The sc-PDB is a collection of 6 415 three-dimensional structures of binding sites found in the Protein Data Bank (PDB). Binding sites were extracted from all high-resolution crystal structures in which a complex between a protein cavity and a small-molecular-weight ligand could be identified. Importantly, ligands are considered from a pharmacological and not a structural point of view. Therefore, solvents, detergents, and most metal ions are not stored in the sc-PDB. Ligands are classified into four main categories: nucleotides (< 4-mer), peptides (< 9-mer), cofactors, and organic compounds. The corresponding binding site is formed by all protein residues (including amino acids, cofactors, and important metal ions) with at least one atom within 6.5 angstroms of any ligand atom. The database was carefully annotated by browsing several protein databases (PDB, UniProt, and GO) and storing, for every sc-PDB entry, the following features: protein name, function, source, domain and mutations, ligand name, and structure. The repository of ligands has also been archived by diversity analysis of molecular scaffolds, and several chemoinformatics descriptors were computed to better understand the chemical space covered by stored ligands. The sc-PDB may be used for several purposes: (i) screening a collection of binding sites for predicting the most likely target(s) of any ligand, (ii) analyzing the molecular similarity between different cavities, and (iii) deriving rules that describe the relationship between ligand pharmacophoric points and active-site properties. The database is periodically updated and accessible on the web at http://bioinfo-pharma.u-strasbg.fr/scPDB/.

A novel method to measure distances between druggable protein cavities is presented. Starting from user-defined ligand binding sites, eight topological and physicochemical properties are projected from cavity-lining protein residues to an 80 triangle-discretised sphere placed at the centre of the binding site, thus defining a cavity fingerprint. Representing binding site properties onto a discretised sphere presents many advantages: (i) a normalised distance between binding sites of different sizes may be easily derived by summing up the normalised differences between the 8 computed descriptors; (ii) a structural alignment of two proteins is simply done by systematically rotating/translating one mobile sphere around one immobile reference; (iii) a certain degree of fuzziness in the comparison is reached by projecting global amino acid properties (e.g., charge, size, functional groups count, distance to the site centre) independently of local rotameric/tautomeric states of cavity-lining residues. The method was implemented in a new program (SiteAlign) and tested in a number of various scenarios: measuring the distance between 376 related active site pairs, computing the cross-similarity of members of a protein family, predicting the targets of ligands with various promiscuity levels. The proposed method is robust enough to detect local similarity among active sites of different sizes, to discriminate between protein subfamilies and to recover the known targets of promiscuous ligands by virtual screening.

Quantification of local similarity between protein 3D structures is a promising tool in computer-aided drug design and prediction of biological function. Over the last ten years, several computational methods were proposed, mostly based on geometrical comparisons. This review summarizes the recent literature and gives an overview of available programs.A particular interest is given to the underlying methodologies. Our analysis points out strengths and weaknesses of the various approaches. If all described methods work relatively well when two binding sites obviously resemble each other, scoring potential solutions remains a difficult issue, especially if the similarity is low. The other challenging question is the protein flexibility, which is indeed difficult to evaluate from a static representation. Last, most of recently developed techniques are fast and can be applied to large amounts of data.Examples were carefully chosen to illustrate the wide applicability domain of the most popular methods: detection of common structural motifs, identification of secondary targets for a drug-like compound, comparison of binding sites across a functional family, comparison of homology models, database screening. /top Single representative entry of ProSurfer database ProSurfer database: ~ 48,000 sites a SiteAlign http://bioinfo-pharma.ustrasbg.fr/template/jd/pages/download/ download.php user-defined site or sc-PDB site user-defined site(s) or sc-PDB site(s) c Ligand binding sites were extracted from the PDB a , the non-redudant PDB b or the sc-PDB dataset c [54].

Predicting off-targets by computational methods is getting increasing importance in early drug discovery stages. We herewith present a computational method based on binding site three-dimensional comparisons, which prompted us to investigate the cross-reaction of protein kinase inhibitors with synapsin I, an ATP-binding protein regulating neurotransmitter release in the synapse. Systematic pair-wise comparison of the staurosporine-binding site of the proto-oncogene Pim-1 kinase with 6,412 druggable protein-ligand binding sites suggested that the ATP-binding site of synapsin I may recognize the pan-kinase inhibitor staurosporine. Biochemical validation of this hypothesis was realized by competition experiments of staurosporine with ATP-γ35S for binding to synapsin I. Staurosporine, as well as three other inhibitors of protein kinases (cdk2, Pim-1 and casein kinase type 2), effectively bound to synapsin I with nanomolar affinities and promoted synapsin-induced F-actin bundling. The selective Pim-1 kinase inhibitor quercetagetin was shown to be the most potent synapsin I binder (IC50  = 0.15 µM), in agreement with the predicted binding site similarities between synapsin I and various protein kinases. Other protein kinase inhibitors (protein kinase A and chk1 inhibitor), kinase inhibitors (diacylglycerolkinase inhibitor) and various other ATP-competitors (DNA topoisomerase II and HSP-90α inhibitors) did not bind to synapsin I, as predicted from a lower similarity of their respective ATP-binding sites to that of synapsin I. The present data suggest that the observed downregulation of neurotransmitter release by some but not all protein kinase inhibitors may also be contributed by a direct binding to synapsin I and phosphorylation-independent perturbation of synapsin I function. More generally, the data also demonstrate that cross-reactivity with various targets may be detected by systematic pair-wise similarity measurement of ligand-annotated binding sites.