Computational prediction of drug response in short QT syndrome type 1 based on measurements of compound effect in stem cell-derived cardiomyocytes

Abstract

Short QT (SQT) syndrome is a genetic cardiac disorder characterized by an abbreviated QT interval of the patient's electrocardiogram. The syndrome is associated with increased risk of arrhythmia and sudden cardiac death and can arise from a number of ion channel mutations. Cardiomyocytes derived from induced pluripotent stem cells generated from SQT patients (SQT hiPSC-CMs) provide promising platforms for testing pharmacological treatments directly in human cardiac cells exhibiting mutations specific for the syndrome. However, a difficulty is posed by the relative immaturity of hiPSC-CMs, with the possibility that drug effects observed in SQT hiPSC-CMs could be very different from the corresponding drug effect in vivo. In this paper, we apply a multistep computational procedure for translating measured drug effects from these cells to human QT response. This process first detects drug effects on individual ion channels based on measurements of SQT hiPSC-CMs and then uses these results to estimate the drug effects on ventricular action potentials and QT intervals of adult SQT patients. We find that the procedure is able to identify IC50 values in line with measured values for the four drugs quinidine, ivabradine, ajmaline and mexiletine. In addition, the predicted effect of quinidine on the adult QT interval is in good agreement with measured effects of quinidine for adult patients. Consequently, the computational procedure appears to be a useful tool for helping predicting adult drug responses from pure in vitro measurements of patient derived cell lines.

Fig 1. Illustration of the computational pipeline.

1) Biomarkers from the cardiac AP are taken from hiPSC-CMs under drug testing. 2) These biomarkers from dose escalation studies are inverted into an SQT1 model of the AP of hiPSC-CMs. Inversion into a matched model provides determination of drug effects on specific channels [23]. 3) The drug effects determined in 2 are inserted into a model of adult CMs with the same SQT1 mutation to give a prediction of drug effect in mature CMs [22]. 4) The adult CM model is converted into pseudo-ECG waveforms for prediction of QT segment changes in SQT1 patients under the estimated effect of the drug.

Fig 2. Illustration of the assumptions underlying the computational maturation approach.

1) The density of different types of ion channels (and other membrane or intracellular proteins) may differ between hiPSC-CMs and adult CMs, but the function of the individual channels is the same. In the model, the density difference is represented by the parameter λ. 2) The SQT1 mutation affects the individual I_Kr channels in exactly the same manner for hiPSC-CMs and adult CMs. In the model, the mutation is represented by an adjusted model for the open probability, o_Kr. 3) The effect of a drug on a single protein is the same for hiPSC-CMs and adult CMs. In the model, the drug effect is represented by the parameter ε.

Fig 3. Illustration of considered biomarkers.

Left: Illustration of the five biomarkers included in the cost function of the inversion procedure: The action potential durations (in ms) at 90% and 50% repolarization (APD90 and APD50, respectively), the maximal upstroke velocity (dvdt, in mV/ms), the action potential amplitude (APA, in mV) and the resting membrane potential (RMP, in mV). Right: Illustration of the QTp interval (in ms) computed from a pseudo-ECG waveform. The QTp interval is defined as the time from the onset of the QRS complex to the peak of the T-wave.

Fig 4. Illustration of the continuation algorithm used for optimization in the inversion method.

A) The problem is defined by a default model and some data we are trying to invert by finding an optimal model parameterization fitting the data. B) In the continuation algorithm, we seek temporary optimal parameters in M iterations (θ-steps). The objective for each θ-step is gradually changed from the default model to the data we are trying to invert. C) In each θ-step, we look for optimal parameters for the temporary objective by drawing N_G random guesses in the vicinity of the optimal parameters from the previous θ-step. For each random guess, we run N_NM Nelder-Mead iterations, and from the result, we select the best fit as the new optimal parameters. D) The final parameterization is given by the optimal parameters found in the last θ-step.

Fig 5. Representation of the SQT1 mutation N588K in the I_Kr model.

Left panel: Steady state I_Kr currents obtained at different fixed values of the membrane potential divided by the currents obtained at v = 0 mV. The results obtained for the fitted WT and SQT1 I_Kr models are compared to corresponding data from [14]. Right panel: Comparison of the steady state inactivation gate in the WT and SQT1 models of I_Kr and steady state inactivation data from [14]. In the SQT1 case, the inactivation is shifted towards higher values of the membrane potential. Data used in this figure can be found in S1 Data.

Fig 6. Properties of the base models for hiPSC-CMs and adult ventricular CMs in the WT and SQT1 cases.

Upper panel: Comparison of the action potentials computed for the WT and SQT1 versions of the hiPSC-CM (left) and adult (center) models, in addition to the pseudo-ECG for the adult model (right). The only difference between the formulations of the WT and SQT1 models is a shift in the inactivation gate of I_Kr as illustrated in the right panel of Fig 5. Lower panel: APD90 values and QTp intervals computed using the WT and SQT1 versions of the models. The computed APD90 values for hiPSC-CMs are compared to data from [9], and the computed QTp intervals for adults are compared to data from [15]. Data used in this figure can be found in S1 Data.

Fig 7. Comparison of measured and computed action potential biomarkers.

We consider the biomarkers reported in the data from [9, 11] (green) and computed from the fitted SQT1 hiPSC-CM models returned by the inversion procedure described in Section 2.3 (purple) for the drugs quinidine, ivabradine, ajmaline and mexiletine. Note that the definition of each of the biomarkers are illustrated in Fig 3 and that RMP data are only included for the quinidine case. Data are shown as the mean ± SEM (standard error of the mean). Data used in this figure can be found in S1 Data.

Fig 8. Result of the inversion procedure applied to data for the drugs quinidine, ivabradine, ajmaline and mexiletine from [9, 11].

Upper panel: Predicted drug effect on individual ion channels in the form of ε-values returned by the inversion procedure. Lower panel: Action potentials of the fitted SQT1 hiPSC-CM models in the control case and for doses of each drug. For comparison, we also show the action potential of the WT hiPSC-CM model. Action potential characteristics of the fitted models in the lower panel are compared to data from [9, 11] in Fig 7. Data used in this figure can be found in S1 Data.

Fig 9. Comparison of reported and estimated block percentages for I_Kr.

We consider the reported block percentages for 10 μM quinidine, 10 μM ivabradine, 30 μM ajmaline and 100 μM mexiletine from [11] and the block percentages estimated based on the IC₅₀ values identified in the inversion procedure. Data are shown as the mean ± SEM. Data used in this figure can be found in S1 Data.

Fig 10. Estimated drug effects for adult patients with the SQT1 mutation.

The figure displays the estimated drug effect of quinidine, ivabradine, ajmaline and mexiletine for adult patients with the SQT1 mutation (and a heart rate of 60 beats/min) based on the result of the inversion of hiPSC-CM data reported in Fig 8. Upper panel: Estimated drug effect on the adult ventricular action potential. Lower panel: Estimated drug effect on the QTp interval. See Fig 3 for an illustration of the definition of the QTp interval. Note that the blue lines represent the adult WT case. Data used in this figure can be found in S1 Data.

Fig 11. Comparison of measured and estimated adult QTp intervals.

We compare the QTp intervals (see Fig 3) estimated by the computational procedure and the measured QTp intervals from [15] for different heart rates. We consider the WT case, the SQT1 case and the SQT1 case with cells exposed to 6.5 μM quinidine. The drug effect estimated by the procedure is based on measured drug effects for SQT1 hiPSC-CMs from [9]. Data used in this figure can be found in S1 Data.