Cryopreserved hiPSC-CMs were thawed and precultured in fibronectin-coated (1:100) T75 flasks for 3 days, in Ncardia’s proprietary Cardiomyocyte Culture Medium, prior to being reseeded onto appropriate plate formats for compound assays. hiPSC-CM cultures were dissociated using 1X TrypLE™ Select Enzyme (Gibco™) and kept in serum-containing culture medium. On day 1, culture medium was switched to serum-free medium composition. hiPSC-CMs were maintained at 37°C and 5% CO₂ with regular medium changes every 48 h. Only for the experiments relating to the assessment of the culture medium compositions, hiPSC-CMs were kept in culture for up to 14 days post seeding. After ascertaining the most suitable timepoint for all subsequent assays (day 8), all compound treatments began on day 7 post seeding, during which 3–6 replicates were exposed to one concentration of a single compound or to vehicle control (0.1% DMSO) for 24 h. Compound solutions were administered at final concentration of 0.1% DMSO (Table 1).

Compound treated hiPSC-CMs seeded onto 384 well µClear black plates (Greiner) at a density of 2,000 cells per well, were stained for selected target structures. Live cell staining–MitoTracker CMXRos Red (25 nM) for 1 h; LysoTracker Red (75 nM) for 30 min at 37°C - was performed preceding fixation of samples using 4% methanol-free paraformaldehyde for 15 min at room temperature (RT). Next, cells were permeabilized with 0.01% Triton-X diluted in DPBS (−/−) for 15 min at RT. Primary antibody (Ab) mixes (Supplementary Table S1) were prepared in blocking solution containing 10% fetal bovine serum using 1:1,000 dilution scheme with the exception of the anti-PMP70 which was diluted to 1:500. Samples were incubated overnight at 4°C then washed thrice with 0.01% Triton-X. Appropriate secondary Ab mixes (1:500 dilution) were added for 2 h at room temperature. Concanavalin (200 μg/mL) staining for 1 h and Wheat Germ Agglutinin (5 μg/mL) for 10 min preceded nuclear staining with DAPI (1:1,000, 15 min at RT). Lastly, cells were washed thrice with 0.01% Triton-X as well as with DPBS (−/−).

High-magnification images (40×) were acquired using an ImageXpress Micro Confocal platform (Molecular Devices) in confocal mode. Images were analyzed in MetaXpress software version 6.6, using the Custom Module Editor, in which unique masks were designed to detect each subcellular component on a single-cell level (Supplementary Figure S1). Signals corresponding to nucleoli (anti-fibrillarin Ab) and DNA damage (anti-γH2AX Ab) were detected within the area corresponding to nuclei (DAPI), and remaining organelles within the segmented cytoplasmic area. During image analysis, the fluorescence threshold was not set as an absolute value for signal quantification but as an intensity difference to local background ensuring comparable noise to signal ratio for each dataset of an immunofluorescence staining experiment.

On day 8, post 24-h compound treatment cell samples were collected from hiPSC-CMs for RNA extraction. For each compound, the treatment concentration was selected as the dose that elicited the largest functional effect on MEA recordings without causing significant loss of viability (Table 1), so as to not bias mechanistic analysis towards cell death pathways and allow elucidation of causative toxicity mechanisms. RNA was isolated using the NucleoSpin RNA kit (Bioke) according to the manufacturer’s protocol.

After RNA extraction, the concentration and quality of the total extracted RNA were evaluated by using the “Quant-it ribogreen RNA assay” (Life Technologies) and the RNA 6000 Nano chip (Agilent Technologies), respectively. Subsequently, 10 ng of RNA was used to perform an Illumina sequencing library preparation using the QuantSeq 3′ mRNA-Seq Library Prep FWD Kit (Lexogen) per manufacturer’s instructions. During library preparation, 17 PCR cycles were used. Libraries were quantified by qPCR, according to Illumina’s protocol ‘Sequencing Library qPCR Quantification protocol guide’, version February 2011. A High Sensitivity DNA chip (Agilent Technologies) was used to control the library’s size distribution and quality. Sequencing was performed on a high throughput Illumina NextSeq 500 flow cell generating 75 bp single reads.

hiPSC-CMs were seeded in a droplet (10,000 cells/droplet), covering all electrodes of fibronectin (1:20) coated Axion CytoView 96 well plates. Cultures were maintained at 37°C, 5% CO₂, with all plates equilibrated for 30 min prior to recordings. Electrophysiological parameters were obtained from field potential recordings of contracting hiPSC-CMs using an Axion Maestro Pro device and the corresponding Axis Navigator and Cardiac Analysis tool. Plates were recorded for 5 min. Parameters analyzed included; active electrodes (number of electrodes detecting beats in percentage with beat detection set at 300 µV), beat rate (beats per minute; BPM), beat rate irregularity (coefficient of variation within each well between the eight electrodes, BRI), beat rate variability (coefficient of variation in between wells of the same plate; BRV CV), field potential duration (ms; FPD), corrected field potential duration (FPDc), FPD detection success rate (percentage of active electrodes which could also detect FPD successfully), FPD variability (coefficient of variation in between wells of the same plate; FPD CV). Calculations were performed following the equations listed: B P M = 60   s B e a t   p e r i o d   s B R I = S t a n d a r d   d e v i a t i o n   S D   b e a t   p e r i o d   p e r   e l e c t r o d e M e a n   b e a t   p e r i o d   p e r   w e l l x   100 B R V   C V = S t a n d a r d   d e v i a t i o n   S D   B P M   p e r   w e l l M e a n   B P M   p e r   p l a t e x   100 F P D c = F P D   m s B e a t   p e r i o d   s ^ 0.192 F P D   C V = S t a n d a r d   d e v i a t i o n   S D   F P D   p e r   w e l l M e a n   F P D   p e r   p l a t e x   100

Prism GraphPad software was used for additional data processing. Error bars on graphs represent the SEM indicating the precision of the estimated population mean, or SD indicating the data variability around the mean.

To ensure predictable compound action, high reproducibility, and more physiologically relevant function for hiPSC-CMs, such as a more mature metabolic phenotype (Feyen et al., 2020), several chemically defined, serum-free hiPSC-CM culture medium compositions were tested as alternatives to standard serum-containing medium (Supplementary Table S2).

The effects of media formulations on cellular electrophysiology were one of the parameters used to assess their suitability. Synchronized electrical activity of cardiomyocytes is essential for producing effective contractions and physiological function in the heart. Thus, propagation patterns of electrical impulses in spontaneously contracting monolayers of Ncardia’s proprietary bioreactor-derived ventricular-like hiPSC-CMs (Ncyte^® vCardiomyocytes, Ncyte CM1) were monitored from the first day of consistent spontaneous activity, i.e., day 4 until day 14, in all media using MEA (Supplementary Figure S2A).

Conditions in which higher rate of decreased electrical activity of hiPSC-CMs (i.e., active electrodes <50%) and arrhythmic contractions–asynchronous contractions between the eight electrodes within each well (i.e., beat-rate irregularity (BRI) >5%) were excluded (Supplementary Figure S2B). The combination of bovine serum albumin (BSA) and Knock-out Serum Replacement (KOSR) enabled stability (Figure 2A–D; Supplementary Figure S2B; time period from day 4 to 14) and metabolic function (Figures 2E–H) comparable to the serum-containing control medium. As expected, several electrophysiological parameters changed over time across the tested media conditions. In the serum-containing control medium, the percentage of active electrodes began to decline after day 8 (Supplementary Figure S2B), accompanied by increased variability in beat rate (BRV CV) in-between individual wells cultured in the same medium and a downward trend in beats per minute (BPM) (Figures 2A,B). The BSA + KOSR-supplemented maturation medium showed similar patterns, nevertheless the timing of decline was delayed. Furthermore, the BSA + KOSR-supplemented maintenance medium maintained stable values for active electrodes, BPM, and BRV CV up to day 12, indicating the most prolonged electrophysiological stability and enabling a stable assay window until day 14 (Figures 2A–D). This data supported the selection of day 8 as the most stable timepoint for further assays (Supplementary Figure S2A).

Cardiomyocytes sustain their electrical activity and physiological functions through an abundant network of mitochondria, which require a continuous supply of respiratory substrates to meet their high energy demands (Li et al., 2020; Yang et al., 2023). Disruption in the ATP-generating pathways or alterations in the mitochondrial network significantly impacts cardiac functionality (Brown et al., 2017). These characteristics can also serve as critical markers for the maturity of hiPSC-CMs. To select which candidate culture condition could facilitate an adult-like metabolic state, metabolic shift from glycolysis to fatty acid oxidation–a hallmark of adult cardiomyocytes (Ahmed et al., 2020) – was monitored. Extracellular acidification rate (ECAR), i.e., cellular glycolysis, decreased over time in all media (Figure 2F), suggesting that prolonged cell culture promotes maturation towards an adult like state. Supplemented serum-free maturation medium showed comparable oxygen consumption rate, as measured by oxidative phosphorylation rate (OCR), to serum-containing medium on day 8 (Figure 2E). However, by day 14, the OCR decreased significantly (p < 0.001). hiPSC-CMs cultured in supplemented serum-free maintenance medium displayed the lowest OCR at all time points (3.5 × 10⁻⁵ ± 3.9^–6 and 3.4 × 10⁻⁵ ± 8.1^–6). As the OCR remained at similar levels with the serum containing media up to day 8, this was chosen as endpoint. Quantitative morphological assessment of mitochondria was consistent with OCR results, showing higher mitochondrial signal intensity (1,674 ± 28.87 vs. 1,134 ± 125.6 RFU) and area coverage (17.93% ± 0.34% vs. 10.07% ± 2.06) in maturation medium (Figures 2G,H).

Cardiac-specific marker troponin T (cTNT) immunostaining at all time points reached cTNT positivity ≥95% (Figures 2I–K), indicating that serum-free media did not prompt preferential growth of non-cardiomyocyte cells.

Maturation promoting medium supplemented with 0.5% total BSA plus 1% total KOSR was selected for further experiments, due to its beneficial effects on metabolism.

Subsequently, to enable scalable morphological profiling of compound-induced cardiotoxicity, high-content imaging (HCI) protocols were optimized for detecting structural changes in subcellular organelles. In comparison with the traditional cell painting assay (Bray et al., 2016; Pahl and Sievers, 2019; Cimini et al., 2023) the panel of fluorescent dyes and antibodies was expanded (Supplementary Table S1). Staining protocols targeting DNA damage (γH2AX), sarcomere (cTNT), gap junctions (connexin 43, CX43), nucleoli, mitochondria, lysosomes, peroxisomes, the Golgi apparatus and endoplasmic reticulum (ER) were validated in a multiplexed 384-well format, forming the basis of our custom organelle-profiling assay.

Protocol optimization involved comparing single versus multiplexed staining strategies using well-characterized positive control compounds for each organelle (Supplementary Figure S3). For instance, since doxorubicin is associated with DNA damage, as well as gap junction and nucleoli defects, corresponding staining protocols were tested. Initial nucleolar visualization employed SYTO14, a nucleic acid dye with a broad excitation/emission spectrum. However, due to significant channel bleed-through in multiplexed conditions, doxorubicin-induced nucleolar alterations were only detectable in single-stain formats. Consequently, SYTO14 was excluded from the final protocol and replaced by a fibrillarin antibody as a more specific nucleolar marker to ensure reliable detection in multiplexed imaging. Further organelle-specific responses were confirmed by known mitochondrial toxicants e.g., rotenone (Won et al., 2015) and chloroquine (Javaid et al., 2022). Rotenone reduced mitochondrial area and number, similarly to chloroquine, which also increased fluorescence intensity of the mitochondrial signal. Further chloroquine is also a lysosomotropic agent that increases pH by accumulating within these organelles as a deprotonated weak base and blocking the binding of autophagosomes to lysosomes (Homewood et al., 1972; Jia et al., 2018), which in this study led to increased lysosomal area and decreased lysosomal fluorescence intensity. Thapsigargin and brefeldin A, elevated ER and Golgi marker signals respectively, while H₂O₂, at all tested concentrations (10, 100 µM), induced a mild reduction in peroxisomal signal. Overall, except for SYTO14, compound-induced changes were comparable across single and multiplexed protocols, validating the robustness of the latter for scalable, phenotypic profiling in hiPSC-CMs. This multiplexed protocol was used in all subsequent compound screening assays studying the effects of a curated library of compounds (Figure 3A).

Seventeen reference compounds were selected from three categories: i) seven highly toxic chemotherapeutic agents from different drug classes ii) seven intermediate/unknown-; antiarrhythmics, anesthetics, antipsychotics, and a cardiac myosin activator iii) three non-toxic compounds; one tyrosine kinase inhibitor (TKI), one platelet aggregation inhibitor and one anti-diabetic agent. This selection was made to capture a broad spectrum of mechanisms, from well-known cardiotoxic effects to less understood or non-toxic profiles. Each compound was tested across a 7-point semi-logarithmic concentration range designed to span the reported clinical maximum total plasma concentration (Cmax) values, when feasible, and to balance clinical relevance with detection of perturbations across a range of exposure levels (Table 1). To counter line-to-line variability, studies were performed in three different hiPSC-CM lines generated using 2D monolayer (NC196) or 3D bioreactor (2 batches of Ncyte CM, NCRM5) differentiation protocols (Supplementary Table S3). All hiPSC-CMs showed high expression of standard cardiomyocyte markers prior to being treated with the compounds listed above (Supplementary Figure S4).

Experiments confirmed that compounds listed as non-toxic, showed no changes in viability between treated samples and concentration-matched DMSO controls. In contrast, across all hiPSC-CMs, significant reduction in nuclear count was observed for doxorubicin and amiodarone. Specifically, doxorubicin reduced the nuclei count by 67% ± 13 at 10 µM (Figure 3B), whereas the same concentration of amiodarone led to a maximal reduction of 31% ± 5 (Figure 3C). Nuclei count served to assess cardiomyocyte numbers. A secondary assay to verify changes to viability was conducted via measuring DNA content (Supplementary Figure S5). It should be noted that actual number of hiPSC-CMs and DNA content could slightly differ by the ratio of multi-nucleated cells which is typically in the range of 8%–20% (da Rocha et al., 2017; Woo et al., 2019).

In the morphological profiling analysis, doxorubicin induced the most pronounced and widespread changes across organelle-specific readouts. γH2AX, a well-established marker of early cellular response to DNA damage and genotoxic stress (Li et al., 2008) showed a clear dose-dependent increase in both fluorescence intensity and area–detected within the nuclear area of the cells–across hiPSC-CM lines. Statistically significant increases in γH2AX intensity were observed at 1 μM and 3 µM (Ncyte CM1: 1.6 ± 0.05, Ncyte CM2: 1.5 ± 0.5, NCRM5: 1.4 ± 0.04, NC196: 1.3 ± 0.14) while nuclear area was significantly elevated from 0.1 µM to 1 µM (Ncyte CM1: 13.4 ± 1.7, Ncyte CM2: 12.1.6 ± 5.2, NCRM5: 8.5 ± 2.1, NC196: 7.9 ± 1.1) (Figures 4A,B). Interestingly, at the highest concentrations, γH2AX signal intensity decreased, likely reflecting progressive cell death rather than diminished DNA damage. This biphasic response is consistent with γH2AX’s role as an early stress marker, preceding overt cytotoxicity.

To further investigate the transcriptional effects underlying the observed DNA damage response, RNA sequencing (RNA-seq) was performed using cells treated with doxorubicin at 0.1 µM (Table 1). At this concentration, selected as having shown phenotypic changes but no cell death that would cloud gene ontology (GO) analysis, a total of 125 genes were differentially expressed compared to vehicle controls, with the majority being downregulated. GO term analysis revealed significant enrichment for pathways related to cell division and nuclear organization, including downregulation of genes involved in mitotic regulation, such as cyclin-dependent kinases (e.g., CDK1) and associated regulators. Notably, FAM111B, a protease implicated in DNA repair and cell survival (Arowolo et al., 2022) showed the largest effect size among all differentially expressed genes. Importantly, TOP2A, the molecular target of doxorubicin and key component of DNA replication and repair was significantly downregulated (logFC: −5.29) (Figure 4F).

Further known toxicities of doxorubicin were detected by imaging, namely; significant increase in lysosomal intensity (p < 0.05 at 3 μM, 10 µM) and nucleolar number (p < 0.05 at 1 μM–10 µM), changes in CX43 area could also be observed, however this effect showed high variability (Figures 4C–E). Differences observed in sensitivity across cell lines highlight the importance of including multiple cell lines with individual genetic backgrounds in compound screening.

MEA experiments using Ncyte CM1 cells, showed doxorubicin compromised electrical activity by increasing BPM and BRI, even at sub-clinical concentrations ≥100 nM (Figure 5A; Supplementary Table S4). Findings from electrophysiology confirmed the highly cardiotoxic potential of doxorubicin, while imaging analysis additionally highlighted the cellular mechanisms involved.

An interesting compound from the intermediate toxicity group, amiodarone, also displayed a distinct morphological phenotype. At the highest concentration tested, the detachment of hiPSC-CMs in culture rendered the collection of adequate imaging data impossible. Similarly, cellular loss has been shown in multiple cell types (Bargout et al., 2000; Balasubramanian et al., 2019). Even in the absence of major hiPSC-CM loss at 1 μM, deregulation in genes involved in cellular adhesion and cell-cell adhesion (e.g., COL14A1↑; KITLG↑, SOX9↓, ACTC1↓, NDRG1↓, KCNJ4 ↓, Figure 6E) was detected.

As a cationic amphiphilic drug, amiodarone promotes the buildup of phospholipids within the endosomal-lysosomal system i.e. phospholipidosis (Buratta et al., 2015; Niimi et al., 2016; Sagini et al., 2021). In accordance with effects observed in primary human cardiomyocytes (Krajcova et al., 2023), increased mitochondrial fluorescence intensity and detected signal area (up to 1.8 ± 0.1 times and 1.7 ± 0.6 times), and lysotoxic effects with decreased fluorescence intensity (up to 0.69 ± 0.1 times) were observed in our study (Figures 6A–D). Additionally, downregulation of genes linked to lysosomal and mitochondrial function, such as P4HA1, were detected. Notably, this gene is also associated with the endoplasmic reticulum lumen cellular component, genes of which (P4HA1, SPON1, P3H2, COL21A1, GPX7, P4HA2, MELTF, ERO1A) were also downregulated (Figure 6E).

In MEA analysis, amiodarone slowed down BR at ≥0.3 µM, without causing FPD prolongation (Supplementary Table S4). Cessation of beating was detected at concentrations ≥3 µM. This aligns with the findings of the CiPA studies (Blinova et al., 2017; Blinova et al., 2018; Blinova et al., 2019). Moreover, treatment also affected processes regulating ion channels (e.g., KCNJ4↓, HCN1↑, CASQ2↓), muscle contraction, and actin filament organization (Figure 6E).

Out of the seventeen compounds tested, ten exhibited effects on any electrophysiology parameters at varying concentrations (Figure 5; Supplementary Table S4). Specifically, seven compounds—doxorubicin, amiodarone, lapatinib, bupivacaine, erlotinib, ponatinib, and dofetilide—were associated with either arrhythmia and/or quiescence (Figure 5). Digoxin shortened the FPD whereas dofetilide and cisplatin caused prolongation. Interestingly, erlotinib displayed a bell-shaped dose-response curve demonstrating complex interactions between dose and target engagement. Chlorpromazine had a mild effect on FPD; however, this effect size was not dependent on concentration. Six compounds had no effect on electrophysiology, including two non-toxic compounds (empagliflozin, aspirin), methotrexate, 5-fluorouracil, propofol, and omecamtiv mecarbil (Supplementary Table S4).

When examining individual morphological parameters, lysosomes were affected by doxorubicin, amiodarone, ponatinib, erlotinib, and lapatinib. Lapatinib also decreased peroxisomal area (Supplementary Figure S6). DNA damage was observed with doxorubicin only, a compound that also impacted nucleoli and CX43 (Figure 4). Mitochondria were most disturbed by amiodarone (Figure 6). Sarcomere alterations (i.e., sarcomere width) were seen with doxorubicin, amiodarone, bupivacaine and dofetilide (Supplementary Figure S6).

Results confirm that morphological analysis detects diverse mechanisms of potential cardiotoxicity and highlight its advantages as a complementary method for providing a more comprehensive cardiotoxicity assessment.

Morphological profiling generates extensive datasets, in this study comprising of 205 parameters (Supplementary Tables S5, S6), enabling the application of bioinformatic clustering techniques.

To explore broad phenotypic effects of the compounds a PCA-based approach was used. All seven concentrations tested per compound were included and treated as a single group, enabling detection of patterns emerging at any exposure level and circumventing the dose-dependent variability in compound response across hiPSC-CM lines. For each compound, a morphological ‘fingerprint’ – reflecting the pattern of morphological perturbation induced by a compound—was generated by analyzing the loading scores of the first principal component (PC1) averaged across cell lines/batches. These scores represent the relative contribution of each morphological feature to the primary axis of variance in the dataset and served as input for clustering analysis, allowing for a normalized data comparison across.

Using this approach, drugs could be divided into 4 clusters (Figure 7). Doxorubicin, as expected, formed its own cluster (cluster 1) characterized by prominent changes to the nuclei and cytoplasm. Interestingly, known non-cardiotoxic compounds (erlotinib, aspirin, and empagliflozin) clustered together with chlorpromazine, an antipsychotic agent and digoxin, a Na/K ATPase inhibitor (cluster 2). A smaller cluster (cluster 3) included ion-channel blockers, amiodarone, dofetilide and bupivacaine. This cluster represents compounds that affect cardiac electrophysiology (Falk and Decara, 2000; Lim et al., 2006; Stoetzer et al., 2016; Vaiciuleviciute et al., 2021; Tagle-Cornell et al., 2024). Evidently, bupivacaine, as observed in the clinic (Cotileas et al., 2000; Osama et al., 2024), led to decreased beat rate (3, 10 µM) (Supplementary Table S4) and increased irregularity (10 µM) (Figure 5). The fourth cluster (cluster 4) was characterized by mixed toxicity on multiple organelles, such as changes in gap junctions. The increase in cytosolic connexin CX43 observed in our study for both ponatinib and doxorubicin is in line with literature reports (Pecoraro et al., 2015; Pecoraro et al., 2017; Madonna et al., 2021) and may relate to heterogeneous post-treatment redistribution (Fontes et al., 2012). Notably, lapatinib was the only compound that changed peroxisomal parameters (Figure 7; Supplementary Figure S6). Correspondingly, GO term analysis showed deregulation of fatty-acid metabolic processes as well as peroxisomal β-oxidation including genes such as ACSL1, ACSL3, and ACOX1 (Supplementary Table S7).

In addition to unsupervised clustering, supervised classification was used to further investigate whether compound-induced phenotypes could be aligned with known clinical cardiotoxicity profiles. To this end, sPLSDA (Perez-Enciso and Tenenhaus, 2003) was applied to replicate-level data to assess the relationship between morphological and electrophysiological alterations and FDA label-based toxicity classifications (Supplementary Table S8). One concentration per compound was selected for this analysis, following the same rationale as for RNA-seq experiments: the dose that elicited the most pronounced functional effect in MEA recordings without significant loss of viability (Table 1).

Lapatinib and doxorubicin, known for their potential effect on left ventricular ejection fraction at least in some patients (Chatterjee et al., 2010; Hsu et al., 2018; Waliany et al., 2023), visibly clustered together when morphological and electrophysiological parameters were combined (Figure 8). However, this similarity was not evident from electrophysiology parameters alone. As a separate example, in the electrophysiology dataset TKIs; erlotinib and ponatinib showed similarity to dofetilide and to amiodarone, respectively. Both are known clinically to be associated with ischemia (Gover-Proaktor et al., 2017; Haguet et al., 2020) yet clustered with arrhythmogenic- and non-toxic compounds, respectively, in the imaging dataset. Unlike other compounds in the ischemia group, ponatinib carries an FDA black box warning for arrhythmia risk, (Vassallo and Trohman, 2007), which could explain its localization alongside the arrhythmia group. From the arrhythmia group, chlorpromazine showed no separation from the non-cardiotoxic compounds in either imaging alone, electrophysiology assays alone, or combined clustering. Similarly, other studies also indicate little effects of chlorpromazine in electrophysiology (Blinova et al., 2017; Blinova et al., 2018; Blinova et al., 2019; Yu et al., 2019). In summary, the addition of functional data to morphology, increased the separation of the clusters formed based on clinical information resulting in 76% of compounds clustered according to their known clinical toxicity (Figure 8C). Misclassified compounds were primarily from the ischemia group. Clustering based on imaging-derived morphological features alone (Figure 8B) produced compound groupings that closely matched those observed in the combined dataset. This convergence suggests that high-content imaging captures phenotypic signatures that are strongly aligned with clinical toxicity classifications, and in some cases may be sufficient for compound-level discrimination without additional functional data. The similarity between these panels underscores the value of imaging as a robust and scalable standalone modality for cardiotoxicity screening, particularly in settings lacking access to electrophysiological platforms. However, for certain compounds with specific functional effects (e.g., ion-channel blockers), electrophysiology provides unique complementary insight (Figure 8A).