A commercially available hiPSC-CM cell line was used (iCell Cardiomyocytes²: Catalog number R1017, Kit/lot 12,012; Fujifilm Cellular Dynamics Inc., Madison, WI, United States) for the study. Cells were pre-plated and seeded in fibronectin-coated 96-well plates at a density suited to form a monolayer and maintained in culture in a stage incubator (37°C, 5% CO₂) according to the instructions of the cell provider. The experiments with test drugs were carried out 5–7 days after plating cells on the plate to have living, beating monolayer cardiomyocytes. The beating monolayer was taken from frozen vials iCell Cardiomyocytes² (≈5 million cells/vial), which were plated onto three 96-well plates (≈50K/well).

To investigate whether there could be lot-to-lot variability with hiPSC-CMs, we tested 55 drugs, known cardiac risks, in one lot of iCell Cardiomyocytes² (Lot 12,012), 19 common reference drugs in another lot of iCell Cardiomyocytes² (Lot105091) as well as in iCell Cardiomyocytes (Lot 103,674) (non-square cells). The experiments with test drugs were carried out 5–7 days for iCell Cardiomyocytes² and 12–15 days (for iCell Cardiomyocytes) after plating cells to have living, beating monolayer cardiomyocytes (≈5 million cells/vial/96-well plates) (≈50K/well). Cells from these lot numbers were from the same donor (Apparently healthy normal, female, Caucasian, age <18) (according to FUJIFILM Cellular Dynamics).

The method for the Ca²⁺ transient assay used in the present study has been described previously (Zeng et al., 2016; Kopljar et al., 2018a; Kopljar et al., 2018b).

On the day of the experiment, the culture medium in the 96-well plates containing the monolayers of hiPSC-CMs was replaced with Tyrode’s solution (Sigma, No. T2397) supplemented with 10 mM HEPES together with KCl to yield isokalemic (4.2 mM K⁺) conditions.

The Ca²⁺-sensitive fluorescence dye Cal-520™ AM (Cat. no.21131; AAT Bioquest) was used to capture the intracellular Ca²⁺ transients in hiPSC-CMs. The protocol used was as described in Kopljar et al. (2018a). Briefly, Cal-520 was incubated for 70 min followed by a wash-out with supplemented Tyrode’s solution and a 30-min recovery before starting the experiments at 37°C. On the day of the experiment, these solutions were further diluted with the supplemented Tyrode’s solution. Compounds were added automatically using the Functional Drug Screen System (FDSS/µCell; Hamamatsu, Japan) head stage by adding 100 µl of the 2-fold concentrated compound solution to wells with hiPSC-CMs already containing a volume of 100 µl of the experimental solution, reaching the final intended test concentration in 0.1% DMSO.

The spontaneous beating activity of hiPSC-CMs was assessed through measurement of the Ca²⁺ fluorescence signal integrated over the whole well. Ten minutes before starting the recordings, the experimental plates were put into the FDSS system for stabilization and a baseline recording was captured for 4 min followed by compound addition. The effect of the compounds was recorded around 30 min after compound addition. The Ca²⁺ transient signals were sampled every 16 ms (frame rate of 62.5 Hz) in hiPSC-CMs maintained at 37°C during the acquisition time. Calcium transient duration at 90% of repolarization (CTD₉₀) is considered to be a surrogate for action potential duration at 90% repolarization or QT-interval of the electrocardiogram (Spencer et al., 2014). In addition, beat rate (BR = peak count/min) and amplitude (amp = difference between max. and min of every beat) were measured, and incidence of cellular arrhythmias, e.g., EAD-like activity (considered a surrogate of TdP), cessation of beating (∼ stop beating), and below beating below quantification level (caused by low amplitude of the Ca²⁺ transient or by ectopic beat: BBQL) were also recorded (Figures 1B,C).

The compounds and their respective test concentrations were selected based on the CiPA compound list (Blinova et al., 2018 #240) and from our earlier publication (Kopljar et al., 2018b). Four DMSO stocks for each drug concentration were prepared and either used on the same day or aliquoted and frozen. Concentrated (2-fold) testing solutions (50-fold for sequential dosing) for each concentration were prepared freshly on the day of testing by diluting DMSO stocks into the experimental medium. Two-fold dilution was done when drugs were added to the experimental well to achieve targeted concentration. For sequential dosing, DMSO concentrations were adjusted sequentially up to 0.1% at the highest concentration to achieve the targeted concentration of each drug. Each compound concentration was tested in 5-6 replicates.

Dofetilide (3 nM), isoprenaline (100 nM), and nimodipine (300 nM) were used as the positive controls, while cetirizine (300 nM) was used as the negative control. A total of 55 reference compounds were used in this study with different known clinical cardiac risks.

For all individual experiments, delta percent (Δ%) at 30 min with respect to the baseline value was calculated [e.g., ∆% for CTD₉₀ = (CTD₉₀ 30 min-CTD₉₀ 0 min)/CTD₉₀ 0 min) × 100]. The following two statistical approaches were used to make decisions about a compound’s effect.

First, the tolerance interval (TI)-based categorization, where the Δ% values of all (pooled) DMSO wells (experiments in this study) were centered around zero (corrected with the mean) and the parametric 90–95% (p = 0.90 & 1-α = 0.95) tolerance interval (TI) was calculated (Liao et al., 2005). TIs indicate an interval where, with a certain confidence level (1- α = 0.95), a specified proportion (p = 0.90) of a sampled population falls. The lower and upper limits based on the TI values were used as cut-offs for the vehicle-corrected net effect (ΔΔ%) values of the compounds to categorize them. These ΔΔ% net effects were the values obtained after normalization for baseline (Δ% vs. baseline) per well and the aggregated compound-treated wells are vehicle corrected by subtraction with the aggregated Δ% of the corresponding DMSO wells from the same plate [e.g., ΔΔ% = median Δ% drug - median Δ% DMSO]. The categorization was made first for each concentration and then aggregated per compound and test site. If a net effect (ΔΔ%) at a given concentration of a compound was below the lower limit of the TI, then it was categorized as “decreased”; if the net effect was within the DMSO TI then it was categorized as “no effect,” and if the net effect was above the upper limit of the TI then it was categorized as “increased.” At the compound level, a compound was categorized as “increased” if it was in a category “increased” in at least one of the four concentrations. Similarly, a compound was categorized as decreased if it was in a category “decreased” in at least one of the four concentrations. Based on TI values, we determined the cut-off values for the hazard score system (Kopljar et al., 2018b).

Data are expressed as means ± SEM. All statistical analyses were performed using SAS^® software (Copyright^© 2002–2012 SAS Institute Inc.).

Significant prolongation of CTD₉₀ observed in this assay was considered a surrogate for drug-induced QT prolongation and TdP risk. Since some drugs cause QT prolongation in humans at overdoses and/or in combination with other drugs, the predictivity values were calculated at free maximal plasma concentration in human (fC_max) and also at a threshold of 10 X fC_max or 30X fC_max. The list of reference drugs that have QT-prolongation and potential risk for TdP in humans was taken from the ParmaPendium^® (www.pharmapendium.com) and from our recent publication (Kopljar et al., 2018b). A true positive (TP) was defined as QT prolonging drugs that significantly prolonged CTD₉₀, while true negative (TN) was defined as non-QT prolonging drugs in humans that did not prolong CTD₉₀ at a tested concentration. A false negative (FN) was defined as a known QT-prolonging drug that did not prolong CTD₉₀, while a false positive (FP) was defined as a non-prolonging QT drug that significantly prolonged CTD₉₀ at > 30-fold fC_max. Sensitivity was calculated as the % of QT-prolonging drugs correctly predicted based on CTD₉₀ prolongation in this assay [TP/(TP + FN)]. Specificity was calculated as the % of non CTD₉₀ prolonging drugs correctly predicted as non-QT prolonging drugs [TN/[(TN + FP)]. Positive predictive value (PPV) was calculated as: PPV = [TP/(TP + FP)], while negative predictive value (NPV) was NPV = [TN/(TN + FN)].

The modeling used in this study was primarily based on the seven predictors from the measurement parameters in MEA as given in the CiPA publication (Blinova et al., 2018), with a slight modification. Data on effects of 55 drugs with known clinical risk of TdP obtained from all experimental sites were used to construct a model that would predict TdP risk category of a drug based on its effects on hiPSC-CMs. Seven modified endpoints from hiPSC-CM experiments used as potential model predictors (Blinova et al., 2018) were: Predictor 1, drug-induced cessation of beating and early afterdepolarization (EAD) at any concentration; Predictor 2, describes the ability of a drug to induce arrhythmia-like events (EAD) in over 40% of wells in hiPSC-CMs; Predictor 3, CTD₉₀ (ms) at the first drug concentration with statistically significant (p ≤ 0.05) prolongation or shortening; Predictor 4, reflects the amount of drug-induced repolarization prolongation (CTD₉₀% of baseline) at the lowest concentration where statistically significant change from the baseline was observed or maximum prolongation at any of the studied concentrations; Predictors 5 and 6, account for concentrations of a drug relative to its fC_max when prolongation of CTD₉₀ or incidence of arrhythmia-like event was first observed; and Predictor 7, an estimated amount of CTD₉₀ prolongation that a drug would induce at its clinical fC_max. We added predictor 8 that includes drug-induced cessation of beating in ≥40% of wells at any concentration, which added an extra value to detect drug-induced slow conduction time or inhibiting spontaneous beating.

Logistic regression models (Model 1 and Model 2) were then used (high or intermediate risk vs. low risk–Model 1, and high vs. low and intermediate-risk vs. low risk–Model 2) on all eight risk predictors according to the CiPA publication (Blinova et al., 2018).

To investigate lot-to-lot variability in hiPSC-CMs, we tested 19 drugs with known clinical cardiac risk in other lots of iCell Cardiomyocytes². Fifty-five reference drugs were tested in iCell Cardiomyocytes² -Lot12012, 18 reference drugs were tested in iCell Cardiomyocytes² (Lot105091), and 18 reference drugs were tested in iCell Cardiomyocytes (Lot103674) (nonsquare cells).

To investigate the reproducibility of the cardiac hazard risk predictions from the different cell lines, we compared the present data with the results of 55 reference drugs obtained from hiPSC-CMs (iCell Cardiomyocytes²) to the data obtained from another cell line which was published previously (Kopljar et al., 2018b).

We also evaluated the translational predictability of the risk scores of NCEs in the CTCM assay to our complementary cardiac safety models: hERG assay, isolated rabbit wedge preparation (ex vivo) (Lu et al., 2016), and anesthetized guinea-pig (in vivo) models (Kopljar et al., 2018b) using the Software - SPEcII (Unlimit-IT, Belgium). All the experiments involving the use of animals have been conducted in accordance with “The provision of the European Convention” on the protection of vertebrate animals that are used for experimental and other scientific purposes, https://rm.coe.int/168007a67b.

We applied our previously reported strategy and methodology for the development of a cardiac hazard scoring system based on phenotypic readouts from Ca²⁺ transient imaging to a different hiPSC-CM cell line (iCell Cardiomyocytes²) (Figure 1A). For a qualitative evaluation of hiPSC-CMs, each test plate (i.e., 96 unique experiments) contained 0.15 DMSO controls, together with several control drugs (cetirizine, isoprenaline, nimodipine, and dofetilide). These drugs have different pharmacological classes and were served as a pharmacological reference set for the development of the hazard scoring system. Cetirizine is a real negative control known not to have any clinical cardiac liabilities. Nimodipine was used as a calcium channel antagonist and isoprenaline is a beta-adrenergic agonist. Dofetilide is a hERG blocker associated with QT prolongation and proarrhythmic TdP risk in humans. Pharmacological effects in hiPSC-CMs were investigated after a 30-min incubation period and normalized against the respective baseline recording, yielding a Δ% change in CTD₉₀, BR, and Amp.

The current hazard scoring outcome was largely similar to our previously reported scoring system for the Cor.4u cell line (Kopljar, 2018 #245, based on the TIs and its modified cut-off values of key parameters: CTD₉₀, beat rate, amplitude, and incidence of arrhythmic events (EAD and cessation of beating). The scoring matrix was determined based on the cut-offs between the different effect zones for each parameter (Figure 2). The “no effect” zone shows minimal changes in a parameter that is within vehicle variability, whereas “mild” and “strong” zones are determined from mild to large changes bidirectionally (e.g., CTD₉₀ shortening and prolongation). The cut-off values are net changes (ΔΔ% changes vs. baseline and vehicle), which are based on the statistical tolerance intervals (TIs) (Δ% changes vs. baseline). TIs indicate an interval where, with a certain confidence level, a specified proportion of a sampled population falls. Vehicle treatments showed low variability in Δ% change for CTD₉₀, BR, and Amp with respect to baseline (Figure 2). Cardiac hazard scoring for the tested compound at a given concentration was given into the following risk classification hazard score (color labels): “No” (green), “Low” (yellow), “High” (red), and “Very High” (black). No hazard labeling means that a tested compound’s effects were within the vehicle variability. Low hazard score indicates that the effects were outside the vehicle variability, with = low or limited risk. High hazard score gives a strong concern whereas the very high hazard score suggests potentials to cause cardiac arrhythmias-EADs.

Based on the outcomes of the positive/negative controls and the reference drug list used in another cell line (Cor.4U) (Kopljar, 2018 #245), we validated the results in this new hiPSC-CM cell line (iCell Cardiomyocytes²) which was essential for using the scoring system for hazard identification of NCEs. Figure 2F shows the hazard distribution for control drugs (at a single concentration) tested in numerous studies. Cetirizine at 1 µM (16-fold the fC_max) as a negative control was exclusively identified to be a no hazard agent. Nimodipine at 300 nM (15-fold fC_max) (n = 124) and isoproterenol (n = 143) at 100 nM (63-fold fC_max) are both cardio-active agents that were identified within the low and high hazard zones (95 and 92%, respectively). Dofetilide at 3 nM (2-fold f C_max) (n = 131) was correctly identified as yellow (8%), high (red color) (21%), or very high (black color) (71%) hazard score.

Following up, we validated the hazard scoring (as defined in Figure 3) using 55 reference drugs and their respective known clinical risk relative to their fC_max. These reference drugs are categorized based on their pharmacological mechanism or level of cardiac side effects in men (Figure 4). Indeed, negative control drugs have no reports of cardiac liabilities in humans and are expected to be proven to be “no hazard” for concentrations up to 10- to 30-fold fC_max. All negative controls were found to have “no hazard” (green). The only special case was raloxifene at clinically irrelevant concentrations (3–10 μM; ≥2500-fold the clinical fC_max).

We investigated the potential hazard scoring system for drugs, known to be associated with QT prolongation and a certain degree of TdP risk in humans (Figure 3) (These drugs were categorized into two groups according to the FDA label or described in www.torsades.org: conditional TdP/potential TdP risk, and known TdP incidence. In cases of overdose, drug–drug interactions, or in certain high-risk individuals (Woosley et al., 2018). The rank order of incidence of EADs or degree of prolongation of CTD₉₀ could be found to be co-related well to known TdP drugs > potential TdP risk or “Conditional TdP risk.” Only the antidepressant olanzapine was not found to be with any hazard level, most likely because of the fact that the highest tested concentration was 100 nM, which is around fC_max. This concentration might be too low, since olanzapine was also only found to significantly inhibit hERG at a concentration of 6 µM, which is at 90-fold fC_max (Titier et al., 2005).

Although the main focus within preclinical safety pharmacology in most pharmaceutical companies is on drug-induced QT prolongation and TdP, other pharmacological mechanisms affecting different cardiac ionic currents can also result in cardiac toxicity such as, e.g., QRS widening or QT shortening. Na⁺ channel blockers such as primidone and phenytoin, which are considered relatively safe in humans, had little impact on hiPSC-CMs (Green score) and another Na⁺ channel blocker, mexiletine, caused some cessation of beating only at a very high concentration (100 µM = 67x fC_max), resulting in a Red score in the current study, similar to that in the CIPA validation study (Blinova et al., 2018). However, most Na⁺ blockers also inhibit hERG current at similar concentrations, which are readily detected as CTD₉₀ prolongation and incidence of EADs leading to “low” to “very high” risk (Figure 3).

Ca²⁺ channel antagonists resulted in strong responses in hiPSC-CMs, showing a marked decrease in amplitude companying with CTD₉₀ shortening and pronounced beat rate increase. The cardiac I_KATP channel opener and the hERG activator (ICA-105574), which shortened CTD₉₀, were also characterized in the present study.

Additional pharmacological mechanisms including beta-adrenergic agonists (isoproterenol, dobutamine, and salbutamol), Na⁺/K⁺ ATPase inhibitors (digoxin and ouabain), calcium channel activator (BAYK8644), and a nonselective 5-HT₄ agonist (tegaserod), which were withdrawn from the market due to adverse cardiovascular events, were also validated in the present study and showed similar acute hazard scoring outcome as determined in Cor.4U cell line. As expected, BaCl₂ (an I_K1 inhibitor) had no effect on hiPSC-CMs since it is known that most current commercial hiPSC-CMs lack functional I_K1 channels (immature phenotype) (Goversen et al., 2018; Pourrier and Fedida 2020). Similarly, I_Ks inhibitor (JNJ-303) was not detected in hiPSC-CMs (Kopljar, 2018 #245).

In the previous score system with Cor.4u cell line (Kopljar et al., 2018b), significant drug-induced decrease in beat rate did not significantly impact hazard score because the maximal decrease in beat rate was only adding 5 points while the threshold for the minimal hazard score (Yellow; low risk) was 10 points. In the current updated scoring system, significant decreases induced by the muscarinic acetylcholine receptor M₂ agonist (carbachol), adenosine receptor agonist (adenosine), and funny channel current (I_f) inhibitor (ivabradine) were identified as potential “low” risk (Yellow) by adding 10 points for the decrease in beat rate by ≥ −30% (delta/delta from baseline and 0.1% DMSO control) (Figure 4). With this update in the scoring paradigm for changes in beat rate, we identified 29 out of ∼3,000 screened reference and proprietary compounds that decreased beat rate by ≥ 30%. These 29 compounds include carbachol, adenosine, zatebradine, ivabradine, arecoline, methacholine, ibutilide, flecainide, and other internal NCEs.

To investigate the lot-to-lot variability in responses/scores of hiPSC-CMs provided by the vendor, some of the reference drugs that prolong QT and/or have TdP risks (total of 19 compounds) were tested in different lots.

The mean baseline Ca²⁺ transient parameter values were amplitude (3,470, 2,854, and 2017 RLU), beat rate (28, 26, and 34 beats/min), and CTD₉₀ (947, 1,006, and 790 ms) for lot 12,012 (iCell Cardiomyocytes²) (n = 2,463), lot 105091 (iCell² Cardiomyocytes²) (n = 521), and lot 103674 iCell Cardiomyocytes) (n = 400), respectively. There were no incidences of EADs or cessation of beating at baseline. The responses to the positive controls (isoproterenol, nimodipine, and dofetilide) and the negative controls (cetirizine or DMSO) were qualitatively similar among the three lots. For example, dofetilide (3 nM) prolonged CTD₉₀ by 305% in lot 12,012 (n = 144), 330% in lot 10,591 (n = 28), and 234% in lot105091 (n = 20), and incidence of EADs was similar among the two iCell Cardiomyocytes² lots and the iCell Cardiomyocytes: 63, 71, and 65% incidence of EADs, respectively.

Responses to three drugs that do not prolong QT or have no TdP risk (captopril, ebastine, atenolol) were similar among the iCell Cardiomyocytes² and iCell Cardiomyocytes² lots: “no hazard”/green color, Figure 4). Interestingly, the cardiac risk scores for 10 drugs known to prolong QT and induced TdP were also similar in two lots of iCell Cardiomyocytes² and iCell Cardiomyocytes. However, EAD incidence for dl-sotalol was much higher in iCell Cardiomyocytes (10 µM) compared to the two lots of iCell-Cardiomyocytes² (3 and 10 µM). Five reference drugs with other mechanisms of acute cardiac risk (hERG activator, IK_ATP channel opener, Ca²⁺ antagonist, and Na⁺ channel blockers) were also similar among the three lots with slight variations acceptable for any in vitro assay (Figure 5).

We analyzed the potential acute cardiac risks of 55 reference drugs with known degrees of risk in humans using the current hazard scoring system based on changes in beat rate, amplitude, CTD₉₀, and incidence of arrhythmias. We compared the cardiac risk of these 55 reference drugs based on our scoring system to the potential for clinical QT-prolongation/TdP risk, QT-shortening, QRS-prolongation, or all potential cardiac arrhythmias (i.e., non- TdP like VT/VF, bradycardia, etc.). Based on the numbers of true positives (TP),true negatives (TN), false positives (FP), and false negatives (FN), we calculated sensitivity (TP/(TP + FN)), specificity (TN/(TN + FP)), and balanced accuracy (TP + TN).

In summary, the total cardiac risks prediction was at 95%: 10 no cardiac risk drugs were all negative (green score), and for QT-prolongation/TdP risk on this CTCM assay with a balanced predictivity of 90%, sensitivity of 85%, and 95% specificity (Figure 6). FP was at 0 out of 43 while FN was at 2 out of 12 compounds (BaCl2 and JNJ-303: experimental drugs).

Fifty-five reference drugs were tested in iCell Cardiomyocytes² in the current study and compared to Cor.4U in the CTCM assay (Kopljar et al., 2018b). The acute cardiac hazard scores of these 55 reference drugs were similar between the 2 cell lines. The prediction values of acute cardiac risk and QT prolongation/TdP risk in both cell lines were >90% (Figure 5A). When taking into account the fold over its fC_max (Figure 7B): <10 -fold fC_max: the predictive value with the Cor.4u cell line was slightly higher, and at <30 or <100 -fold fC_max, the iCell Cardiomyocytes² had slightly higher predictive values compared with the Cor.4u cell line. However, both cell lines were equal in defining cardiac risks in general.

In addition, we used statistical modeling of proarrhythmic potential for the 55 drugs with known clinical risk for QT-prolongation and TdP based on the results from CTCM assay and hazard scoring system. Eight endpoints from the CTCM assay were used for potential mode predictors that include seven predictors defined in the CiPA publication ((Blinova et al., 2018)), and an additional predictor: cessation of beating was included (Lu et al., 2019): Predictors 1 and 2 indicate a test drug that elicited an arrhythmia-like event (i.e., early afterdepolarization: EAD): Predictor 1 (binary): did any drug induce EAD at the tested concentration [0 = no EAD; 1 = EAD)]; Predictor 2 (binary): EAD observed at any concentrations in ≥40% wells (0 = no; 1 = yes); Predictor 3: first drug concentration with statistically significant (p ≤ 0.05) prolongation or shortening of repolarization; Predictor 4: defined the maximal drug-induced changes in repolarization (CTD₉₀) observed at any concentration; Predictor 5: defined drug concentration (multiple of fC_max) at the first statistically significant (p < 0.05) prolongation of CTD₉₀; Predictor 6: accounted for the first drug concentration relative to its clinical fC_max to induce EAD; Predictor 7: had drug-induced changes in CTD₉₀ (ms) at ∼ its free C_max; Predictor 8 (binary): caused cessation of beating at any concentration in ≥40% of wells (0 = no; 1 = yes).

As described in the CiPA publication, two logistic regression models for the potential TdP risk group were used: Model 1: had high or intermediate risk vs. low risk and Model 2: had high risk vs. low risk and intermediate risk vs. low risk.

The results of the 55 reference drugs including a large set of CiPA compounds in the statistical model are shown in Figure 7. The models predicted well for high- and intermediate-risk drugs as well as for low-risk drugs with identified incidence in humans. There were very limited drugs overlapping between the intermediate- and low-risk drug (e.g., raloxifene). However, the concentrations used in the study were substantially higher compared with its f C_max in humans (1.2 nM). Raloxifene is a selective estrogen receptor modulator that is used for women with postmenopausal osteoporosis and is not associated with QT-prolongation in clinic. Raloxifene was reported to inhibit I_Kr with an IC₅₀ at 1.1 µM and I_Ks with an IC₅₀ at 4.8µM, and voltage-gated Na⁺ current with an IC₅₀ at2.8 µM (Liu et al., 2010). This may explain the identified activity in the current assay at much higher concentrations compared with free C_max (>1,000 fold its C_max) that could be associated with some risk for QT-prolongation.

Sequentially, we investigated 839 NCEs including Janssen discovery research projects based on molecules (Figure 8A). Compounds were tested at a wide concentration range (0.1–10 µM). Hazard evaluation identified that the most of NCEs were classified within the “no hazard” group at concentrations less than 10 μM, with a portion showing risk at the highest concentration of 10 µM. 839 NCEs (54%) were found to be no hazard over the entire tested concentration range. Additionally, for compounds with identified hazard, the levels increased in a concentration-dependent manner. At 1 µM and higher, over 50% of the compounds did show some effects associated with a certain hazard level in hiPSC-CMs, with the “high hazard” mainly at 10 µM. The proportion in the “very high” compounds, linked to arrhythmic-like EADs, was relatively small at 10 µM with some incidence also observed at lower concentrations.

Although the hazard score classification is used as an initial indicator for potential for cardiac toxicities, further analysis in other models is necessary to help to better understand the pharmacological mechanisms involved, ultimately leading to a comprehensive integrated hazard identification. Therefore, we separated the low from high hazard groups (at 1, 3, and 10 µM) in terms of the directional effect on CTD₉₀ (Figure 8B). CTD₉₀ prolongation was mainly cause of hazard labeling among compounds for all hazard-concentration combinations (Figure 8B). At 1 and 3 μM, high hazard identification was mostly caused by CTD₉₀ prolongation, and by the combination of CTD₉₀ prolongation and beat stop (cessation of beating) at 10 µM.

A sum of 121 compounds were tested in both the hERG patch-clamp and CTCM assays at the similar concentration ranges (Figure 8C). There were 66 compounds with no risk in the CTCM assay: 53 out of 66 compounds showed hERG inhibition ≥25% while 13 of 66 compounds showed no effect in the hERG assay. There were 55 of 121 compounds that had low to very high hazard score in the CTCM assay, and only 5 of these 55 compounds (9%) showed no effect in the hERG assay, while 50 of 55 compounds (91%) were confirmed active in the hERG assay, indicating that the drug-induced hazard scores were primarily due to inhibition of hERG current. The difference between the CTCM assay and the HERG assay could be caused by the effects on the CTCM assay not only mediated by affecting HERG current but also on other cardiac ion channels/receptors, or due to small % of false positive and negative on HERG assay.

Seventy-three compounds (acute hazard scores from the CTCM assay) were also tested in the isolated rabbit ventricular wedge preparation (Figure 8D). From 52 of these 73 compounds that had hazard scores in the CTCM assay, 96% (50 out of 52 compounds) had effects on either QT interval or QRS or both QRS and QT-interval in the isolated rabbit wedge assay, while only two (4%) had no effects on QRS or QT-interval in the isolated wedge preparation. Of the 21 of 73 compounds that had no hazard score in the CTCM assay 43% (9 of 21 compounds) were confirmed to have no effects in the isolated rabbit wedge preparation while 57% (12 of 21 compounds) had effects on QT interval and/or QRS: 8 compounds had QT effects; 3 had both QT and QRS effects, and 3 widened only QRS duration. However, the effects of these compounds on QT interval and QRS duration in the wedge preparation were relatively small.

Twenty-six compounds that had hazard scores in the CTCM assay were also tested in anesthetized guinea pigs (in vivo) (Figure 8E): Of the compounds that had no risk (green) in the CTCM assay, eight were also found to have no effects in anesthetized guinea-pigs in terms of changes in QRS duration and QTcB interval. One compound (terfenadine) slightly shortened CTD₉₀ and caused cessation of beating in the CTCM assay, but significantly prolonged QRS duration and QTcB interval in the anesthetized guinea pig at much higher plasma levels than that of the CTCM assay. Terfenadine is known to have solubility issues in vitro and possibly resulting in less than nominal concentrations in the CTCM assay (Lu et al., 2012). In addition, terfenadine is rapidly metabolized in the liver to an active metabolite. Of seventeen compounds that scored yellow/red/black risk hazard scores in the CTCM assay, 71% were confirmed as positive in the anesthetized guinea pig. However, 29% of these 17 compounds (5 of 17 compounds) did not have significant effects in anesthetized guinea pigs in terms of changes in QRS duration and QTcB interval. Explanations for some of these discrepancies include: One compound slightly shortened CTD₉₀ and tended to shorten QTcB interval in vivo; the free plasma levels of another four compounds (high plasma protein binding in vivo) were much lower than in the CTCM assay.