Computational Prediction of the Site(s) of Metabolism and Binding Modes of Protein Kinase Inhibitors Metabolized by CYP3A4

Protein kinase inhibitors (KIs), which are mainly biotransformed by CYP3A4-catalyzed oxidation, represent a rapidly expanding class of drugs used primarily for the treatment of cancer. Ligand- and structure-based methods were applied here to investigate whether computational approaches may be used to predict the site(s) of metabolism (SOM) of KIs and to identify amino acids within the CYP3A4 active site involved in KI binding. A data set of the experimentally determined SOMs of 31 KIs known to undergo biotransformation by CYP3A4 was collated. The structure-based (molecular docking) approach employed three CYP3A4 X-ray crystal structures to account for structural plasticity of this enzyme. Docking pose and SOM predictivity were influenced by the X-ray crystal template used for docking and the scoring function used for ranking binding poses. The best prediction of SOM (77%) was achieved using the substrate (bromoergocryptine)-bound X-ray crystal template together with the potential of mean force score. Binding interactions of KIs with CYP3A4 active site residues were generally similar to those observed for other substrates of this enzyme. The ligand-based molecular superposition approach, using bromoergocryptine from the X-ray cocrystal structure as a template, poorly predicted (42%) the SOM of KIs, although predictivity improved to 71% when the docked conformation of sorafenib was used as the template. Among the web-based approaches examined, all web servers provided excellent predictivity, with one web server predicting the SOM of 87% of the data set molecules. Computational approaches may be used to predict the SOM of KIs, and presumably other classes of CYP3A4 substrates, but predictivity varies between methods.