Predictions Of P-gp Inhibitors/Substrates And Molecular Dynamics Simulations

The over-expression of P-gp leads to the multidrug resistance (MDR) of tumorcells, which is a major obstacle to effective treatment of cancer. Thus, designingeffective P-gp inhibitors is one of the important contents in anti-cancer drug designs. Inthis paper, homology modeling, QSAR, molecular docking and molecular dynamicssimulations were employed to develop classification models and explore the interactionmodes of P-gp inhibitors and substrates. Therefore, the results obtained in this paper areof both theoretical significance and practical values in the screening of P-gp inhibitors.The main contents and achievements:â‘ Two human P-gp homology models templated by the crystal structures of miceP-gp were constructed and then optimized by Amber force field. The results of modelevaluations showed that the P-gp homology models are reasonable and reliable inconsideration of structure and energy aspects, and can be further applied in theresearches of molecular simulation.â‘¡Both SVM and molecular docking were used to establish predictive models ofP-gp inhibitors. According to the results of10-fold cross-validation, an optimal linearSVM model with only3descriptors was derived on857training samples, of which theoverall accuracy, sensitivity, specificity, and MCC were0.840,0.873,0.813, and0.683,respectively. The SVM model was further validated by418test samples with the overallaccuracy, sensitivity, specificity, and MCC of0.868,0.938,0.738, and0.704,respectively. The results of surflex-dock showed that the total scores have a satisfiedperformance in distinguishing P-gp inhibitors from non-inhibitors. Further studies showthat SVM and surflex-dock gave consensus prediction results for77.3%of the samples,of which the overall accuracy is93.1%. At the same time, the docking results showedthat Phe699, Phe303, Phe695, Phe945and Phe39are important binding sites of P-gpinhibitors.â‘¢Both SVM and molecular docking were used to establish classification modelsof P-gp substrates. According to the results of10-fold cross-validation, an optimalclassification model with5descriptors was obtained, of which the overall accuracy,sensitivity, specificity, and MCC of1731training samples were0.69,0.85,0.47, and0.34, respectively. The SVM model was further validated by193test samples with theoverall accuracy, sensitivity, specificity, and MCC of0.71,0.84,0.49and0.35, respectively. The results of molecular docking showed that the total scores can notdistinguish P-gp substrates from non-substrates. In addition, there is no significantdifference in the binding sites between substrates and non-substrates.â‘£Based on the homology models established,10ns molecular dynamicssimulations of membrane-embedded P-gp was performed to investigate the influence ofcell membrane and ligands on P-gp conformation. The results of molecular dynamicssimulations showed that the structure of membrane-embedded P-gp was stabilized after10ns molecular dynamics simulations. Then, P-gp substrates and non-substrates(n=1929) were docked into the P-gp with the lowest energy conformation once again.The results showed that the total scores of surflex-dock are still unable to effectivelydistinguish P-gp substrates from non-substrates. The pool performance of surflex-dockmay be caused by structural diversities and different interaction modes of P-gpsubstrates.