A New Approach Methodology (NAM) for the Prediction of (Nor)Ibogaine-Induced Cardiotoxicity in Humans

The development of non-animal-based new approach methodologies (NAMs) for chemical risk assessment and safety evaluation is urgently needed. The aim of the present study was to investigate the applicability of an in vitroin silico approach to predict human cardiotoxicity of the herbal alkaloid ibogaine and its metabolite noribogaine, which are promising anti-addiction drugs. Physiologically based kinetic (PBK) models were developed using in silico-derived parameters and biokinetic data obtained from in vitro liver microsomal incubations and Caco-2 transport studies. Human induced pluripotent stem cell-derived cardiomyocytes combined with a multi-electrode array (MEA) assay were used to determine in vitro concentration-dependent cardiotoxicity reflected by prolongation of field potential duration, which was subsequently translated to in vivo dose-dependent prolongation of the QTc (heart rate corrected duration from ventricular depolarization to repolarization) using PBK modeling-based reverse dosimetry. Results showed that the predictions matched well with in vivo kinetic data and QTc data for ibogaine and noribogaine available in the literature, indicating a good performance of the NAM. Benchmark dose analysis of the predicted dose response curves adequately predicted the onset of in vivo cardiotoxicity detected by QTc prolongation upon oral exposure to ibogaine and noribogaine. The present study provides an additional proof-of-principle of using PBK modeling-based reverse dosimetry as a NAM to predict human cardiotoxicity. This is an Open Access article distributed under the terms of the Creative Commons Attribution 4.0 International license (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution and reproduction in any medium, provided the original work is appropriately cited.

In our previous work, we demonstrated that in vivo methadone-induced QTc prolongation (heart rate corrected duration from ventricular depolarization to repolarization) can be adequately predicted based on in vitro cardiotoxic effects of methadone on human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) combined with PBK modeling (Shi et al., 2020a). Given that cardiotoxicity is one of the most common toxic endpoints and a main reason for discontinuing drug development (Ovics et al., 2020;Pang et al., 2019), it is of importance to validate the applicability and accuracy of this newly developed NAM by providing additional proof of principle for its prediction of cardiotoxicity. Thus, the aim of the present study was to apply the developed PBK modeling-based reverse dosimetry approach to predict the potential cardiotoxicity of ibogaine and its metabolites, which are plant-based substances that have attracted spe-

Introduction
In line with the 3Rs principle, the development of non-animal-based novel methods towards new approach methodologies (NAMs) for chemical risk assessment and safety evaluation has become a leading research topic (Andersen et al., 2019;ICCVAM, 2018;Taboureau et al., 2020). NAMs based on in vitro and in silico models generate data with high throughput and reduce animal use and costs in predicting human toxicity (Bos et al., 2020;Patterson et al., 2020). Within the framework of NAMs, in vitro toxicity assays characterize the biological effects of chemicals with a target organ specificity that reflects the relevant mode of action, while in vitro and/or computational models can capture the biokinetics related to absorption, distribution, metabolism, and excretion (ADME) (Andersen et al., 2019;Punt et al., 2020). One good example of such a NAM approach is socalled physiologically based kinetic (PBK) modeling-based reverse dosimetry, which has been shown to be of use for quantitative in vitro to in vivo extrapolation (QIVIVE), enabling the

A New Approach Methodology (NAM) for the Prediction of (Nor)Ibogaine-Induced Cardiotoxicity in Humans
noribogaine on QTc prolongation in opioid-dependent patients. The observed QT prolongation could be associated with the potential inhibitory effect of both compounds on human ether-àgo-go-related gene (hERG) channels that play a critical role in cardiac repolarization (Martin et al., 2004). Based on in vitro studies using the electrophysiological-based patch clamp technique, both ibogaine and noribogaine were reported to block the hERG channels with similar potency (Alper et al., 2016;Koenig et al., 2014;Rubi et al., 2017).
In the present study, to predict cardiotoxicity upon oral exposure to ibogaine or noribogaine using PBK modeling-based reverse dosimetry, the in vitro cardiotoxicity of the two compounds was quantified in hiPSC-CMs using the multiple-electrode array (MEA) technique. This data was combined with in vitro biokinetic data and used in a PBK modeling-based reverse dosimetry approach to predict human dose-response curves. The hiPSC-CM MEA assay quantifies the extracellular field potential duration corrected for beat rate (FPDc) as an in vitro surrogate of QTc prolongation in the human electrocardiogram (ECG). The biokinetic data describing absorption and metabolism of ibogaine and noribogaine obtained from in vitro experiments were integrated in a PBK model for ibogaine with a submodel for noribogaine in humans. Subsequently, the PBK model was used to translate the in vitro concentration-dependent cardiotoxicity to the predicted dose-dependent cardiotoxic effects on QTc prolongation. The obtained data were compared to reported human in vivo data on ibogaine and noribogaine-induced QTc prolongation. Benchmark dose (BMD) analysis of the predicted in vivo dose-response curves was performed to allow comparison to available data on human cardiotoxicity and validate the predictions. From the predicted dose-response curve, points of departure (PoDs) can be derived to define a safe level of exposure and to evaluate the risk by comparing the PoD values with exposure levels, thus providing insight into the use of the NAM to evaluate the risk of ibogaine and noribogaine-induced cardiotoxicity in the clinical setting.
cial attention due to cardiotoxicity occurring in the clinical setting (Schep et al., 2016).
Ibogaine is an indole alkaloid naturally occurring in the West African shrub Tabernanthe iboga, which was used as a traditional medicine and at religious ceremonies (Davis et al., 2017;Goutarel et al., 1993;Litjens and Brunt, 2016;Mash et al., 2018). Nowadays, ibogaine is banned or allowed only under medical supervision in most countries except for New Zealand due to its psychoactive properties (Noller et al., 2018). A single high dose of ibogaine can reduce drug-induced withdrawal symptoms in humans (Alper et al., 1999;Mash et al., 2018;Noller et al., 2018). Ibogaine is mainly metabolized by hepatic P450 cytochromes (CYPs) via O-demethylation to the primary metabolite noribogaine, which is also psychoactive and has pharmacological effects (Obach et al., 1998;Glue et al., 2016;Mash et al., 2016;Litjens and Brunt, 2016). The major enzyme involved in the conversion of ibogaine to noribogaine has been identified in both in vitro and in vivo studies to be CYP2D6 (Obach et al., 1998;Glue et al., 2015b), with minor contributions from CYP2C9 and CYP3A4 (Obach et al., 1998). Subsequently, noribogaine is cleared via glucuronidation to noribogaine glucuronide (Glue et al., 2015a(Glue et al., ,b, 2016 (Fig. 1).
The efficacy and safety of ibogaine and noribogaine for the treatment of drug addiction has been under debate for decades. Despite evidence obtained in preclinical and clinical studies showing promising pharmacological efficacy, cardiotoxicity was identified as the major safety concern for its clinical use (Schep et al., 2016). Several fatalities and case reports describe patients who received a high dose of ibogaine after which they experienced prolonged QTc intervals that could further develop to cardiac arrythmia and even sudden death (Asua, 2013;Grogan et al., 2019;Hildyard et al., 2016;Hoelen et al., 2009;O'Connell et al., 2015;Paling et al., 2012;Pleskovic et al., 2012;Steinberg and Deyell, 2018;Vlaanderen et al., 2014). Furthermore, Glue et al. (2016) observed a dose-dependent effect of Abbreviations ADME, absorption, distribution, metabolism and excretion; BMC, benchmark concentration; BMCL10, lower 95% confidence limit of the benchmark concentration resulting in 10% effect above background; BMD, benchmark dose; BMDL10, lower 95% confidence limit of the BMD resulting in 10% effect above background; CYP, cytochrome P450; ECG, electrocardiogram; FPDc, field potential duration corrected for beat rate; hERG, human ether-à-go-go-related gene; hiPSC-CM, human induced pluripotent stem cellderived cardiomyocytes; MEA, multi-electrode array; NAM, new approach methodology; QIVIVE, quantitative in vitro to in vivo extrapolation
The precoating of MEA chips (60-6well MEA200/30iR-Ti-tcr, MCS GmbH, Reutlingen, Germany) and the culturing of Pluricyte ® cardiomyocytes were done according to the manufacturer's protocol as previously described in Shi et al. (2020a). At day 7-9 post-seeding, extracellular field potentials of cardiomyocytes ( Fig. 3) were accessed using the MEA2100-system (MCS GmbH) equipped with a chamber and a heating controller that guaranteed a stable atmosphere (37°C with 5% CO 2 ). Cells were equilibrated for 20 min prior to the compound/vehicle exposure. Measurements started by replacing half of the medium in each well with fresh medium containing 0.1% (v/v) acetonitrile or 0.2% (v/v) DMSO to reach a final concentration of 0.05% (v/v) acetonitrile or 0.1% (v/v) DMSO, which was defined as baseline condition. Following the same procedure, each concentration of the model compounds was cumulatively added to the well, increasing the concentration at each subsequent addition (Ando et al., 2017;Nozaki et al., 2017). In each MEA chip, five of the six wells were (Woburn, MA, USA). Pooled human plasma and rapid equilibrium dialysis (RED) materials, including RED inserts, RED base plates and sealing tape, were obtained from Thermo Fisher Scientific (Bleiswijk, The Netherlands).

General outline of the PBK modeling-based reverse dosimetry approach
The PBK modeling-based reverse dosimetry approach to predict in vivo dose-response curves from in vitro cardiotoxicity concentration-response data included the following steps: i) establishment of the in vitro concentration-response curves for ibogaine and its metabolite noribogaine in hiPSC-CM using the MEA, ii) development of a PBK model for ibogaine and noribogaine in humans using biokinetic data obtained from in vitro liver microsomal incubations and Caco-2 transport studies, and parameters derived from in silico simulations and the literature, iii) evaluation of the PBK model, iv) translation of in vitro concentration-response curves to in vivo dose-response curves using the PBK model, and v) evaluation of the PBK modeling-based reverse dosimetry approach by comparing predicted dose-response data to data obtained from the literature on the effects of ibogaine and noribogaine on QTc prolongation. The general outline of the PBK modeling-based reverse dosimetry approach is illustrated in Figure 2.

In vitro cardiotoxicity of ibogaine and noribogaine in the hiPSC-CM MEA assay
Exposure and MEA recording The exposure and recording of the spontaneous beating of hiPSC-CMs was conducted as previously described (Shi et al.,  independent replicates. One-way analysis of variance (ANOVA) followed by Dunnett's test was conducted to assess the effect of ibogaine and noribogaine on FPDc using GraphPad Prism 5.0. Significance was determined at p < 0.05.

In vitro experimental biokinetic parameters for the PBK models
In vitro intestinal transport studies To estimate the intestinal absorption of ibogaine and noribogaine, a Caco-2 transport study was performed to determine the apparent permeability coefficients (P app ) using a previously developed method with minor modifications (Hubatsch et al., 2007;Strikwold et al., 2017). Caco-2 cells (passages 10-15 post-thawing) were seeded onto the inserts of Costar 12-well transwell plates (Corning, 0.4 µm pored polycarbonate membrane, 12 mm diameter). The apical compartment of each well was filled with 0.5 mL of a 4 x 10 4 cell suspension in culture medium consisting of DMEM supplemented with 10% (v/v) FBS, 1% (v/v) P/S and 1% (v/v) NEAA, and the basolateral compartment was filled with 1.5 mL culture medium. Given that FBS is an animal-derived component, whose production may be associated with pain and suffering, FBS-free culture medium for Caco-2 cells, e.g., serum-free DMEM containing insulin, transferrin, selenium (ITS) or MITO+™ serum extender (ITS plus growth factors) (Ferruzza et al., 2013), could be considered for future Caco-2 transport studies. The plate was incubated under atmospheric conditions at 37°C and 5% CO 2 for 21-23 days, which allows cells to grow and differentiate into a confluent monolayer. Cell culture medium was changed every 2 days.
Stock solutions of ibogaine were prepared in acetonitrile/water (50/50 v/v). Noribogaine, methadone (reference compound), antipyrine (passive transcellular control for the transport studies) and fluorescein (passive paracellular control for the transport studies) were dissolved in DMSO. Stock solutions were diluted 200-fold resulting in a final test concentration of 50 µM (125 nM for fluorescein) at which cytotoxicity was not observed (data not shown). The final concentration of acetonitrile or DMSO was 0.25% (v/v) or 0.5% (v/v), respectively.
For the transport experiment, HBSS containing 10 mM methanesulfonic acid, NaHCO 3 (final concentration 0.35 mg/mL) and the respective test compounds was used as exposure medium (pH 6.5) in the apical compartment. HBSS containing 30 mg/mL BSA was used as medium in the basolateral compartment (pH 7.4). 1.5 mL basolateral transport medium was added first, and the experiment was initiated by adding 0.5 mL exposure medium to the apical compartment. The plate was incubated at 37°C with 5% CO 2 for 20 min. Subsequently, an aliquot of 75 µL sample was collected from the basolateral compartment and transferred to a tube containing 150 µL ice-cold acetonitrile. To measure the mass balance (recovery), 75 µL exposure buffer was also collected from the apical compartment before and after the incubation and added to the tubes containing 150 µL ice-cold acetonitrile. The insert fil-exposed to model or reference compounds while one well served as the vehicle control analyzed during a similar period as the other wells to enable corrections for time-, addition-and vehicledependent effects on the field potentials. At each concentration, exposure was conducted for 15 min followed by a 2 min recording of extracellular field potentials. Data were recorded using Cardio 2D 2.12.0 software (MCS GmbH) with a sample frequency of 10 kHz and a 0.1-3.5 kHz band-pass filter.

Data analysis and statistics
Raw data obtained from the hiPSC-CM MEA assay was analyzed using MultiwellAnalyzer 1.8.6.0 software (MCS GmbH). Electrodes with field potentials of good quality, i.e., those with amplitudes of depolarization and repolarization peaks higher than 200 µV and 20 µV, respectively, were selected for further analysis (Ando et al., 2017;Sala et al., 2017;Shi et al., 2020a). Subsequently, two parameters reflecting electrophysiological activity of cardiomyocytes, namely field potential duration (FPD, duration between the beginning of the sodium spike and the repolarizing peak) and RR interval (duration between two depolarization peaks), were measured from the 2 min recording for each concentration.
In line with other MEA studies, the Fridericia formula (Eq. 1) was applied to correct for the effect of beat rate on FPD (Ando et al., 2017;Kitaguchi et al., 2017;Vandenberk et al., 2016): where the FPD and RR-interval are expressed in seconds.
Well-based FPDc were determined by calculating the relative percentage of FPDc for the exposure measurements compared to FPDc at baseline conditions (0.05% (v/v) acetonitrile or 0.1% (v/v) DMSO) set at 100%. The potential effects of time, addition, and vehicle on well-based FPDc were corrected for by subtracting the responses obtained from the corresponding timematched vehicle control well. Irregular waveforms (Fig. S1 1 ) of the field potential including arrhythmia-type changes, a flattened unclear second peak, and beating arrest may occur at high concentrations (Kitaguchi et al., 2017;Shi et al., 2020a;Zwartsen et al., 2019). For deriving concentration-response curves of model compounds, concentrations inducing irregular waveforms were excluded as the FPD and RR interval could not be defined.
To compare the potency of model compounds, BMD analysis was used to derive effective concentrations causing a defined % increase in the FPDc compared to the baseline control as described in the "Benchmark dose analysis" section. The concentration-response curves were plotted using GraphPad Prism 5.0 with four-parameter logistic fit (GraphPad Software Inc., San Diego, USA). Data were collected from three independent experiments with six well replicates (two replicates for dofetilide and isoproterenol) in each experiment. Each data point is presented as the mean value ± standard deviation (SD) of the three Eq. 1 cubated at 37°C for 1 min, and a final concentration of 0.25 mg/ mL liver microsomal protein was added to initiate the reaction. After 2.5 min incubation in a shaking water bath at 37°C, the reaction was terminated by adding 40 µL ice-cold acetonitrile. Control incubations were performed in the absence of NADPH, which was replaced with Tris-HCl. The formation of the glucuronide of noribogaine was investigated in a similar manner as described above. A total volume of 200 µL sample contained (final concentrations) UGT reaction system (5 mM uridine 5′-diphosphoglucuronic acid (UDPGA), 8 mM MgCl 2 , 25 µg/mL alamethicin) and 0.5 mg/mL liver microsomal protein in 100 mM Tris-HCl (pH 7.4). Samples were placed on ice for 30 min, which allows the pore-forming peptide alamethicin to boost the glucuronidation activity (Fisher et al., 2000;Ning et al., 2017). After pre-incubating samples at 37°C for 5 min, reactions were started by adding noribogaine from 100-times concentrated stock solution in DMSO at final concentrations ranging from 0-750 μM. The incubation time was 40 min, and reactions were terminated by adding 50 µL ice-cold acetonitrile. Control incubations were performed in the absence of UDP-GA, which was replaced with Tris-HCl.
Test concentrations of ibogaine and noribogaine were chosen at relevant human plasma concentrations and enabled the determination of Michaelis-Menten kinetic parameters. Optimizations of experiments demonstrated that the reaction rates for the formation of noribogaine were linear with time up to 20 min and with protein concentration of up to 1 mg/mL. For the formation of noribogaine glucuronides, reaction rates were shown to be linear with time up to 120 min and with protein concentration up to 0.75 mg/mL. After termination, samples were centrifuged at 18,000 x g for 5 min at 4°C. The supernatant was collected for the quantification of ibogaine, noribogaine and noribogaine glucuronide by liquid chromatography-mass spectrometry (LC-MS/ MS) as described in the "LC-MS/MS analysis" section.
For the kinetic analysis, the data for the formation of noribogaine and noribogaine glucuronide were fitted to the Michaelis-Menten Eq. 3 using GraphPad Prism 5.0 (GraphPad Software Inc.): ters were washed with PBS, then cut out and transferred to 200 µL ice-cold acetonitrile for a 15 min sonification. Samples were left on ice for 20 min and centrifuged for 45 min at 18,000 x g, after which supernatants were collected for LC-MS/MS analysis.
To check the linearity of transport of test compounds, 75 µL sample was collected from the basolateral compartment at 1, 5, 10, 20, 30 and 60 min of incubation and added to 150 µL ice-cold acetonitrile. After each collection, an equal volume of basolateral transport medium was added back as compensation. The calculation for each time point was corrected for the dilutions.
The integrity of the Caco-2 cell monolayer was checked by measuring the transepithelial electrical resistance (TEER) using a Millicell ERS-2 Volt-Ohm meter (EMD Millipore Corporation, California, USA). Prior to the transport experiment, cells with a TEER value between 600 and 800 Ωcm 2 were selected for the transport study. After the experiment, cells with a change in the TEER value of more than 20% were excluded from further analysis of the P app value (Bentz et al., 2013). Additionally, the integrity was checked by measuring the transport of fluorescein, and cells with a transport of fluorescein lower than 1% were considered suitable for the analysis.
The P app (cm/s) value was calculated using Eq. 2 as follows (Hubatsch et al., 2007;Strikwold et al., 2017): where ΔQ/Δt (nmol/s) is the amount of the test compound transported to the basolateral compartment over the incubation time, A (cm 2 ) is the surface area of the filter, and C 0 (µM) is the initial concentration of the test compound in the apical compartment.
The mass balance was calculated by comparing the total amount of compound in the apical compartment, basolateral compartment, and filter to the initial amount. The results were obtained from three independent experiments including four replicates for ibogaine and noribogaine and two replicates for methadone and antipyrine in each independent experiment. Data are presented as the mean value ± SD.
Using methadone as a reference compound, the in vivo oral absorption rate constant (ka) for ibogaine and noribogaine were calculated by multiplying the corresponding ratio of P app to P app of methadone by 0.59/h, which is the mean ka value of methadone reported in human studies (Foster et al., 2000;Wolff et al., 2000).

In vitro microsomal incubations
To determine kinetic parameters for the conversion of ibogaine to noribogaine, in vitro microsomal incubations were performed as described before (Shi et al., 2020a) with minor modifications. Incubation mixtures contained the NADPH regeneration system (final concentrations: 1.3 mM NADP + , 3.3 mM glucose-6-phosphate, 0.4 U/mL glucose-6-phosphate dehydrogenase and 3.3 mM MgCl 2 ) and ibogaine at final concentrations from 0 to 25 µM in 100 mM Tris-HCl buffer (pH 7.4) with a total volume of 160 µL. Stock solutions of ibogaine were prepared in acetonitrile/water (50/50 v/v) and were diluted 50-fold to result in the designated test concentrations in the incubation mixtures. Samples were pre-in- where [S] is the substrate concentration (µM) and v is the rate of metabolite formation (nmol/min/mg protein). V max is the apparent maximum velocity (nmol/min/mg protein) and K m the apparent Michaelis-Menten constant (µM). The in vitro catalytic efficiency expressed in µL/min/mg microsomal protein was calculated by dividing V max by K m . Data were collected from three independent experiments, and each data point is presented as the mean value ± SD.

Determination of plasma protein and in vitro medium binding
The rapid equilibrium dialysis (RED) assay was conducted to determine the unbound fraction (f u ) of ibogaine and noribogaine in pooled human plasma and in the in vitro medium used in the initial concentration was 100% mobile phase A, linearly changing to 100% mobile phase B over 7 min, which was held for 1 min. Then mobile phase B dropped to 0% over 1 min followed by equilibration of the system for 4 min. Total run time was 13 min. The injection volume was 1 µL, and the temperature of the column was kept at 40°C. The retention times for ibogaine, noribogaine, noribogaine glucuronide, methadone and antipyrine were 5.6, 5.0, 4.9, 6.3 and 5.1 min, respectively, determined using commercially available reference compounds. Quantification was performed by comparing the respective peak areas of the total ion chromatogram (TIC) to the TIC peak areas of corresponding linear calibration curves of reference compounds (R 2 > 0.999) using Browser analysis in the software LabSolution (Shimadzu).

PBK models development and evaluation
A PBK model consisting of multiple organ compartments was developed to describe the ADME of ibogaine and its metabolite noribogaine upon oral administration (Fig. 4). Noribogaine has also been reported to cause the prolongation effects on the QTc interval in humans (Glue et al., 2016). Therefore, an oral administration route was included in the submodel of noribogaine (Fig.  4), which enables modeling of noribogaine kinetics and prediction of its cardiotoxicity upon oral administration. A full description of the model development can be found in Text S1 and Ta To evaluate the model, the predicted blood concentrations and area under the blood concentration-time curve (AUC) of ibogaine and noribogaine were compared with the in vivo data reported in clinical studies (Glue et al., 2015a(Glue et al., ,b, 2016. The reported plasma-based kinetics of ibogaine and noribogaine were extracted using WebPlotDigitizer Version 4.4. (Rohatgi, 2020) and converted to blood-based kinetics by multiplying with the respective bloodto-plasma ratio (BPr) values. The evaluation was performed according to the specifications (body weight and oral dose) of in vivo studies as summarized in Table 2. A local parameter sensitivity analysis was conducted to estimate to what extent the model parameters can influence the model output. A full description of the sensitivity analysis and results can be found in Text S3 and Figure S5 1 .

QIVIVE using PBK modeling-based reverse dosimetry
Given that the in vitro endpoint FPDc can be considered a surrogate endpoint for the QTc interval in the human ECG (Shi et al., 2020a;Zwartsen et al., 2019), in vitro concentration-response curves for FPDc obtained in the hiPSC-CM MEA assay were translated to in vivo dose-response curves for QTc using PBK modeling-based reverse dosimetry. In the case of oral administration of ibogaine, a toxic equivalence (TEQ) approach was applied to combine the cardiotoxicity of ibogaine and noribogaine. Assuming that the cardiotoxicity of ibogaine and noribogaine are additive to hiPSC-CM, the internal unbound TEQ concentration hiPSC-CM MEA assay as previously described (Shi et al., 2020a). The stock solutions of ibogaine prepared in acetonitrile/water (50/50 v/v) and noribogaine dissolved in DMSO were diluted 100-fold in human plasma or in vitro medium to reach the final concentration of 10 µM in test sample solutions. A 300 µL aliquot of test sample solution was added to the sample chamber, and 500 µL PBS was added to the buffer chamber of the RED insert, after which dialysis was performed on an orbital shaker at 37°C at 250 rpm for 5 h. Following dialysis, an aliquot of 25 µL was collected from both the sample and buffer chamber and diluted with 25 µL PBS (to the aliquot from the sample chamber) and plasma or in vitro medium (to the aliquot from the buffer chamber) to eliminate potential matrix effects. Subsequently, post-treatment samples were precipitated using 300 µL cold acetonitrile/water (50/50 v/v) and left on ice for 30 min followed by centrifugation for 45 min at 18,000 x g. Supernatants were collected for LC-MS/ MS analysis. The fraction unbound (f u ) was calculated with Eq. 4 (van Liempd et al., 2011;Waters et al., 2008): The measurements were performed in triplicate in three independent experiments. Results are presented as the mean value ± SD.

LC-MS/MS analysis
The identification and quantification of compounds in samples from the Caco-2 transport studies, microsomal incubation and RED assay were performed by LC-MS/MS analysis using a Shimadzu Nexera XR LC-40D SR UPLC system coupled with a Shimadzu LCMS-8045 mass spectrometer (Kyoto, Japan). The compounds were separated by a Phenomenex Kinetex ® C18 column (2.1 x 50 mm 1.7 μm, 100 Å) connected to a precolumn and detected by a Shimadzu LCMS-8045 triple quadrupole with electrospray ionization (ESI) interface. The instrument was operat-  (Chang et al., 2011;Glue et al., 2016).
Mobile phase A was nanopure water containing 0.1% (v/v) formic acid and mobile phase B was acetonitrile containing 0.1% (v/v) formic acid. A gradient elution at a flow rate of 0.3 mL/min was applied for the analysis with the program set as follows: the f u = concentration in buffer chamber concentration in sample chamber Eq. 4 When oral administration of noribogaine was considered, the translation was performed by setting the in vitro unbound concentrations of noribogaine detected in the hiPSC-CM MEA assay equal to the unbound C max of noribogaine in the heart venous blood with a correction for the fraction unbound in human plasma and conversion from plasma to blood using BPr as described in Eq.7: was the combination of the unbound concentration of ibogaine and noribogaine in the heart venous blood taking the corresponding toxic equivalency factors (TEFs) into account. The TEFs of ibogaine and noribogaine were calculated based on their cardiotoxic potencies obtained in the hiPSC-CM MEA assay, with the TEF for ibogaine defined as 1.00. Then QIVIVE was performed by assuming the in vitro unbound concentration of ibogaine equal to the C max of unbound ibogaine expressed in TEQ concentrations in the heart venous blood as shown in Eq. 5 and 6: C total,in vitro,ibo × f u,m,ibo = C unbound,human blood,TEQ Eq. 5

Fig. 4: Schematic diagram of the PBK model for ibogaine with a submodel for noribogaine
where C total, in vitro, ibo and f u, m, ibo are the in vitro ibogaine concentration and unbound fraction of ibogaine in the in vitro exposure medium, respectively. BPr ibo and BPr nor are the blood to plasma ratio of ibogaine and noribogaine. f u, p, ibo and f u, p, nor are the respective unbound fraction of ibogaine and noribogaine in human plasma. C total, human blood, ibo and C total, human blood, nor are the concentrations of ibogaine and noribogaine in the heart venous blood, respectively. TEF ibo and TEF nor are the TEF values of ibogaine (defined as 1.00) and noribogaine (defined based on its relative potency in the MEA assay). C unbound, human blood, TEQ values were converted to in vivo oral doses of ibogaine by PBK modeling-based reverse dosimetry, using a body weight of 70 kg (Brown et al., 1997).
Eq. 6 C unbound,human blood,TEQ = C total,human blood,ibo × f u,p,ibo BPr ibo × TEF ibo + C total,human blood,nor × f u,p,nor BPr nor × TEFnor where C total, in vitro, nor and f u, m, nor are the in vitro noribogaine concentration and unbound fraction of noribogaine in the in vitro exposure medium, respectively. C total, human blood, nor values were extrapolated to in vivo oral doses of noribogaine by PBK modeling-based reverse dosimetry, using a body weight of 70 kg (Brown et al., 1997). The calculations using Eq. 5 and 6 or 7 were performed for each of the in vitro concentrations of ibogaine or noribogaine tested in the hiPSC-CM MEA assay, which enables the translation of the entire in vitro concentration-response curve to a predicted in vivo dose-response curve.

Validation of the PBK modeling-based reverse dosimetry approach
To validate the performance of the PBK modeling-based reverse dosimetry approach, the predicted dose-response curves for QTc prolongation upon exposure to ibogaine and noribogaine were compared to the respective in vivo dose-response data obtained from case reports and clinical studies (Asua, 2013

In vitro experimental biokinetic parameters for PBK models
In vitro intestinal transport studies Table 1 shows the P app values obtained from Caco-2 transport studies and the ka values of the test compounds derived from these P app values based on comparison to the data for methadone. All test compounds were rapidly transported with the largest P app value of 47.0 x10 -6 cm/s for antipyrine and the smallest P app value of 20.8 x10 -6 cm/s for methadone. The P app value of noribogaine was 1.5-fold higher than that of ibogaine.
The transport of all test compounds was within the linear range al., 2012; Steinberg and Deyell, 2018;Vlaanderen et al., 2014). Since in most case reports ibogaine administered to the patients was internet-purchased with unknown purity, the reported doses were converted to effective doses by multiplying with the lower (15%) and upper value (50%) of purity reported in Alper et al. (2012). The predicted dose-response curve of noribogaine for the validation was made using the body weight 81.9 kg, which was the average body weight of subjects as reported in Glue et al. (2016). The details of the in vivo studies are summarized in Table S1 1 . The in vitro absolute FPDc values and the in vivo QTc data were expressed as relative percentages by comparing the post-treatment data to the baseline values for a straightforward comparison.

Benchmark dose analysis
BMD analysis of reported and predicted dose-response curves for ibogaine and noribogaine was performed to derive a lower 95% confidence limit of the BMD resulting in 10% cardiotoxicity (BMDL 10 ), which can be used as the point of departure (PoD) for the risk assessment and safety evaluation of ibogaine and noribogaine. As previously described (Shi et al., 2020a), an effective level of 10% was chosen based on the fact that a BMDL 10 value is generally considered a dose level similar to a no observed adverse effect level (EFSA, 2017) and that 10% change in QTc interval over the population baseline of 407 ms, being a QTc of 450 ms, can be used as a threshold to evaluate the abnormal QTc prolongation (ICH, 2005;Mujtaba et al., 2013;Wedam et al., 2007;Florian et al., 2012). The European Food Safety Authority web-tool based on R-package PROAST version 69 (Dutch National Institute for Public Health and the Environment, RIVM, The Netherlands) was used for the BMD analysis (Shi et al., 2020a). BMD analysis of in vitro concentration-response data was performed to calculate the benchmark concentrations (BMC) resulting in a 10% change in the FPDc with lower 95% confidence limit (BMDL 10 ). The obtained BMDL 10 values were used to compare the potency and derive the TEF of noribogaine relative to the TEF of ibogaine set at 1.00. Figure 5 shows that ibogaine and its metabolite noribogaine induced a significant concentration-dependent prolongation of FPDc. The BMDL 10 is 0.11 µM for ibogaine and 0.15 µM for noribogaine, which results in TEF values of ibogaine and noribogaine being 1.00 and 0.65, respectively. Arrhythmia-type waveforms were observed upon treatment with 1 µM ibogaine and 3 µM noribogaine. The repeated addition of vehicle controls (0.05% (v/v) acetonitrile and 0.1% (v/v) DMSO) did not significantly influence the FPDc (Fig. S2 1 ). The two reference compounds dofetilide and isoproterenol both prolonged the FPDc and increased beat rates in a concentration-dependent manner, indicating an adequate performance of the hiPSC-CM MEA assay (Fig. S3 1 ).

Fig. 5: Concentration-response curves of the effect of ibogaine (black circles and line) and noribogaine (green squares and line) on FPDc in hiPSC-CM detected by the MEA
The response of the baseline condition (0.05% (v/v) acetonitrile for ibogaine, 0.1% (v/v) DMSO for noribogaine) was set at 100%. Data represent the mean of results obtained from three independent experiments each containing six well replicates. Each data point represents the mean ± SD. ***, p < 0.001.
In vitro microsomal incubations Figure S4 1 shows the substrate concentration dependent metabolism of ibogaine and noribogaine by human liver microsomes. The obtained results follow Michaelis-Menten kinetics. The apparent V max and K m derived from these data for the formation of noribogaine from ibogaine were 0.17 ± 0.033 nmol/min/mg microsomal protein and 0.63 ± 0.005 µM, respectively. The apparent V max and K m for the conversion of noribogaine to its glucuronide were 0.036 ± 0.0008 nmol/min/mg microsomal protein and 305 ± 15.8 µM, respectively. The catalytic efficiency (V max /K m ) for the formation of noribogaine was 269.8 µL/min/mg microsomal protein, which was 2,248-fold more efficient than that for formation of noribogaine glucuronide of 0.12 µL/min/mg microsomal protein. This explains the relatively higher plasma concentrations of noribogaine than of ibogaine upon dosing ibogaine (see below).

Determination of plasma protein and in vitro medium binding
The unbound fraction of ibogaine and noribogaine in the in vitro medium and human plasma were determined by the RED assay for calculation of the unbound concentration of the compounds for the in vitro to in vivo extrapolation. The unbound fractions of ibogaine and noribogaine in the in vitro medium were comparable, amounting to 0.71 ± 0.01 and 0.80 ± 0.03, respectively. The unbound fraction of ibogaine in human plasma was determined to be 0.04 ± 0.017, which is 6.5-fold lower than that for noribogaine (0.26 ± 0.05).

Evaluation of the PBK model
The PBK models of ibogaine and noribogaine were evaluated against in vivo data reported in clinical studies. Figure 6 shows that the developed PBK model accurately predicted the time-dependent change in the blood concentrations of ibogaine and noribogaine upon oral administration of ibogaine (Glue et al., 2015b) and noribogaine (Glue et al., 2015a(Glue et al., , 2016 with differences between the predicted and reported blood C max and AUC being less than around 2-fold, which is generally considered an adequate predictive performance (Badhan et al., 2019;WHO, 2010). The detailed comparisons between predicted and reported blood C max and AUC values are summarized in Table 2. For ibogaine, the predicted blood C max and AUC were 1.7-fold and 2.1-fold higher than the reported values, respectively. For noribogaine, the prediction shows an average 0.9-fold and 1.3-fold difference both in blood C max and AUC values.

Contribution of ibogaine and noribogaine to blood ibogaine equivalents
To further illustrate the contribution of ibogaine and noribogaine to ibogaine-induced cardiotoxicity in humans, a dose-dependent comparison was made between the predicted unbound blood concentration of ibogaine and noribogaine considering their TEF values. Figure 7 shows that, upon dosing ibogaine, the  (Glue et al., 2015b). Solid lines and dashed lines represent the predictions for ibogaine and noribogaine, respectively. (b) Dots, squares, triangles and crosses represent the reported blood concentrations after an oral dose of 3, 10, 30 and 60 mg noribogaine (Glue et al., 2015a), respectively, with the solid lines being the predicted blood concentrations for the corresponding doses. (c) Dots, squares and triangles represent the reported blood concentrations after an oral dose of 60, 120 and 180 noribogaine (Glue et al., 2016), respectively, with the solid lines being the predicted blood concentrations of the corresponding doses.
To further evaluate the model predictions and derive PoDs for risk assessment, BMDL 10 values were calculated from both predicted and reported dose-response curves. The BMDL 10 value of noribogaine derived from the clinical study of Glue et al. (2016) was 163 mg for the subjects with an average body weight of 81.9 kg, which is 1.5-fold higher than the predicted BMDL 10 value, amounting to 110 mg for an 81.9 kg person. This shows that PBK modeling-based reverse dosimetry can adequately predict the in vivo cardiotoxicity of noribogaine. The predicted BMDL 10 value contribution of noribogaine to the C max expressed in unbound ibogaine equivalents is substantially higher than the contribution of ibogaine itself at all dose levels evaluated. The relative contribution of ibogaine to the ibogaine equivalents increases with increasing oral dose but is still about 8-fold lower than that of noribogaine at 500 mg despite the only limited difference in the TEF value between the two compounds. This observation can be ascribed to the fact that the concentration of noribogaine is higher than that of ibogaine (see also Fig. 6a) due to the fact that the catalytic efficiency of ibogaine O-demethylation to noribogaine is more efficient than the clearance of noribogaine by glucuronidation.

QIVIVE using PBK modeling-based reverse dosimetry, its validation and BMD analysis
By applying reverse dosimetry, the in vitro concentrationresponse curves of ibogaine and noribogaine obtained in the hiPSC-CM MEA assay were translated to predicted in vivo dose-response curves for the QTc prolongation upon oral administration of ibogaine (Fig. 8a) or noribogaine (Fig. 8b). Subsequently, the predicted data were compared to the in vivo dose-response data for QTc prolongation obtained from case studies and clinical studies to evaluate the performance of the PBK modeling-based reverse dosimetry predictions (Fig. 8). Given the unknown purity of internet-purchased ibogaine described in the case studies, a range of 15 to 50% was used to correct for the effective doses. Figure 8a reveals that the predicted dose-dependent QTc prolongation for ibogaine is best in line with the reported data when a purity of 15% is considered. For noribogaine, the predicted dose-response curve is comparable with the reported dose-response data on QTc prolongation (Fig. 8b).  (Glue et al., 2016;Glue et al., 2015a), assuming subjects have similar demographic characteristics. c Subjects were administered ibogaine.

Fig. 7: Predicted dose-dependent relative contribution of ibogaine (solid line) and noribogaine (dashed line) to the Cmax expressed in unbound ibogaine equivalents for 70 kg body weight
effects on hERG channels as detected using human mammalian cell lines heterologously expressing hERG channels (Alper et al., 2016;Koenig et al., 2014;Rubi et al., 2017). We also found that ibogaine was 1.4-fold more potent than noribogaine in prolonging FPDc, which is in contrast to the hERG inhibition studies, which reported a lower (1.3-fold) potency of ibogaine compared to noribogaine. The discrepancy might be ascribed to the use of different cell models. Unlike the transfected cell lines overexpressing a single type of ion channel, hiPSC-CMs express the major cardiac ion channels and receptors present in human cardiomyocytes (Karakikes et al., 2015;Kussauer et al., 2019;Ma et al., 2011), and thus the observed results of ibogaine and noribogaine could be the result of multiple ion-channel effects. Rubi et al. (2017) assessed the cardiotoxicity of ibogaine and noribogaine in hiPSC-CMs using one concentration of each compound that prolonged the action potential duration at 90% repolarization by respectively 14.2% and 15.5%, indicating comparable potency. Additionally, arrhythmia-type waveforms were observed upon treatment of hiPSC-CMs with ibogaine at high concentrations, which is in line with case studies where cardiac arrhythmia has been associated with the intake of an overdose of ibogaine (Asua, 2013;Paling et al., 2012). Furthermore, at clinically relevant doses of ibogaine for the treatment of drug addiction (typically 500-1000 mg), the total blood concentration of noribogaine ranged from 0.7 to 4.5 µM (Mash et al., 2018), and unbound blood concentrations ranged from 0.08 to 0.47 µM when taking the unbound fraction in plasma and BPr into account. These values cover the unbound in vi-for ibogaine-induced QTc prolongation was 108 mg for 70 kg body weight, which is similar to that for noribogaine (94.2 mg, based on 70 kg body weight), indicating a comparable potency of the two compounds in inducing QTc prolongation.

Discussion and conclusion
The aim of the present study was to provide additional proof-of-principle for the prediction of potential human cardiotoxicity via QTc prolongation by combining an in vitro cardiotoxicity assay, in vitro-derived biokinetic parameters, and PBK modeling-based reverse dosimetry into a NAM for human risk and safety assessment. Two herbal alkaloids, ibogaine and noribogaine, which are promising anti-addiction drugs, were selected as model compounds since their cardiotoxicity is a major safety concern related to their clinical use, while so far not well studied. In addition, available in vivo kinetic and QTc data available for human subjects who had been exposed to the compounds enabled the evaluation of the developed NAM against clinically observed responses.
In the current paper, the electrophysiological cardiotoxicity of ibogaine and noribogaine was assessed using hiPSC-CMs on a MEA platform, which has been used previously for detecting drug-induced QTc prolongation and proarrhythmia (Shi et al., 2020a,b;Satsuka and Kanda, 2020). The results reveal that ibogaine and noribogaine prolong the FPDc in a concentration-dependent manner, which could be explained by their inhibitory  (Alper et al., 2012). Dots in (b) represent the in vivo dose-response data for noribogaine reported in Glue et al. (2016). ibogaine related parameters, including metabolic parameters, were not influential to the unbound TEQ concentration expressed in ibogaine equivalents, which however was sensitive to noribogaine related parameters. Furthermore, results show that the unbound TEQ concentration and C max of noribogaine in the heart venous blood were affected by these noribogaine related parameters to a similar extent.
Upon evaluation of the newly defined PBK models, they were used to translate the in vitro concentration response curves obtained in the hiPSC-CM-MEA assay to in vivo dose response curves for cardiotoxicity of ibogaine and noribogaine. To evaluate the predictions of the PBK modeling-based reverse dosimetry NAM, the predicted dose-response curves of ibogaine and noribogaine were compared to available in vivo data. The results obtained show that the predicted dose-response curve for ibogaine is in line with the reported QTc prolongation data, especially when the reported doses were corrected for a purity of 15%, which could be considered as a representative and realistic purity of internet-purchased ibogaine as reported by Hoelen et al. (2009). It is important to note the large variation in QTc prolongation data at similar oral doses observed in some case reports, which may be related to the individuals' diverse demographic characteristics and/or potential QTc prolonging risk factors, which were not well-documented in these studies. When applicable, applying exclusion or inclusion criteria to the reported data may better illustrate the dose-dependent effects of ibogaine and thus further improve the accuracy of the evaluation. For noribogaine, the predicted dose-response curve matches well with reported dose-response curves for QTc prolongation, with a difference in the BMDL 10 values derived from the predicted and reported data being less than 1.5-fold, which further demonstrates that the developed QIVIVE approach can adequately predict the in vivo cardiotoxicity for humans.
Based on human experiences, the administered dose of ibogaine for treating drug addiction has a wide range and varies from 6 to 30 mg/kg body weight (equal to 420 to 2100 mg for 70 kg body weight) (Alper et al., 1999;Davis et al., 2017;Mash et al., 2018;Noller et al., 2018;Schep et al., 2016). These dose levels are 4-to 21-fold higher than our predicted BMDL 10 value (96.9 mg for 70 kg body weight) for the ibogaine dependent induction of QTc prolongation. Since a BMDL 10 value generally represents a dose level at which the adverse effect is considered negligible (EFSA, 2017), the prolonged QTc would be expected at doses higher than the predicted BMDL 10 values, which is in line with the observation that QTc prolongations and arrythmia were observed in the subjects administered dose levels of ibogaine above the predicted BMDL 10 and that doses higher than 20 mg/kg bw ibogaine (equal to 1400 mg for 70 kg body weight) were associated with fatalities (Mash et al., 2018). Our predictions also confirm that ECG monitoring is essential for patients receiving ibogaine (Glue et al., 2016). In addition, many studies demonstrate that ibogaine and noribogaine have different neurobiological profiles (Baumann et al., 2001a,b;Maciulaitis et al., 2008), indicating that noribogaine might be more efficient than ibogaine in mediating certain pharmacological effects (Mash et al., 2016). Based on our model predictions and reported obser-tro effective concentration (BMCL 10 ) of noribogaine of 0.12 µM, a value corrected for the unbound fraction in the in vitro medium obtained in the current study, indicating that the cardiotoxicity of noribogaine contributes to ibogaine-induced cardiotoxicity and should thus be considered in the reverse dosimetry.
Results obtained from in vitro microsomal incubations reveal a high catalytic efficiency for the metabolism of ibogaine to noribogaine, which is in accordance with published data indicating that ibogaine is a compound with high intrinsic clearance with the majority being metabolized to noribogaine (Obach et al., 1998). Based on studies with human liver microsomes, Obach et al. (1998) suggested that two enzymes with different activities were involved in the formation of noribogaine, and the apparent V max and Km for the high-affinity enzyme appeared comparable with our data. However, the reported biphasic kinetics were not observed in our incubations. Furthermore, we found that the catalytic efficiency for the conversion of noribogaine to noribogaine glucuronide was quite low, which is consistent with the fact that only small amounts of noribogaine glucuronides were reported to be formed following an oral dose of noribogaine (Glue et al., 2015a). The substantially higher catalytic efficiency for conversion of ibogaine to noribogaine than for glucuronidation of noribogaine is also reflected in the kinetic data of the present study and explains why upon dosing ibogaine, plasma levels of noribogaine exceed those of ibogaine itself. Additionally, by comparing the unbound blood concentration corrected for TEF values of ibogaine and noribogaine, we found that noribogaine is predicted to be a major contributor to the unbound TEQ concentration expressed in ibogaine equivalents. Therefore, it can be speculated that noribogaine instead of ibogaine itself plays a dominant role in the in vivo cardiotoxicity upon the oral administration of ibogaine. The discrepancy between the relatively high cardiotoxic potency of ibogaine observed in the hiPSC-CM MEA assay and a relatively small contribution to the in vivo cardiotoxicity could be explained by our findings that i) ibogaine can be efficiently and extensively metabolized to noribogaine, ii) ibogaine highly binds to the plasma protein, resulting in a low unbound internal concentration of ibogaine, which is the fraction generally assumed to be responsible for the therapeutic or toxic effect of drugs (Smith et al., 2010), iii) clearance of noribogaine to its glucuronide is less efficient, and iv) the protein binding of noribogaine is somewhat less than that of ibogaine, resulting in higher unbound concentrations.
The blood C max and AUC values of ibogaine and noribogaine predicted by the PBK model are comparable with the literature data, with differences between the predicted and reported values being less than 2-fold, indicating an adequate predictive performance of the developed PBK models.
Based on the sensitivity analysis (Text S3 1 ), the influence of metabolic parameters on the predicted C max of ibogaine in heart venous blood showed a dose-dependent change. When the oral dose of ibogaine increased to 500 mg, the influence of V max increased while K m was less influential. This may be explained by the fact that at high dose levels metabolism gets saturated, which reduces the influence of the K m with metabolic clearance being dependent on V max . It is also of interest to note that most of the formance of the developed in vitro-in silico approach. Obtained predictions also reveal a similar in vivo cardiotoxicity potency upon oral administration of ibogaine and noribogaine, while noribogaine might play a substantial role in ibogaine-induced QTc prolongation. Altogether, the present study shows additional proof-of-principle for using a NAM consisting of PBK modeling-based reverse dosimetry of hiPSC-CM MEA assay data for the prediction of human cardiotoxicity, which can be used for cardiac safety evaluation.
vations (Glue et al., 2016), a 1.2 to 1.6-fold higher dose of ibogaine is needed to reach the same C max of noribogaine compared to the direct intake of noribogaine. Considering a similar in vivo potency observed in the current study for the predicted in vivo induction of the unwanted side effect of QTc prolongation by the two compounds (similar predicted BMDL 10 values), noribogaine would be a safer option.
In the present study, we consider that the relatively small contribution of ibogaine itself to the in vivo cardiotoxicity could be due to its low unbound fraction in plasma and its extensive metabolism. However, as ibogaine is a basic compound, the plasma protein binding could be influenced by the level of α1-acid glycoprotein, which has an up to 10-fold variation in human plasma (Smith and Waters, 2019). Furthermore, it has been reported that the internal concentration of ibogaine was up to 43-fold higher in poor CYP2D6 metabolizers compared to extensive metabolizers (Mash et al., 2001;Glue et al., 2015b). Glue et al. (2015b) also reported that the C max of ibogaine was 26-fold higher in CYP2D6 extensive metabolizers who took CYP2D6 inhibitors compared to the ones who took placebo. Based on these observations, the cardiotoxicity of ibogaine might be more apparent in these sensitive individuals. Moreover, due to limited information on the metabolism of noribogaine, only the glucuronidation was included in the current model. Other metabolic reactions such as sulfation might also be involved in the clearance of noribogaine (Glue et al., 2016). Considering the relatively high potency of noribogaine in in vivo cardiotoxicity, it would be of interest to have a comprehensive metabolic profile for noribogaine defining also potential minor pathways for its clearance.
Based on findings in the present study, the developed in vitro-PBK modeling based approach could be extended to a higher throughput NAM for a broader range of chemicals inducing QTc prolongation, which ultimately would contribute to the use of non-animal based NAMs for human cardiac risk assessment and safety evaluations of chemicals. This can be achieved by employing a higher-throughput platform of the MEA (e.g., 96-well plate) (Kraushaar and Guenther, 2019), which allows the hiPSC-CM MEA assay to be used for rapidly detecting cardiotoxicity for large numbers of chemicals. Additionally, the development of a generic PBK model that only includes a limited number of, and thus the most influential, input parameters could accelerate the efficiency of PBK modeling, contributing to enhancing the throughput of the developed NAM. As illustrated in the present study and based on the sensitivity analysis (Text S3 1 ), such input information could include physicochemical parameters, intestinal permeability, unbound fraction in plasma, and kinetic parameters for hepatic metabolic clearance. The development of high-throughput biokinetic assays to derive these parameters would further contribute to the efficiency of model parameterization In conclusion, we have demonstrated that integrating in vitro cardiotoxicity data obtained with hiPSC-CMs in the MEA assay, in vitro biokinetic data, and PBK modeling can be a promising NAM to predict the in vivo dose-dependent cardiotoxicity of ibogaine and noribogaine in humans. Comparison of the predictions with the available in vivo data indicated the adequate per-