificity.
Two relatively small targets from the NESG highlighted different challenges in structure prediction. The structure of an uncharacterized domain from a human gut flora protein revealed a dimeric PDZ fold with low sequence similarity to other PDZ domains and distinct loop orientations — both factors creating a difficult prediction target. The PBS linker domain from the ApcE protein of a cyanobacterial photosystem also proved difficult to predict due to an exposed hydrophobic surface and many possible packing arrangements for its eight α-helices and two small β-strands.
The MCSG examined the structure of a kinase from an alternative glycolytic pathway in Klebsiella pneumoniae. Members of this protein family are composed of two domains that surround a deep active site groove. Although the global structure was correctly predicted, unique insertions and deletions resulted in an incorrectly predicted substrate-binding pocket. In a final illustrative example, the SGC solved crystal structures of two Hsp90-type co-chaperones from Plasmodium falciparum that have significant structural homology to yeast proteins. The reliable templates allowed for successful predictions despite low sequence homology.
Novel interactions or previously uncharacterized folds were significantly represented in targets submitted by individual research groups. These included a functionally critical coiled-coil, a unique disulfide connectivity pattern, a specific hydrophobic substrate binding pocket, and a fold that recognizes a DNA base unique to Trypanosoma brucei. The authors concluded that while sequence and structural conservation often allow successful predictions of the overall fold, there is a need for improvement in methods for predicting unexpected features and interactions, which are often the details critical to a protein's functional specificity.
V. Mariani et al. Assessment of template based protein structure predictions in CASP9.
Proteins. 79, 37-58 (2011). doi:10.1002/prot.23177
A. Kryshtafovych et al. Target highlights in CASP9: experimental target structures for the critical assessment of techniques for protein structure prediction.
Proteins. 79, 6-20 (2011). doi:10.1002/prot.23196