Structure based prediction and identification of novel peptidase substrates using computational protein design

Rapid and accurate prediction of the extended substrate specificity of proteases would greatly aid in developing custom proteases that cleave biotechnologically or pharmaceutically relevant substrates, and for uncovering targets of therapeutically relevant protease enzymes.  However,current in silico approaches for specificity prediction rely on pattern recognition in known experimental data and are thus limited by the availability of experimental cleavage data for each enzyme variant.  We present a method for predicting peptidase substrates de novo based on the structure-guided modeling and biophysical evaluation of peptide-peptidase complexes.  We develop a new structure-guided sequence sampling technique based on self-consistent mean field theory for rapid specificity profile prediction and show that computationally calculated specificity profiles closely match the experimentally determined ones.  Further, we develop adiscriminatory scoring function, using the macromolecular modeling software Rosetta and Amber-MMPBSA, that ranks putative peptide substrates in order ofcalculated interaction energy with the peptidase active site and demonstrate its (i) accuracy and generality: by evaluating its ability to classify cleaved and uncleaved peptides for all major protease mechanistic classes–serine, cysteine, aspartyl and metallo-proteases, (ii) higher classification performance compared to sequence-based statistical inference methods, and (iii) predictive ability: by designing and experimentally testing novel substrates for the HCV NS3 proteases.  The presented approach should allow the identification of novel targets of newly discovered, functionally uncharacterized peptidases, and, conversely, permit the design of “protein knockouts” using designed peptidase enzymes targeting chosen substrates to conditionally inactivate toxic or unwanted proteins.