Article pubs.acs.org/jmc
Comprehensive Analysis of Three-Dimensional Activity Cliffs Formed by Kinase Inhibitors with Different Binding Modes and Cliff Mapping of Structural Analogues Norbert Furtmann,†,‡,§ Ye Hu,†,§ and Jürgen Bajorath*,† †
Department of Life Science Informatics, B-IT, LIMES Program Unit Chemical Biology and Medicinal Chemistry, Rheinische Friedrich-Wilhelms-Universität, Dahlmannstrasse 2, D-53113 Bonn, Germany ‡ Pharmaceutical Institute, Rheinische Friedrich-Wilhelms-Universität, An der Immenburg 4, D-53121 Bonn, Germany ABSTRACT: Kinases are among the structurally most extensively characterized therapeutic targets. For many kinases, X-ray structures of inhibitor complexes are publicly available. We have identified all threedimensional activity cliffs (3D-cliffs) formed by kinase inhibitors. More than 1300 X-ray structures of unique kinase-inhibitor complexes and associated activity data were analyzed. On the basis of binding mode comparison and 3D similarity calculations, 105 3D-cliffs were detected for type I, type II, or type III inhibitors of 13 different kinases. Many of these activity cliffs revealed clear interaction differences between highly and weakly potent inhibitors. More than 200 structural analogues of 3D-cliff compounds were identified whose structure−activity relationships (SARs) can be further explored in three dimensions on the basis of the corresponding 3D-cliffs. In addition to SAR exploration, 3D-cliffs provide useful interaction hypotheses for structure-based design. The kinase inhibitor and activity cliff information is made freely available as a part of our study.
■
by comparing compound binding modes6 and calculating their 3D similarity.6,7 These similarity calculations were based upon atomic property density functions by taking conformational, positional, and atomic feature differences into account.7,8 Furthermore, 216 3D-cliffs involving 38 targets belonging to 17 families (including selected kinases) were extracted from X-ray structures.9 This study has led to a first categorization of 3D-cliffs according to observed interaction differences between highly and weakly potent cliff forming compounds and enabled a comparison of 3D-cliffs and 2D-cliffs defined on the basis of fingerprint similarity calculations.9 It was shown that only a limited number of 3D-cliffs could be reconciled by molecular graph-based similarity assessment.9 Taken together, this limited number of studies has provided the first detailed views of 3D-cliffs on the basis of X-ray data. In this study, we have focused 3D-cliff analysis on protein kinases for which many X-ray structures of inhibitor complexes have become publicly available10 and have been organized according to various structural criteria.11 On the basis of structural data, different modes of action of kinase inhibitors have been characterized12,13 that can be related to compound binding patterns.13−16 We have aimed to systematically identify 3D-cliffs formed by currently available protein kinase inhibitors. To these
INTRODUCTION Activity cliffs are generally defined as pairs or groups of structurally similar or analogous compounds that share the same specific activity but have a large difference in potency.1,2 Structurally similar compounds might be identified by wholemolecule similarity calculations and are not necessarily structural analogues only distinguished by small chemical modification(s). Typically, activity cliffs are assessed on the basis of molecular graphs, especially in the practice of medicinal chemistry. To identify activity cliffs, compound similarity is mostly evaluated on the basis of 2D molecular representations,1,2 for example, by calculating similarity values using fingerprints (or other molecular descriptors) or by determining substructure relationships as a measure of similarity.1,2 Going beyond conventional 2D similarity assessment, only a few attempts have thus far been made to study activity cliffs in three dimensions on the basis of 3D structures of protein−ligand complexes.2 In an early study, exemplary activity cliffs were analyzed on the basis of crystallographic binding modes of compounds active against the same target to rationalize ligand− target interactions that might be responsible for cliff formation.3 In a different investigation, computed ligand−target interaction energies were used to generate interaction fingerprints for compounds from structures of complexes and identify activity cliffs by calculating interaction fingerprint similarity4 and structure−activity landscape index values.4,5 In another study, 3D activity cliffs (3D-cliffs) were obtained from X-ray structures of inhibitor complexes of β-secretase 1 and factor Xa © XXXX American Chemical Society
Special Issue: New Frontiers in Kinases Received: June 18, 2014
A
dx.doi.org/10.1021/jm5009264 | J. Med. Chem. XXXX, XXX, XXX−XXX
Journal of Medicinal Chemistry
Article
(assay-dependent) IC50 values. After removing structures containing only ionic ligands or duplicate complexes, 1023 complexes with available Ki and/or IC50 annotations remained. The potency annotations were manually curated by eliminating all approximate measurements such as “>”, “
