Computational prediction of the sites of metabolism (SOM) of protein kinase inhibitors

Introduction. Small molecule protein kinase inhibitors (KIs) are an effective targeted therapy for multiple types of cancers. KIs are mainly biotransformed through oxidation reactions catalysed by CYP3A4. SOM prediction is a useful tool for identifying metabolically labile sites of KIs (and other drugs) in the drug discovery pipeline.
Aims. This study sought to predict the SOM of KIs using a range of computational methods and to identify amino acids important for KI binding within the CYP3A4 active site.
Methods. SOMs were collated for a dataset of 31 marketed KIs metabolised by human CYP3A4. A range of computational approaches were evaluated for SOM prediction: molecular docking (using three CYP3A4 X-ray crystal structure templates); molecular superpositioning (using 4 ligand templates); and Web-based methods (using three algorithms). Molecular docking additionally identified amino acids involved in KI binding within the CYP3A4 active site.
Results. Since CYP3A4 is known to exhibit plasticity in the catalytic site, three X-ray crystal structures were investigated as templates for molecular docking. Docking in the bromoergocryptine-bound structure (3UA1) provided superior SOM prediction (77%) compared to the unliganded and ritonavir-bound structures (74% and 68%, respectively). Of the various scoring functions investigated, the PMF-score showed more consistent SOM prediction. The web-based SOM prediction algorithms provided marginally better predictivity (77%-87%), whereas the substrate superpositioning (molecular overlay) approach using 4 different compounds as templates was less effective (42%-71% prediction accuracy). Docking of the KIs in the CYP3A4 active sites identified Glu37 Phe57, Asp76, Arg105, Arg106, Ser119, Arg212, Phe215, Thr224, Arg372, and Glu374 as important residues for substrate binding.
Discussion. The study demonstrated that SOM prediction of KIs was dependent on the CYP3A4 X-ray crystal structure employed as the template, consistent with the known plasticity of this protein. Web-based SOM algorithms provided the best predictivity. Hydrophobic, hydrogen-bonding, and charge interactions contribute to KI binding in the CYP3A4 active site. The approaches adopted here are likely to be applicable to SOM prediction of other CYP3A4 substrates.
Figure 1