Terfenadine, an antihistamine drug widely prescribed in the 1990s, was found to exhibit cardiotoxicity due to hERG channel inhibition only after it came to market. Terfenadine undergoes metabolism to Fexofenadine, which is also a potent histamine antagonist but does not inhibit hERG. This metabolite replaced the parent drug after it was famously withdrawn from the market in 1998.
While the hERG-channel inhibition assay has become one of the standard early tests in drug discovery, the backlog of early screening can be alleviated using in silico tools. Running these two compounds through the ACD/Tox Suite hERG Inhibitors module, we see that Terfenadine would be expected to be a hERG inhibitor (predicted hERG inhibition probability 0.98) while Fexofenadine would not (predicted probability 0.16), both predictions being highly reliable (Figure 2).
Genotoxicity of chemicals also often involves metabolic activation. For example, polycyclic aromatic hydrocarbons are only weak mutagens, but their metabolites, mostly epoxides, are strong mutagens. Therefore, genetic toxicity experiments, like the Ames test, examine compounds with and without metabolic activation. Given this fact, ACD/Tox Suite predicts genotoxicity while taking metabolic activation into account.
When using in silico tools, it is important to have realistic expectations of results. During the evaluation, one should keep in mind that a QSAR model is only as good as the quality of the associated data and is applicable only for the chemical space covered by the training set. For some desired endpoints there may be little available experimental data upon which to build a viable model.
When enough data is available, its quality may be insufficient to build a reliable model. The best scenario for building a predictive model is to use a global dataset, with local models “sitting on top” applying corrections for a series of structurally similar compounds. Ideally, the user could add in-house data to improve prediction accuracy. The trainable model feature of many ACD/ADME Suite and ACD/Tox Suite modules allows such customization of models.
Reliability of predictive models will continue to be an area for development and improvement. With drug discovery and development organizations facing increasing pressure to deliver investigational new drugs with favorable safety and toxicity metabolite profiles, we anticipate greater application of in silico tools. They can support experimental efforts by predicting superior compounds for synthesis and reduce the burden on initial high-throughput screens.
These and other examples suggest that early prediction of possible metabolites and their properties may alert researchers to safety risks not directly associated with the drug substance itself but with products of its biotransformation. To avoid failures, especially in late development or after-market, the key would be to use all the tools at our disposal for metabolite safety research, including predictive software.