Prediction and Experimental Evaluation of the hERG Blocking Potential of Drugs Showing Clinical Signs of Cardiotoxicity

Background: A large scale experimental validation conducted at the National Center for Advancing Translational Sciences (NCATS/NIH, USA) confirmed the predictions of our 3D-SDAR model of hERG blockage and phospholipidosis induction. It was demonstrated that both hERG blockage and phospholipidosis induction are driven by a common three-center toxicophore composed of two aromatic rings and an amino group. This work extends our earlier efforts by predicting the hERG blocking potential of pharmaceuticals from two additional datasets: i) one comprised of 106 drugs with reported clinical signs of cardiotoxicity from the AZCERT database and ii) a dataset of 54 FDA-approved tyrosine kinase inhibitors (TKIs). Methods: A bagging-like 3D-SDAR algorithm aggregating predictions from 100 randomized models was used to predict the hERG blocking potential of all 160 drugs. All 106 drugs from the AZCERT dataset were further evaluated for their hERG inhibition at NCATS using a thallium flux assay. Results: Comparison of the predicted hERG class against the results of the thallium flux qHTS assay resulted in an overall predictive accuracy of 0.736 and the area under the ROC curve of 0.780. Factors such as the generation of false negatives by the thallium flux assay, proximity to the cut-off, use of conformations that may differ from the biologically relevant ones, and the lack of structurally similar compounds in the modeling set could explain the somewhat reduced predictive performance compared to that of the original model. The original 3D-SDAR model was also used to evaluate the TKIs ability to block hERG. Comparing our predictions to class assignments based on IC50 values with a 30 μM cut-off, an accuracy of 0.850, sensitivity of 0.906, and specificity of 0.625 were achieved. Conclusion: 3D-SDAR provides a reliable platform for the prediction of hERG blockage. Particular attention should be paid to all investigational new drugs containing our three-center hERG toxicophore, especially those having highly flexible molecules. Particular scrutiny should be given to the tyrosine kinase inhibitors, which represent a therapeutic class possessing all structural characteristics previously associated with an increased potential to block hERG.