On-chip spatiotemporal electrophysiological analysis of human stem cell derived cardiomyocytes enables quantitative assessment of proarrhythmia in drug development

We examined a simultaneous combined spatiotemporal field potential duration (FPD) and cell-to-cell conduction time (CT) in lined-up shaped human embryonic stem cell-derived cardiomyocytes (hESC-CMs) using an on-chip multielectrode array (MEA) system to evaluate two origins of lethal arrhythmia, repolarization and depolarization. The repolarization index, FPD, was prolonged by E-4031 and astemizole, and shortened by verapamil, flecainide and terfenadine at 10 times higher than therapeutic plasma concentrations of each drug, but it did not change after lidocaine treatment up to 100 μM. CT was increased by astemizol, flecainide, terfenadine, and lidocaine at equivalent concentrations of Nav1.5 IC50, suggesting that CT may be an index of cardiac depolarization because the increase in CT (i.e., decrease in cell-to-cell conduction speed) was relevant to Nav1.5 inhibition. Fluctuations (short-term variability; STV) of FPD and CT, STVFPD and STVCT also discriminated between torsadogenic and non-torsadogenic compounds with significant increases in their fluctuation values, enabling precise prediction of arrhythmogenic risk as potential new indices.


Principle
The extracellular field potential(FP) is the electrical potential produced by the inward and outward ion currents of cardiomyocytes. In electrophysiological conventions, a negative current value or downward deflection of a current trace are typically referred to as an inward current, reflecting either the movement of positive ions into the cell or negative ions out of the cell. A positive current value or upward deflection of the current trace (i.e., outward current) also reflect either the movement of positive ions out of the cell or negative ions into the cell. Similar to the action potential(AP), the FP origin is the electrical potential difference across the plasma membrane of excitable cells caused by the inward/outward ion current of cells. Hence, for individual cells, the time course of FP, V f , is proportional to the transmembrane ion current of cells, I ion , which is the AP time differential, φ a , of cells. The relationship of V f to φ a is theoretically described as, where, R is the resistance of the environment and measurement system. Time course V f amplitude is the value of the FP microelectrode measurement. Because phenotypes and electrophysiological changes of cardiomyocytes can be distinguished by APs, in practice, FPs can be used for long term screening of the electrophysiological properties of cells on microelectrodes. The actual field potential waveform measured by MEA is affected by the electrical resistance and capacitance between the cell membrane and the surface of electrode 41 , however, as shown in Fig. 1, FP waveforms of phenotypes can be estimated by the time differential of APs, and the experimental results of FPs were similar to their time differentials of APs.

Materials and Methods
Test compounds. The following six compounds in four categories were chosen and applied for experiments: E-4031; hERG inhibitor, astemizole and terfenadine; multi-channel blockers, lidocaine and flecainide; sodium channel blocker, and verapamil; calcium channel blocker. The vehicle control was dimethyl sulfoxide (DMSO). E-4031 was synthesized in Daiichi Sankyo Co., Ltd (Tokyo, Japan). The other compounds were purchased from Sigma (St. Louis, MO). Each compound was dissolved in DMSO at a thousand-fold concentration to prepare stock solutions. The highest concentration of each compound was 30 100 times their effective therapeutic plasma concentration (C eff ) in consideration of the solubility.
MEA cardiomyocyte lined-up network chip. The MEA chip surface was prepared as follows. First, the chip surface was coated with collagen type I-C (Nitta Gelatin, Japan). After drying, the chip surface was coated with a 0.1% agarose gel. A part of thin agarose layer was carefully melted along three parallel lines of 8 × 8 multi-electrodes with spot heating of infra-red laser (1480 nm), which selectively removed agarose gel, but not collagen gel (Fig. 2). At >30 min before the start of experiment, the medium was changed to 2 mL pre-warmed fresh maintenance medium per chip.
Field potential recordings. Extracellular potential recordings of the lined-up hESC-CMs were performed using the MEA system at a sampling rate of 10 kHz with a low path filter of 2 kHz and high path filter of 1 Hz, and signals were amplified by 100-50,000 using the amplifier. The MEA chip with lined-up hESC-CMs was set in the FP measurement device and placed in a humidified cell culture incubator at 37 °C, 5% CO 2 . The signals were analyzed by Igor Pro 6.0 (WaveMetrics, Inc., USA) and FlexPro 7.0 (Weisang GmbH, Germany).
Drug administration protocol. We adopted the same drug administration protocol as we examined previously 40 (Fig. 2B). Briefly, the MEA chips of lined-up cells for experiments were selected from the chips having 0.6-1.1 Hz beating frequency and ventricle-like waveforms of FP recordings (see Fig. 1B,D). The MEA chip was placed in the holder of the FP measurement device, and equilibrated for 5 min, and then the control FP waveforms were recorded for 10 min. Subsequently, the drug was applied to the medium at 1% dilution in serially increasing additions, and the FP waveforms were recorded for 10 min at each concentration. Finally, the drug-containing medium was replaced with fresh medium after washing three times. The last 50 beats extracted from 10 min recorded FP waveform data was used for beat rate, FPD (the time from first depolarization peak to broad repolarization peak) and CT (the time difference of the first peaks between separately placed electrodes in lined-upped cell-network) measurement at each concentration (Fig. 2C). Data analysis. The FPD was defined as the duration time between the initial field potential deflection and the peak of inward current of depolarization mainly caused by potassium ion channels. It was normalized (cFPD) to the beating rate of cardiomyocytes using Bazett's correction formula based on the comparison of the slope of a plot of inter-spike interval and FPD, or corrected FPD by Bazett's or Fridericia's formulae (Fig. 3). The short-term variability (STV) of cFPD and CT, which was defined as the mean distance of points perpendicular to the line of identity in the Poincaré plot, was calculated ( , where D n represents the FPD and CT of n-th beating). STVs were last 50 beat data from 5 min extraction of cFPD and CT. To assess the arrhythmogenicity risk of test drugs, the mean values of cFPD, STV cFPD , CT and STV CT of each drug were plotted. Vehicle, verapamil and lidocaine were recognized as safe drugs, and E-4031, astemizole, terfenadine and flecainide were recognized as the arrhythmogenic drugs. In CT-STV CT and STV cFPD -STV CT plots, the line segregates the safe and arrhythmogenic drugs. The slope and intercept of the line were determined as the maximum distances from the line to the nearest point in each safe and arrhythmogenic drug data by Microsoft © Office Excel 2013 solver.

Statistical analysis.
All values are presented as mean ± S.E.M. (unless stated otherwise). All drug effects were evaluated using the Dunnett's multiple comparison test when comparing multiple groups with time-matched vehicle control. P < 0.05 was considered as statistically significant.

Results
The effect on cFPD and STV cFPD . To evaluate the potential to detect the effects on FPD of the lined-up cardiomyocytes network, ca. 1000 hESC-CMs were seeded on the rectangular agarose-free region of the MEA chip and incubated until the cells attached to the microelectrodes efficiently enough to acquire their field potential signals (Fig. 2C). As shown in the field potential graphs, waveforms of hESC-CMs in this experiment were similar to the ventricle type (Fig. 1B).
We first tested QT-prolonging drug E-4031 with hERG channel blocking as positive control. E-4301 prolonged and was 30% in corrected FPD with Bazett's formula (cFPD)(n = 13, P < 0.05, compared to DMSO by Dunnett's Multiple Comparison Tests) at 0.1 μM and 47% (n = 12, P < 0.05) at 1 μM (blue symbols in Fig. 4A,B). Verapamil, calcium channel blocker with an almost equivalent hERG blocking action, shortened cFPD by 26.6% at 10 μM (n = 3, P < 0.05, Fig. reff3C). STV of cFPD (STV cFPD ), which is the fluctuation of the repolarization of cells on the electrode, was more than doubled by 1 μM of E-4031, whereas no significant change in STV cFPD was observed at concentrations up to 10 μM of verapamil ( Fig. 4B,C, red symbols). Responses to cFPD in this lined-up hESC-CMs and cluster were similar 40 .
The effect on CT and STV CT . First, we evaluated the CT between two neighboring electrodes with 150 μm of inter-electrode distance and its STV (STV CT ) as the index of spatiotemporal fluctuation. As expected, CT did not significantly change at any concentration by pure hERG blocker E-4031 or Ca 2 + and hERG blocker, verapamil (Fig. 5B,C), which was similar to the results of DMSO as the vehicle control (Fig. 5A). However, STV CT was increased by the 336% in comparison with the control condition (n = 13, P < 0.05) at 1 μM of E-4031 (red symbols in Fig. 5B) and by 103% at 1 μM of verapamil (red symbols in Fig. 5C).
Astemizole increased CT by 126% at 1 μM (n = 3, P < 0.05, Fig. 5F) and terfenadine also increased CT by 189% at 3 μM (n = 4, P < 0.05, Fig. 5G). IC 50 of Nav1.5 by astemizole and terfenadine was 2.2 and 4.4 μM, respectively (Table 1) Risk assessment of arrhythmogenecity. In Table 1, all drugs used in the MEA assay employing lined-up hESC-CMs are summarized with arrhythmogenic markers (cFPD and STV cFPD ) and other parameters (CT, STV CT , Inter-spike interval, first peak amplitude, and arrest case). A cFPD and STV cFPD plot at the highest dose of each drug did not discriminate non-arrhythmic lidocaine from pro-arrhythmic drugs (E-4031, astemizole, terfenadine, and flecainide) (Fig. 6A). In contrast, a CT and STV CT or a plot of STV CT and STV cFPD could clearly discriminate between arrhythmogenic drugs (E-4031, flecainide, astemizole and terfenadine) and safe drugs (lidocaine and verapamil) clearly (red dashed line in Fig. 6B,C).

Discussion
The electrophysiological analysis of hPSC-CMs is currently the most ideal strategy for in vitro QT/arrhythmogenic risk assessment. In particular, FPD assessment in the MEA assay using hPSC-CMs shows the potential to predict the QT/arrhythmogenic risk with much better reliability compared with other conventional in vitro assays [6][7][8]10,49,50 , although only FPD prolongation cannot fully predict lethal arrhythmia especially for evaluation of drugs with multiple ion channel effects 26 . Either spatial heterogeneity or temporal fluctuation would lead VT and lethal arrhythmia [51][52][53] . Regarding the temporal fluctuation, not only the conventional STV FPD or STV cFPD analysis but also the analysis including the individual or experimental differences Pueyo2016 or other cardiovascular related parameters, such as left ventricular pressure in in vivo 54 have been developed for the TdP risk related parameters. However, there is no report that simultaneously measures the spatial fluctuation and temporal fluctuation in hPSC-CMs, and little is known about the combination potential of those indexes on arrhythmogenic risk assessment. The MEA assay has the advantage of observing multipoints in the cell community. Nevertheless, there are poor insights into the spatial fluctuation of conduction in the MEA assay. In the present study, we showed that not only the temporal fluctuation of cFPD, but also the spatial fluctuation of conduction time (CT) of the lined-up hPSC-CMs in the MEA assay can be additional indices as proarrhythmic markers. Our results indicated that the lined-up hESC-CMs can be used to evaluate cell-to cell conduction in combination with QT prolongation as well as discriminate between arrhythmogenic and safe drugs with multiple ion channel effects (Fig. 6A-C) by spatiotemporal fluctuation parameters, STV cFPD and STV CT , or CT and STV CT .
In this study, our lined-up hESC-CMs provided reliable FPD response similar to conventional two-dimensional (2D) cell sheets exposed to well-known hERG channel blocker E-4031 as an FPD prolonger. FPD prolongation by E-4031 was demonstrated at the relevant concentration to the IC 50 of hERG. Calcium channel blocker verapamil as an FPD-shortening drug also showed reasonable responses similar to the previously reported cFPD-shortening concentration response in 2D cell sheet or clusters of hPSC-CMs at 1-10 μM 8,40 . However, the effects of astemizole on cFPD were seen at relatively higher concentrations compared with previous studies, especially for astemizole 6 , and at least about a 160 times higher concentration to the IC 50 of hERG.
Regarding to the discrepancy of the drug response in cFPD among our results and previous reports using the MEA assay with hPSC-CMs, the difference of cell origins (hESC-CMs vs. hiPSC-CMs), cell arrangements (lined-up, vs. 2D cell sheet), cell density (274 ± 59 cells/mm with 60 μm width in the lined-up system), and the duration of drug exposure, might be related, although there is no obvious explanation for this difference.
Furthermore, the present assay conditions did not detect cFPD prolongation of terfenadine or flecainide, which is in contrast to previous studies 5, 55 . Terfenadine showed FPD shortening at concentrations more than 10 μM (180 times higher than its hERG IC 50 ) in this study. In addition, flecainide did not show FPD prolongation at concentrations of up to 100 μM (55 times higher than its hERG IC 50 ). It has been shown that drugs with a CaV1.2 inhibition potential have a counter effect on FPD prolongation by their hERG inhibition. IC 50 of Cav1.2 can explain our observation because it is approximately four times higher than the IC 50 s of hERG for flecainide and terfenadine. In addition, terfenadine shows a bell-shape change in cFPD prolongation 5,55 , which cause difficult in evaluating its effect on cFPD. In fact, some reports indicate that terfenadine prolongs cFPD 4,7,8,49 whereas others reported that terfenadine had little effect on cFPD 50 . Terfenadine is also reported to be one of the difficult drugs to detect the arrhythmogenicity of within a short period of exposure [56][57][58][59] , and this can be detected only after longer exposure of more than 12 h 60 . The time-course analysis would be required to detect arrhythmogenicity more precisely.
To investigate the CT and variability of CT (STV CT ), we focused on two types of well-known sodium channel blocker, flecainide (class Ic) with a slowdown effect on the conduction velocity and lidocaine (class Ib) with little such effect. From the results, flecainide tended to increase CT and STV CT at the second highest concentration, which is approximately equivalent to the therapeutic concentration, whereas lidocaine had a lesser effect than flecainide even at the highest concentration that is approximately 10 times higher than the therapeutic concentration (Fig. 5D,E). In addition, multi-ion channel blockers astemizole and terfenadine showed marked increases in Figure 4. Effects of the arrhythmic drugs to the repolarization function and its temporal fluctuation on the line-shaped model using hESC-CMs. Dose-dependent changes of the field potential duration corrected by beating frequency using Bazett's formula (cFPD, blue filled circles and lines) and its short-term variability (STV cFPD , red open triangles and lines) are plotted on DMSO as a vehicle control (A), E-4031 as a I Kr blocker (B), verapamil as a Cav1.2 blocker (C), lidocine and flecainide as I Na blockers (D and E), astemizole and terfenadine as multi-ion channel blockers (F and G). cFPD and STV cFPD were normalized to control condition. Symbols and error bars indicate mean values and standerd error (SE). 10% cFPD prolongation (cFPD > 1.1) and 90% STV cFPD increase (STV cFPD > 1.9) are shown in blue or red dashed lines, respectively. The green, orange, and purple bars show hERG, Cav1.2, and Nav1.5 IC 50 of each drug, respectively. CT and STV CT at their equivalent concentrations to IC 50 s of Nav1.5 (Fig. 5F,G and Table 1). All drugs with Nav1.5 inhibition showed increases in CT at almost the same concentration as their IC 50s of Nav1.5, suggesting that CT, a parameter of cell-to cell conduction, would be relevant to Nav1.5 inhibition. This finding is well concordant with the recent report using the 2D-sheet of iPS-CMs 61 .
Moreover, Izumi-Nakaseko and colleagues stated that conduction properties of the cell sheet may largely depend on the extent of Nav1.5 availability as is the case in the human ventricle. Interestingly, a plot of CT and STV CT discriminated between torsadogenic and non-torsadogenic compounds. Nevertheless, a plot of FPD and STV FPD resulted in a false-positive judgment as arrhythmogenic for lidocaine in the lined-up cell network model (Fig. 6).
E-4031 and verapamil did not increase the mean value of CT at any tested concentration. However, both drugs increased the temporal beat-to-beat fluctuation, that is, STV CT at the highest concentration of E-4031 or from the lowest concentration of verapamil (Table 1). It appears that the increase in STV CT has no relationship to Nav1.5 inhibition, because IC 50 s of Nav1.5 in both E-4031 and verapamil are higher than their highest tested concentrations in this study.
Quality and quantity control of cardiomyocyte network should also be considered for establishment of reproducible on-chip in vitro assays. The influence of fibroblasts has been examined in on-chip cell-network assay, indicating that fibroblasts reduce the ability of synchronization of cardiomyocytes 62,63 . Heterogeneity of cardiomyocytes also should be a issue for improvement. We have previously evaluated the single cells obtained from single hESC-CMCs visually and concluded that 73% of total cells beat spontaneously, however their action potential analysis of APD 20-40 /APD  and dV/dt max revealed that they would have at least more than two major phenotypes, arbitrary "ventricular-like" and "atrial-like" cells 64 . To overcome this problem of mixture of phenotypes, computational simulation might be one of the possible solutions. For example, the role of community effect of single phenotype cardiomyocyte network was simulated successfully with fluctuation-dissipation theorem 65 .
As a limitation of this study, the following four aspects should be noted. We used only one positive (flecainide) and one negative (lidocaine) control for conduction-slowing effect. Further study using diverse blockers, such as other types of sodium channel blockers and gap junction inhibitors, would deepen the understanding of each  Table 1. Effects of ion channel blockers on cFPD, STV cFPD , conduction time (CT), STV CT , inter-spike interval (ISI) and beat rate.
parameter recorded in our MEA assays. Second, because we cannot deny the possibility of drug-induced chronotropic effect on CT evaluation, pacing would be helpful to eliminate chronotropic effect, which would enable us to evaluate the drug effects on CT directly. Third, a comparison of the expression balance of each cardiac ion channel in each cell arrangement model (lined-up, clusters and 2D cell sheets) is also important, because there are differences in response in cFPD prolongation among those assays 64 . Finally, we need to consider the importance of community effects of cardiomyocytes, such as the cell number, spatial arrangement, and cell contamination by non-cardiomyocytes.
In conclusion, the lined-up cell-network model of the MEA assay using hESC-CMs allows small scale in vitro simultaneous evaluation of FPD and CT, that is, two important indexes of lethal arrhythmia occurrence, repolarization and depolarization. STV FPD and STV CT also discriminated between torsadogenic and non-torsadogenic compounds with significant increases in both of their fluctuation values, indicating they are potential new indexes to predict torsadogenic lethal arrhythmia risk more precisely.