The substrate specificity of proteases underpins their diverse and crucial biological roles and provides a basis for the design of inhibitors that are commonly used as anti-viral drugs. Furthermore, proteases designed to target specific substrate sequences will be attractive therapeutic leads for specifically degrading any given disease-associated protein. We used the TEV protease, a commonly used laboratory reagent for the removal of affinity tags from recombinantly-expressed proteins, as a platform system to develop a structure-based predictive model for protease specificity. The enzyme functions as a highly specific cysteine protease and recognizes the canonical sequence Glu-Asn-Leu-Tyr-Phe-Gln-(Gly/Ser), with the cleavage occurring between the Gln-Gly or the Gln-Ser peptide bond. However, similar sequences can also be cleaved, and lists of cleavable and uncleavable substrate sequences have been experimentally determined. We used the Rosetta macromolecular modeling program to generate structural models of 70 cleaved and 342 uncleaved sequences, and used the components of the energy interaction between the protease and peptides to train support vector machines for developing a predictive classifier. On an independent test set, a preliminary classifier was able to predict cleaved sequences with 84.45% accuracy with a third degree polynomial kernel and with 81.46% accuracy with a radial based function kernel. Refinement and further testing of the model to recapitulate the specificities of variant proteases is ongoing. These studies may lead to the development of a new structure-based protocol for the design of proteases with novel specificities, which will serve as leads for a new class of therapeutic drugs.

Structure-based predictive model for the substrate specificity of the tobacco etch virus protease

247th National Meeting (2014)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.