KLIFS: a structural kinase-ligand interaction database

Abstract

Protein kinases play a crucial role in cell signaling and are important drug targets in several therapeutic areas. The KLIFS database contains detailed structural kinase-ligand interaction information derived from all (>2900) structures of catalytic domains of human and mouse protein kinases deposited in the Protein Data Bank in order to provide insights into the structural determinants of kinase-ligand binding and selectivity. The kinase structures have been processed in a consistent manner by systematically analyzing the structural features and molecular interaction fingerprints (IFPs) of a predefined set of 85 binding site residues with bound ligands. KLIFS has been completely rebuilt and extended (>65% more structures) since its first release as a data set, including: novel automated annotation methods for (i) the assessment of ligand-targeted subpockets and the analysis of (ii) DFG and (iii) αC-helix conformations; improved and automated protocols for (iv) the generation of sequence/structure alignments, (v) the curation of ligand atom and bond typing for accurate IFP analysis and (vi) weekly database updates. KLIFS is now accessible via a website (http://klifs.vu-compmedchem.nl) that provides a comprehensive visual presentation of different types of chemical, biological and structural chemogenomics data, and allows the user to easily access, compare, search and download the data.

Figure 1.

The data collection, processing, annotation and analysis workflow of KLIFS.

Figure 2.

Annotation structural kinase-ligand interaction data in KLIFS. (A) Consistent structure-based kinase binding site residue annotation (top) and careful curation of the chemical topology and protonation of kinase ligands (bottom) enable the systematic analysis of kinase-ligand IFPs of kinase-ligand complexes (middle), illustrated for 7 of the 85 binding site residues of a phthalazine inhibitor (PDB ligand identifier: A17) bound p38a structure (PDB: 3DS6). (B) Automated assessment of ligand-targeted subpockets in protein kinase binding sites. Spatial probes (0.5 Å grid spacing) are: (i) placed around the conformations of kinase-bound ligands that are superposed according to the structural alignment of the corresponding binding sites; (ii) scored according to the ratio of close contacts (<1.0 Å) with training sets of ligands that bind a specific subpocket and ligands that do not bind this subpocket; (iii) highest ranked probes are selected as descriptors to assign subpockets for new kinase-ligand complexes. (C) A 10-fold cross-validated decision tree model derived from systematic analysis of the Accessible Surface Area (ASA) (32) of kinase binding site residues and vectors defining the directionality of D^xDFG.81 and F^xDFG.82 side chains enable the automatic assessment of the DFG conformation of kinase structures with an accuracy > 99%. The decision tree algorithm identified the ASA values of D^xDFG.81 and y-values of D^xDFG.81 and F^xDFG.82 vectors (obtained by subtracting the coordinates of Cγ by Cα) as the most predictive descriptor combination for the discrimination of DFG conformations.

Figure 3.

KLIFS entry example (sorafenib (BAX) bound KDR, PDB: 3WZE), schematic overview of KLIFS website search options (gray boxes) and 3D kinase-ligand binding mode visualization (overlay sorafenib gray carbon atoms, ligand from search example PDB: 2BAK magenta carbon atoms). All kinases with an entry in KLIFS are indicated in the interactive human kinome (red dots), and search results can also be highlighted. Top right panels: sorafenib binding mode views showing (color-coded) secondary structure elements, sliced binding site surface and user-customizable 3D-rendering settings (blue box). Further details regarding structural information, ligand search/binding mode/affinity, pocket alignment/kinase-ligand interaction pattern browse/search options are explained in the text.