Advancing Cardiotoxicity Screening with 3D Cardioids

Overview

To improve the predictive accuracy of in vitro cardiotoxicity screening, we conducted a comprehensive benchmarking study comparing compound responses observed in 3D Cardioids with conventional 2D iPSC-derived cardiomyocytes. By evaluating both model systems under matched experimental conditions, this study aims to highlight the physiological relevance and sensitivity of Cardioids in detecting compound-induced cardiotoxic effects.

‍

Study Design

Using a commercially available, curated library of reference compounds and controls, we assessed the functional responses of 3D Cardioids and 2D cardiomyocytes side by side.

Both 2D and 3D models were generated in-house from the same batch of human iPSCs to ensure consistency. The 2D iPSC-derived cardiomyocytes were produced using an established protocol (Campostrini et al., 2021, Nature Protocols). Cardioids were generated using proprietary protocols from HeartBeat.bio. Each system was prepared in three independent biological replicates, with four technical replicates per compound in each replicate.

‍

Screening Workflow

We utilized a well-characterized compound library from Enzo Life Sciences, composed of agents targeting a variety of known cardiotoxic pathways. Compounds were tested at a final concentration of 10 µM. Prior to treatment, the culture media was replaced with a BSA-free formulation, and a baseline (“PRE”) scan was captured for every well. Functional readouts were collected at 30, 45, and 90 minutes post-treatment using the FLIPR system from Molecular Devices, with calcium transients serving as a proxy for functional cardiomyocyte activity.

‍

Data Analysis: Multi-Parametric Toxicity Scoring

To quantify cardiotoxic effects, we developed a custom multi-parametric scoring system based on a number of functional key features — each measured at the 30-, 45-, and 90-minute timepoints. Each compound was scored based on its euclidean distance from the control baseline, with a higher score indicating a greater deviation and stronger toxic effect. While this scoring method does not isolate the specific parameter contributing to the response, it efficiently flags compounds causing disruption across any functional aspect.

Compounds were classified as cardiotoxic “hits” if their score exceeded two standard deviations above the control mean at any timepoint — a threshold selected to balance sensitivity and specificity, and adjustable for future analyses.

‍

Key Findings

Our results show that 3D Cardioids are significantly more effective in detecting functional changes upon compound administration. Out of 130 compounds tested, 39 compounds were identified as cardiotoxic exclusively in the Cardioid model whereas 35 compounds were identified as hits in both the 2D cardiomyocyte system and 3D Cardioid model.

Except for one compound, all toxic responses were detected using Cardioids. Compounds that did not produce changes in calcium dynamics were not captured, a limitation that needs to be tackled by following up with orthogonal assays such as multi-electrode arrays, biomarker release assays (ELISAs), histology, or contraction analysis using videos as necessary. For instance, several compounds not classified as hits — primarily located in the bottom-left quadrant of our response plots — primarily affect contractile strength via calcium-independent pathways. Finally, our internal and external (library-provided) positive and negative controls worked as expected, thereby validating our experiments.

‍

Conclusion

This benchmarking study confirms the added value of 3D Cardioids as a highly sensitive and physiologically relevant platform for cardiotoxicity screening. Compared to traditional 2D models, Cardioids consistently demonstrated improved fidelity in detecting cardiotoxic effects, supporting their role as a powerful tool for early-stage safety evaluation in both preclinical research and drug development pipelines. These findings also underscore the potential for expanding assay capabilities in cardiotox applications beyond calcium dynamics to capture a broader spectrum of cardiac responses.

Figure 1: Log-transformed multiparametric scores for compounds tested on 2Dcardiomyocytes (x-axis) versus 3D Cardioids (y-axis). The grey quadrant presents compounds that did not induce significant changes, falling within two standard deviations of the control. The red quadrant highlights compounds that elicited significant responses exclusively in Cardioids, while the blue quadrant includes compounds that showed strong responses in both 2D and 3Dsystems. The white quadrant captures compounds with significant effects detected only in 2D cardiomyocytes. Each point represents a compound, with the hollow point indicating the negative control (DMSO only) and the solid redpoint marking the positive control (Bay K8644-treated).