Predicting the Interactions between PDZ Adapter Domains and Disordered Peptides

Abstract

PDZ domains, one of the most ubiquitous and important scaffolding modules in human proteins, bind the disordered C-terminus of plasma membranes, mediating protein-protein interactions. Experiments have demonstrated that dissimilar C-terminal peptides bind to the same PDZ domain and different PDZs can bind the same peptides. Crystallographic studies revealed that binding to the PDZ domains requires a four residue long strand anchored by a C-terminal hydrophobic residue. Based on this information, we developed a novel semi-flexible docking method to model the peptide-PDZ complex structure and estimate its absolute affinity. The method has been tested on a set of 126 15-residue long natural peptides binding to PDZ3 of PSD95. The resulting sensitivity and specificity rates were 90.91%/79.13% by defining a kinetic and a thermodynamic threshold. Moreover, complex structures of 5 different peptides bound to PDZ domains were successfully recovered as the top ranked predicted models. This general structure-based technology is the first de novo approach to dock disordered peptides, providing a needed complementarity to proteomic assays to mine GeneBank for new targets of scaffold proteins and to predict novel protein-protein interactions. Our findings also reveal that the four-residues C-terminal recognition motif leads to only a weak non-specific binding intermediate complex, while an extended network of contacts established by the next three to five unconstrained residues determines the high specificity of the complex.