In vitro to in vivo extrapolation from 3D hiPSC-derived cardiac microtissues and physiologically based pharmacokinetic modeling to inform next-generation arrhythmia risk assessment

Abstract

Proarrhythmic cardiotoxicity remains a substantial barrier to drug development as well as a major global health challenge. In vitro human pluripotent stem cell-based new approach methodologies have been increasingly proposed and employed as alternatives to existing in vitro and in vivo models that do not accurately recapitulate human cardiac electrophysiology or cardiotoxicity risk. In this study, we expanded the capacity of our previously established 3D human cardiac microtissue model to perform quantitative risk assessment by combining it with a physiologically based pharmacokinetic model, allowing a direct comparison of potentially harmful concentrations predicted in vitro to in vivo therapeutic levels. This approach enabled the measurement of concentration responses and margins of exposure for 2 physiologically relevant metrics of proarrhythmic risk (i.e. action potential duration and triangulation assessed by optical mapping) across concentrations spanning 3 orders of magnitude. The combination of both metrics enabled accurate proarrhythmic risk assessment of 4 compounds with a range of known proarrhythmic risk profiles (i.e. quinidine, cisapride, ranolazine, and verapamil) and demonstrated close agreement with their known clinical effects. Action potential triangulation was found to be a more sensitive metric for predicting proarrhythmic risk associated with the primary mechanism of concern for pharmaceutical-induced fatal ventricular arrhythmias, delayed cardiac repolarization due to inhibition of the rapid delayed rectifier potassium channel, or hERG channel. This study advances human-induced pluripotent stem cell-based 3D cardiac tissue models as new approach methodologies that enable in vitro proarrhythmic risk assessment with high precision of quantitative metrics for understanding clinically relevant cardiotoxicity.

Keywords: cardiac arrhythmias; cardiovascular toxicity; human risk assessment; in vitro to in vivo extrapolation; physiologically based pharmacokinetics; tissue engineering.

Fig. 1.

Cardiomyocyte differentiation, microtissue generation, and metrics of proarrhythmic risk. a) Human-induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) are differentiated using a small-molecule modulation of the Wnt signaling pathway followed by lactate-based metabolic selection and microtissue formation. b, c) hiPSC-CMs are combined with 5% primary human cardiac fibroblasts in agarose molds with 35-microwells to allow for self-assembly of cardiac microtissues over 1 wk of culture prior to (d) optical mapping. e) Action potentials (APs) are recorded via optical mapping of voltage-sensitive dye signals before extraction of key AP metrics (Kofron et al. 2021; Soepriatna et al. 2023).

Fig. 2.

Point of departure (POD) estimation. a) Normalized concentration-response data is used to estimate POD across molds and differentiation batches (given example is cisapride APD_tri). Vehicle controls are treated as having a concentration one-half lower than the lowest tested concentration on a log scale (left-most point, black). b) Simulated data sets are generated by randomly selecting microtissues with replacements to generate bootstrapped samples of an equivalent size. c) The data are weighted by mold, averaged, and a cubic spline is fit to the resampled data prior to (d) fitting a sigmoidal curve. This sequential curve fitting ensures the fit sigmoid does not approach an asymptote without a saturated concentration response. The POD is estimated as the concentration where the fit sigmoid is 5% greater than the baseline response (dashed orange line, equivalent to a ∼10 to 20 ms change in APD_mxr). e) This process is repeated 10,000 times to determine the confidence interval of the POD represented by the experimental data.

Fig. 3.

In vitro to in vivo extrapolation. Physiological parameters as well as parameters describing the absorption, distribution, metabolism, and elimination of the chemical through the system were used to develop a physiologically based pharmacokinetic model that can be used to predict the population distribution of tissue concentration from any given daily dose. Reverse dosimetry predicts administered doses equivalent to an in vitro active concentration (i.e. point of departure), which can be compared with in vivo doses. Figure adapted from Bell et al. (2018).

Fig. 4.

Action potential (AP) concentration responses of engineered human cardiac microtissues. a–d) Representative AP traces of microtissues exposed to (a) quinidine, (b) cisapride, (c) ranolazine, and (d) verapamil for 15 min at indicated concentrations. Darker colors indicate higher concentrations. (e–h) Normalized concentration responses of AP duration to a maximum rate of repolarization (APD_mxr) and (i–l) action potential triangulation (APD_tri) from all microtissues. The left-most point (black) indicates the pre-treatment control with a normalized value of 1 and standard deviation of zero placed one-half lower than the lowest tested concentration on the log scale where it was used for point of departure (POD) calculation. The gray-shaded regions indicate therapeutic range of maximum heart concentration predicted by pharmacokinetic modeling. Data are presented as mean ± standard deviation for n = 51 to 95 microtissues across 3 molds and 2 to 3 differentiations per compound.

Fig. 5.

Point of departure (POD) estimation. PODs were estimated from resampled normalized concentration-response data. PODs were calculated from each of 10,000 data sets generated via bootstrapping and the 90% confidence intervals were calculated from the 5th and 95th quantiles. Data are displayed as a kernel density estimation. Light and dark colors represent POD estimates from APD_mxr and APD_tri concentration responses, respectively.

Fig. 6.

Proarrhythmic risk assessment via in vitro to in vivo extrapolation. Margins of exposure (MOEs) were calculated for each compound using both (left) APD_mxr and (right) APD_tri data. Points of departure (PODs) were adjusted for in vitro binding and converted to a human equivalent dose (HED) using a physiologically based pharmacokinetic model. Mean MOEs (points) were calculated using central estimates of POD and HED at the average therapeutic dose. Lower and upper estimates (error bars) were calculated using worst-case (low POD and HED estimates for the highest therapeutic dose) and best-case (high POD and HED estimates for the lowest therapeutic dose) scenarios, respectively. A Log₁₀MOE equal to zero occurs when the HED and therapeutic dose are equivalent, marking the transition from higher to lower risk (dashed vertical line).