IntroductionDetection of drug-induced pro-arrhythmic risk is a primary concern for pharmaceutical companies and regulators. Increased risk is linked to prolongation of the QT interval on the body surface ECG. Recent studies have shown that multiple ion channel interactions can be required to predict changes in ventricular repolarisation and therefore QT intervals. In this study we attempt to predict the result of the human clinical Thorough QT (TQT) study, using multiple ion channel screening which is available early in drug development.MethodsIon current reduction was measured, in the presence of marketed drugs which have had a TQT study, for channels encoded by hERG, CaV1.2, NaV1.5, KCNQ1/MinK, and Kv4.3/KChIP2.2. The screen was performed on two platforms — IonWorks Quattro (all 5 channels, 34 compounds), and IonWorks Barracuda (hERG & CaV1.2, 26 compounds). Concentration–effect curves were fitted to the resulting data, and used to calculate a percentage reduction in each current at a given concentration.Action potential simulations were then performed using the ten Tusscher and Panfilov (2006), Grandi et al. (2010) and O'Hara et al. (2011) human ventricular action potential models, pacing at 1 Hz and running to steady state, for a range of concentrations.ResultsWe compared simulated action potential duration predictions with the QT prolongation observed in the TQT studies. At the estimated concentrations, simulations tended to underestimate any observed QT prolongation. When considering a wider range of concentrations, and conventional patch clamp rather than screening data for hERG, prolongation of ≥ 5 ms was predicted with up to 79% sensitivity and 100% specificity.DiscussionThis study provides a proof-of-principle for the prediction of human TQT study results using data available early in drug development. We highlight a number of areas that need refinement to improve the method's predictive power, but the results suggest that such approaches will provide a useful tool in cardiac safety assessment.

Cell lines expressing ion channels (IC) and the advent of plate-based electrophysiology device have enabled a molecular understanding of the action potential (AP) as a means of early QT assessment. We sought to develop an in silico AP (isAP) model that provides an assessment of the effect of a compound on the myocyte AP duration (APD) using concentration-effect curve data from a panel of five ICs (hNav1.5, hCav1.2, hKv4.3/hKChIP2.2, hKv7.1/hminK, hKv11.1). A test set of 53 compounds was selected to cover a range of selective and mixed IC modulators that were tested for their effects on optically measured APD. A threshold of >10% change in APD at 90% repolarization (APD(90)) was used to signify an effect at the top test concentration. To capture the variations observed in left ventricular midmyocardial myocyte APD data from 19 different dogs, the isAP model was calibrated to produce an ensemble of 19 model variants that could capture the shape and form of the APs and also quantitatively replicate dofetilide- and diltiazem-induced APD(90) changes. Provided with IC panel data only, the isAP model was then used, blinded, to predict APD(90) changes greater than 10%. At a simulated concentration of 30 μM and based on a criterion that six of the variants had to agree, isAP prediction was scored as showing greater than 80% predictivity of compound activity. Thus, early in drug discovery, the isAP model allows integrating separate IC data and is amenable to the throughput required for use as a virtual screen.

IntroductionUnwanted drug interactions with ionic currents in the heart can lead to an increased proarrhythmic risk to patients in the clinic. It is therefore a priority for safety pharmacology teams to detect block of cardiac ion channels, and new technologies have enabled the development of automated and high-throughput screening assays using cell lines. As a result of screening multiple ion-channels there is a need to integrate information, particularly for compounds affecting more than one current, and mathematical electrophysiology in-silico action potential models are beginning to be used for this.MethodsWe quantified the variability associated with concentration-effect curves fitted to recordings from high-throughput Molecular Devices IonWorks® Quattro™ screens when detecting block of IKr (hERG), INa (NaV1.5), ICaL (CaV1.2), IKs (KCNQ1/minK) and Ito (Kv4.3/KChIP2.2), and the Molecular Devices FLIPR® Tetra fluorescence screen for ICaL (CaV1.2), for control compounds used at AstraZeneca and GlaxoSmithKline. We examined how screening variability propagates through in-silico action potential models for whole cell electrical behaviour, and how confidence intervals on model predictions can be estimated with repeated simulations.ResultsThere are significant levels of variability associated with high-throughput ion channel electrophysiology screens. This variability is of a similar magnitude for different cardiac ion currents and different compounds. Uncertainty in the Hill coefficients of reported concentration-effect curves is particularly high. Depending on a compound’s ion channel blocking profile, the uncertainty introduced into whole-cell predictions can become significant.DiscussionOur technique allows confidence intervals to be placed on computational model predictions that are based on high-throughput ion channel screens. This allows us to suggest when repeated screens should be performed to reduce uncertainty in a compound’s action to acceptable levels, to allow a meaningful interpretation of the data.

The use of computational models to predict drug-induced changes in the action potential (AP) is a promising approach to reduce drug safety attrition but requires a better representation of more complex drug-target interactions to improve the quantitative prediction. The blockade of the human ether-a-go-go-related gene (HERG) channel is a major concern for QT prolongation and Torsade de Pointes risk. We aim to develop quantitative in-silico AP predictions based on a new electrophysiological protocol (suitable for high-throughput HERG screening) and mathematical modeling of ionic currents. Electrophysiological recordings using the IonWorks device were made from HERG channels stably expressed in Chinese hamster ovary cells. A new protocol that delineates inhibition over time was applied to assess dofetilide, cisapride, and almokalant effects. Dynamic effects displayed distinct profiles for these drugs compared with concentration-effects curves. Binding kinetics to specific states were identified using a new HERG Markov model. The model was then modified to represent the canine rapid delayed rectifier K(+) current at 37°C and carry out AP predictions. Predictions were compared with a simpler model based on conductance reduction and were found to be much closer to experimental data. Improved sensitivity to concentration and pacing frequency variables was obtained when including binding kinetics. Our new electrophysiological protocol is suitable for high-throughput screening and is able to distinguish drug-binding kinetics. The association of this protocol with our modeling approach indicates that quantitative predictions of AP modulation can be obtained, which is a significant improvement compared with traditional conductance reduction methods.