The blood-brain barrier (BBB), which tightly regulates the exchange of substances between the blood vessel and the brain, maintains central nervous system (CNS). Therefore, the development of BBB in vitro model is expected to significantly contribute to the CNS-targeted drug development. Organ-on-a-chip systems using microfluidic devices under flow conditions have recently attracted attention for enabling cell culture environment that closely resemble in vivo. In particular, organ-on-a-chip systems utilizing closed, dual-channel microfluidic devices with a porous membrane are valuable tools for analyzing inter-organ interactions, such as those at the BBB. In this study, we aimed to investigate the influence of shear stress, induced by perfusion, on the gene expression in human induced pluripotent stem cell-derived brain microvascular endothelial cells (iBMECs) cultured in such a device for the development of a BBB-on-a-chip.iBMECs were seeded into the upper channel of a closed, two-channel microfluidic device and cultured under static conditions for 12 h. Following this, the culture medium was perfused for 48 h, and the cells were then for gene expression analysis. As a control, iBMECs cultured under static conditions using conventional cell culture inserts were similarly analyzed.We first optimized the perfusion parameters to prevent cell detachment. As a result, a trend toward increased expression of genes, including those encoding drug efflux transporters, was observed in the perfused group compared to the static culture. Notably, gene expression of Breast Cancer Resistance Protein markedly increased to levels comparable to those observed in vivo. In addition, gene expression of a glycosyltransferase C1GALT1, which has been reported to contribute to the barrier function of BBB, also showed a tendency to increase in the perfused group.In this study, we successfully established perfusion culture condition for iBMECs within a microfluidic device and demonstrated that flow-induced mechanical stimuli can significantly enhance the expression of key BBB-related genes. These findings suggest that the perfusion culture of iBMECs is a valuable approach for developing a functional BBB-on-a-chip model.
Cardiovascular diseases, with high mortality and incidence rates, drive development of drugs and treatments. Off-target effects resulting from insufficient models are responsible for 45% of drug withdrawals. Engineered heart tissues (EHT) from hiPSC cardiomyocytes are gaining traction in New Approach Methodologies (NAM) as they better recapitulate human physiology and functions, however, lack of standardization and throughput are impeding their industrial adoption. Objectives: Using the Ethica M platform, with the MChip to:1.Study tissue formation and contractile maturation of EHTs2.Validate function with force frequency relationship (FFR) and compound addition3.Validate the ability of an automated high-throughput system to generate robust, reproducible EHTs for drug screening applicationsCell/gel matrix was dispensed and tissues formed in 2 days. Spontaneous beating was tracked and videos recorded daily during the study. Continuous electrical pacing started on day 9 up to endpoint assays of FFR and compound exposure. After initial tissue formation (85%), irregular spontaneous beating began at d2, stabilizing at 0.6 Hz (day 5). Maturation was assessed by recording contraction amplitude, beat rate and pulse width. Within 14 days, spontaneous contraction amplitudes of previously stimulated tissues increased by 64%, while pacing the same tissues increased amplitudes by 165%. Contraction increases were accompanied by a beat rate reduction of 38% (0.91–0.56 Hz) and a 15% pulse width decrease, indicating a shift from fetal to adult phenotype as maturation processes in sarcomere structures, electrophysiology and calcium handling result in stronger, shorter contractions and a lower beat rate. Tissues were exposed to inotropic compounds (d31). Isoprenaline showed increased contraction strength (positive inotrope), EC50 = 9 ± 2 nM, while Nifedipine showed negative inotropy with an amplitude reduction, IC50 = 41 ± 3 nM. In both cases, tissues showed dose-dependent responses mimicking the response of mature cardiomyocytes. Functional maturity was confirmed by FFR. Electrical stimulation ranged from 0.8 to 4 Hz showing positive FFR correlation for frequencies up to 2.2 Hz, recapitulating the response of human myocardium. We demonstrate the ability of Ethica M to produce mature EHTs suitable for screening workflows that may recapitulate expected drug compound and FFR responses of the human myocardium.
In recent years, brain models derived from pluripotent stem cells have become a fundamental tool for studying common neurological disorders, such as epilepsy, Alzheimer's disease, and Parkinson's disease. The ability to measure the electrical activity of human iPSC-derived neurons in real time and label-free can provide much-needed insights into the complexity of the neuronal networks. Nowadays, combining single cell resolution with high-throughput physiological assays, which can potentially deepen our understanding of subtype-specific neuronal activity, is especially valuable and yet difficult to achieve.In this study, high-density microelectrode array (HD-MEA) platforms (MaxWell Biosystems, Switzerland) were used to perform in vitro extracellular recordings of action potentials at different scales, ranging from network-level dynamics to single-neuron and subcellular features. Moreover, we showed that the high resolution of HD-MEA systems featuring 26,400 electrodes per well is crucial for increasing the statistical power of the data collected from iPSC-derived neurons over multiple days/weeks, and for reducing variability between samples. This enabled the detection of clear, dose-dependent changes in cell activity upon compound application.Finally, we characterized the function and axonal structure of different iPSC-derived neuronal cell lines, as well as neuronal development over time, using the Axon Tracking Assay. By extracting metrics such as action potential conduction velocity, axonal length, and number of axonal branches, we could quantify this development, enabling long-term studies of neuronal maturation.These HD-MEA platforms and the extracted metrics, such as firing rate, spike amplitude, and network burst profile, among several others, provide an extremely powerful and user-friendly approach for in vitro drug screening and disease modelling.
Cardiovascular safety findings encompass structural and functional changes. These perturbations can manifest acutely or chronically and may be driven by the target or off-target mechanisms. Transcriptomics analysis of cardiac in vitro models can provide a means for predicting cardiotoxicity early in drug discovery programs and can afford mechanistic insights to derive hypotheses behind observed perturbations. Below we outline 2 example case studies of these applications. Firstly, we are applying high throughput transcriptomics to our cardiac microtissue model to identify predictive gene signatures of structural cardiotoxicity. Bulk mRNA sequencing was performed on microtissues treated with 35 compounds at 2 concentrations for 48 h. DEG analysis showed 9412 genes changed, with 312 as predictors. Application of ML models identified a random forest approach that separated none from structural cardiotoxins with a 79% sensitivity and 87% specificity. Ongoing work to increase model throughput, profile expanded compound numbers and concentrations is ongoing to enable predictions at scale for early drug discovery. Secondly, we are utilizing transcriptomics to derive mechanistic insights behind functional perturbations. Targeting epigenetic regulators can result in delayed onset effects that occur after days of compound treatment. We have profiled compound X over time in an impedance-based assay using hiPSC-CMs and found a concentration-response decrease in beat rate without changes in viability over the 7-day timeframe. A 7-day repeat-dose rat telemetry study showed a reduction in heart rate and ECG changes. Transcriptomics analysis of compound X-treated hiPSC-CMs showed GO pathway enrichment including calcium ion transport, ion channel activity and heart contraction with the enrichment score gradually decreasing as the concentration of compound increased. Downregulated genes within these pathways are responsible for generation of different phases of the action potential and contraction process. Further work to compare hiPSC-CM to rat heart transcriptomics will enable hypothesis generation that allows further investigations to understand the molecular mechanisms responsible, in vitro to in vivo translation and cross-species comparison. Successful integration of these approaches into the early drug discovery pipelines and eventual incorporation of systems pharmacology modelling will facilitate improved drug design, faster timelines to candidate nomination, and improved quantitative mechanistic understanding to predict outcomes in patients.
Drug-induced liver injury (DILI) is a serious adverse reaction that can lead to liver failure in severe cases. Despite non-clinical assessments using experimental animals, DILI remains a major cause of drug withdrawal from the market, highlighting the need for more predictive models. Recently, new approach methodologies (NAMs), such as human cells and three-dimensional (3D) culture systems, have been developed and applied in safety pharmacology. Several studies have demonstrated that 3D-cultured primary human hepatocytes (PHHs) are promising tools for DILI risk prediction; however, they have some limitations including lot-to-lot variability and limited availability. HepaSH cells, human hepatocytes derived from chimeric mice with humanized liver, are consistently available and exhibit drug-metabolizing enzyme expression comparable to PHHs, with minimal lot variations. Here, we investigated whether HepaSH cells can be useful for predicting DILI risk in humans. HepaSH cells were cultured under 2D and 3D conditions using Cellartis and William's E media, followed by evaluation of gene expression and hepatic functions. Drug-induced cytotoxicity was assessed by intracellular ATP quantification after exposure to 16 compounds selected from the FDA's DILIrank database (high risk: 6, moderate risk: 5, no risk: 5 compounds) for either 2 or 14 days. We found that 3D-cultured HepaSH cells showed higher levels of key cytochrome P450 (CYP) enzymes, transporters and albumin, consisting with increased CYP3A4 activity and albumin secretion compared to 2D-cultured cells. Cytotoxicity assessment revealed that 3D-HepaSH cells cultured in William's E medium were the most sensitive to high-risk DILI drugs. The toxic concentrations observed were comparable to those reported in 3D-cultured PHHs. These findings suggest that 3D-cultured HepaSH cells are a promising in vitro model for assessing DILI risk in humans. Further large-scale screening studies would be required to validate their predictive performance as a novel non-clinical testing platform for human DILI risk.
Cardiotoxicity assessment still represents a critical step in drug development process. At present, preclinical investigation mostly relies on in vitro and animal experimentations, known for lack of specificity and poor recapitulation of human heart behavior. Notably, the development of mature, chamber-specific (Atrial/Ventricular) cardiac models using human-induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) remains a significant challenge. Hence, advancing these models holds great potential. Here we present a chamber specific(Atrial/Ventricular) human functional 3D cardiac model developed within a beating Organ-on-Chip (OoC) platform, named uHeart, integrating fit-to-purpose assays for assessing microtissue contractility and electrophysiology and exploited for detecting drug-induced functional alterations. The model was developed by culturing Atrial/Ventricular human induced pluripotent stem cells derived cardiomyocytes(h-iPSC-CMs, AXOL Bioscience) combined with human cardiac fibroblasts (h-CFs, Innoprot) in a 75%–25% ratio, embedded in fibrin hydrogel (100 × 106 cells/mL) and cultured for up to 11 days in static and dynamic conditions. Dynamically-cultured microtissues were subjected to mechanical stimulation through a patented technology, an actuating mechanism integrated in uHeart that provides a physiological uniaxial cyclic strain (i.e., 10% strain, 1 Hz, 50% duty cycle). Following biological characterization, a toxicological evaluation was performed, assessing microtissues' contractility and electrophysiological changes upon 4-Aminopyridine and Dofetilide administration. Real-time q-PCR showed increased gene expression of light-chain-myosin MYL7 and ratio of heavy-chain-myosins MYH7/MYH6 in all dynamically-cultured microtissues. Atrial microtissues also featured increased gene expression of MYL2, sodium (SCN5A), potassium (KCNH2) and calcium (CACNA1C) channels. Immunofluorescence demonstrated chamber-specific phenotypes maintenance (atrial microtissues expressed atrial Sarcolipin, ventricular microtissues exhibited ventricular Light-Chain-Myosin). Video analysis of spontaneously beating microtissues showed greater contraction synchronicity for dynamically-cultured microtissues (Correlation Coefficient: atrial-static = 0.36,atrial-dynamic = 0.81;ventricular-static = 0.44,ventricular-dynamic = 0.93). Electrophysiological characterization showed that ventricular microtissues had a shorter beating period (BP-ventricular = 1.7 s), increased spike amplitude (SA-ventricular = 219 μV) and field potential duration (FPD-ventricular = 0.83 s) than atrial microtissues (BP-atrial = 2.9 s,SA-atrial = 144 μV,FPD-atrial = 0.66 s). Toxicological evaluation revealed that 4-Aminopyridine and Dofetilide prolonged the atrial and ventricular FPD, respectively (FPD variation respect to control: FDP-atrial = 22.4% at C_4AP = 100 μM,FPD-ventricular = 24% at C_Dofetilide = 2 nM). The study suggests that mechanical stimulation benefits atrial and ventricular microtissues' functionality. The preliminary pharmacological tests demonstrated the suitability of atrial and ventricular uHeart to be exploited for cardiotoxicity testing.
A major challenge in the clinical translation of cancer therapies is the toxicity in healthy tissues, particularly the gastrointestinal (GI) tract. Adverse events such as diarrhea, mucosal ulceration, and intestinal inflammation are frequently associated with tyrosine kinase inhibitors (TKIs) and immunotherapies, including bispecific antibodies. These effects are often linked to compromised epithelial barrier function and immune-mediated tissue damage. Given the intestine's central role in drug absorption and metabolism, it is essential to assess compound-induced GI toxicity (GIT) early in the drug discovery and development pipeline. Here, we utilize 3D human induced pluripotent stem cell-derived intestinal organoids (HIOs) as a physiologically relevant preclinical model to evaluate gastrointestinal toxicity. To quantify toxicity of clinically relevant and diarrheagenic small molecules (Afatinib, CPT11 and 5FU), we applied complementary cell viability assays and image-based analyses, including Ki-67 immunostaining to assess epithelial proliferation and cleaved caspase-3 staining to evaluate apoptosis of drug-treated HIOs across a concentration–response range (0.0001 to 100 μM) of a single dose treatment (day 1) followed by viability assessment on day 3 post drug treatment. HIOs viability of tested compounds (IC50, Afatinib 0.002 μM; CTP11 0.46 μM, 5FU n.d.) were normalized to clinical Cmax exposure and the resulting IC50/Cmax ratio plotted to score diarrheagenic and non-diarrheagenic drugs. Furthermore, we established a co-culture system of HIOs with peripheral blood mononuclear cells (PBMCs) to model immune-mediated epithelial injury and assess the GIT potential of bispecific antibodies (EpCam, 0.1 to 10 μg/mL). Our results demonstrate that human intestinal organoids provide a scalable and sensitive platform for detecting epithelial damage, capturing both direct cytotoxic and immune-related adverse effects. This organoid-based assay framework offers a powerful tool for early-stage screening of therapeutic candidates, enabling prediction and mitigation of GI liabilities prior to clinical translation.
Animal models are widely used in cardiovascular safety pharmacology but have limitations in translatability and scalability. New Approach Methodologies (NAMs) such as Engineered human myocardium (EHM) are introduced to overcome these caveats. EHM are created from induced pluripotent stem cell (iPSC)-derivatives, including cardiomyocytes, fibroblasts, and endothelial cells. For an improved design, we have created protocols for the derivation of highly enriched cardiomyocyte and fibroblast with >95% purity. For controlled vascularization, similarly purified microvascular endothelial cells are required.To establish and validate a robust differentiation protocol for the derivation of microvascular endothelial cells from iPSC as purified starting material for EHM tricultures.Three wild-type hiPSC lines (two male [myrWT1, myrWT5], one female [myrWT3]) were differentiated into iPSC-EC by a modified mesoderm induction protocol. Endothelial identity was confirmed by flow cytometry, immunocytochemistry, qPCR and morphological assessment.All three hiPSC lines yielded iPSC-EC exhibiting canonical cobblestone morphology and strong expression of mature endothelial markers, including CD144 and vWF, on a transcript and protein levels in a reproducible manner. Expression of the EC progenitor marker CD34 remained low. Phenotypic profile of iPSC-EC was comparable to commercially available primary human cardiac microvascular endothelial cells isolated from a patient heart, confirming robust and physiologically relevant endothelial differentiation. We have reached >90% purity in myrWT1, 3 and 5 models, respectively.We established and validated a iPSC-EC differentiation in three independent iPSC-models. Next steps will include comparisons between spontaneous vs defined EHM tricultures as well as validations in safety and efficacy drug screens in a high-through-put myrTissue/myrImager format.
Cardiac safety in drug development faces new challenges, as novel therapeutic modalities with immune pharmacodynamics have clinically exhibited myocarditis and other immune-mediated adverse reactions. To evolve preclinical cardiac safety assessment toward incorporating these liabilities in animal-free humanized models. We used our routine 3D cardiac microtissues (Hu-CMT) and co-cultured them with immune cells in vitro for assessing the immunogenicity and infiltration in response to tool compounds. Our work shows that human peripheral blood mononuclear cells (PBMC) significantly infiltrate 3D Hu-CMTs in response to immune activation. Establishing these readouts additionally revealed that cardiac immunocompetent co-cultures require donor-compatibility based on HLA genotypes, as our 3D-CMT (triculture composed of cardiomyocytes, endothelial cells and fibroblasts from different donors) needed to be HLA-matched with the PBMC donors (mono-allelically in HLA-A/B/C) to decrease baseline infiltration and to allow experimental resolution. To assess the applicability of the cardio-immune platform we benchmarked using 2 drug modalities with known myocarditis risks: Adeno-associated viruses (AAV), and immune checkpoint inhibitors (ICI). Both AAV-treated and ICI-treated 3D-CMTs were significantly more infiltrated by PBMCs than controls. To phenotype cardiac infiltrating immune populations, imaging mass cytometry revealed increased CD8+ T cell infiltration in response to the widely used ICI atezolizumab. We also incorporated macrophages into 3D-CMTs to mimic a tissue resident cardio-immune compartment, and to assess their immunomodulatory role upon PBMC infiltration. Macrophages not only increased the cardiac maturity markers of our 3D-CMTs but also refined the immunogenicity of the model in response to inflammatory agents, changing the selectivity of the PBMC infiltration assay in response to different compounds. Our data supports the incorporation of immune cells in cardiac assays to identify mechanistic safety concerns linked to therapy-induced myocarditis, and contributes to strengthen preclinical cardiac safety assessment in face of novel modalities. Current focus on validation against clinical datasets will allow development of a quantitative risk assessment modelling tool.
Cardiotoxicity is a life-threatening side effect of anti-cancer drugs like doxorubicin (DOX). Currently, there are no effective treatments to prevent or mitigate drug-induced cardiac damage. To discover novel therapeutic compounds that prevent DOX-induced cardiac dysfunction. We established a high-throughput screening platform combining machine learning-based evaluation of cardiac damage in human iPSC-derived mature ventricular cardiomyocytes with functional assessment of three-dimensional (3D) mature cardiac tissues. We engineered 3D cardiac tissues by co-culturing iPSC-derived cardiomyocytes with epicardial cells and developed a robust protocol to efficiently promote the maturation of these cardiac tissues by supplementing with various maturation factors, including lipids, hormones, and small molecules. These 3D mature cardiac tissues served as a cardiotoxicity testing platform. Upon chronic DOX exposure, the tissues exhibited hallmark features of cardiotoxicity, including sarcomere disorganization, elevated release of cardiac damage biomarkers, impaired calcium handling, and reduced contractile function. To enable high-throughput analysis, we developed a supervised machine learning algorithm that quantitatively scores sarcomere disorganization in 2D mature ventricular cardiomyocytes, allowing precise assessment of structural damage. Using this system, we conducted a drug screening and identified a novel compound that protects against DOX-induced sarcomere damage. The cardioprotective effect of this compound was validated in 3D mature cardiac tissues, where it significantly improved contractile function under chronic DOX exposure. We further developed a co-culture model of 3D cardiac tissues and cancer cells to assess the effect of DOX and the candidate compound on cancer cell viability. Importantly, the compound did not interfere with DOX's anti-cancer efficacy. Finally, in a DOX-induced cardiomyopathy mouse model, the treatment with the compound enhanced cardiac contractile function, resulting in significantly improved survival.This study highlights the utility of in vitro 3D mature cardiac tissues for modelling cardiotoxicity and discovering cardioprotective drugs. The identified compound holds promise as a novel therapeutic candidate for the treatment of DOX-induced cardiomyopathy.
To assess evolving practices in Investigative Toxicology (I-Tox) across the pharmaceutical industry, a 30-question survey was conducted in 2025, following earlier editions in 2015 and 2020. Seventeen mid- to large-sized pharmaceutical companies participated, all active in both traditional (NCEs, NBEs) and emerging modalities. Respondents included in vitro toxicologists from the Investigative Toxicology Leadership Forum, providing company-level input on team structure, objectives, assay capabilities, and future outlook. Most companies reported a dedicated I-Tox function embedded within nonclinical safety organizations. While I-Tox teams remain lean—around 1% of R&D staff—their focus has shifted toward high-impact project support, with greater reliance on CROs and GLP-compliant outsourcing. Internal laboratory activities have become more streamlined, but scientific scope remains broad, with growing emphasis on general toxicology and in silico approaches. I-Tox involvement now occurs earlier in discovery to enable proactive safety de-risking. Core I-Tox contributions span the R&D continuum, from target selection to clinical support. Compared to 2015, greater emphasis is placed on early-phase activities, including SAR guidance, off-target risk assessment, and chemistry support. In later phases, I-Tox focuses on elucidating mechanisms of toxicity, translational relevance, and signal interpretation in both nonclinical and clinical settings. The growing proportion of GLP work managed by I-Tox prompted further exploration of adjacent disciplines. Safety Pharmacology (SP) and Genetic Toxicology (GT) are now integrated into I-Tox functions in 50% and 75% of companies, respectively. These functions are supported by GLP-compliant assays conducted internally (25%), at CROs (58%), or through a combination of both (17%). Notably, one-third of respondents reported incorporating SP into I-Tox within the past five years. Assay availability has improved over the past decade, particularly for in vitro and in silico platforms targeting key organ systems. However, translational confidence remains a limiting factor. Technologies such as iPSC models and high-content imaging are now routinely applied, while others—like organ-on-chip and metabolomics—are still maturing. Respondents also highlighted emerging tools with near-term disruptive potential. Overall, I-Tox continues to evolve as a strategic enabler of drug safety, increasingly contributing to early de-risking, mechanistic insight, and the integration of innovative, non-animal technologies across pharmaceutical R&D.
CIPN is one of the most common side effects associated with certain classes of anti-neoplastic agents. It results in a range of sensory, motor, and autonomic symptoms which not only significantly impact patients' quality of life but often lead to dose reduction or discontinuation of anticancer therapy. While assessing CIPN preclinically is challenging, hiPSC-derived neurons offer a promising in vitro model for phenotypic screening of compounds with CIPN liability. Here, we treated hiPSC-derived motor neurons, or glutamatergic neurons co-cultured with astrocytes, with known CIPN positive controls Bortezomib, Cisplatin and Vincristine at clinically relevant concentrations. The integrity of neuronal structure was monitored in real-time with IncuCyte_S3 imager. Culture medium was collected at the end of treatment for cytotoxicity (LDH) assessment. Cells were then immuno-stained for the neuronal marker b-III tubulin, and cell count, neurite length/area and branch points were quantified with Opera Phenix high content imaging. While count of cell bodies or clusters remained relatively stable, a time- and concentration dependent reduction in neurite length and branch points was observed. Imaging via Opera Phenix, especially characterizing the branch points was more sensitive for detecting neurite loss relative to that via IncuCyte (change in branch points at 72 h treatment: −73% Opera Phenix, −47% IncuCyte). A concentration-dependent increase in LDH release was also observed, indicating neurite toxicity. Spontaneous neuronal firing was analyzed for up to 75 h on a multielectrode array (MEA, Maestro Pro) system and all three control drugs caused a similar, time- and concentration-dependent increase followed by a decrease in firing frequency or duration of spikes, bursts and network bursts. The percentage of spikes participating in bursts or network bursts was decreased as was the level of synchronized network firing. Collectively, these phenotypic changes indicate early hyperexcitability of neurons followed by decreased activity due to neurite damage and diminished network activity, which is consistent with clinical manifestation of CIPN.
Understanding interspecies differences in Drug-Induced Liver Injury (DILI) events is critical for translational risk assessment and reducing compound attrition during drug development. Although non-clinical safety testing in two animal species persists, the ability to predict human responses alongside animal is critical to progress drug candidates to the clinic, or to pause and better understand adverse outcomes if required. Organ-on-chip (OOC) systems are essential to bridge the translational gap between animal models and human outcomes, improving predictivity of non-clinical safety assessments. To address this, we employed the PhysioMimix® OOC System to develop a DILI assay optimized for primary hepatocytes derived from human, rat, and canine. 3D liver microtissues were formed and cultured in the liver microphysiological system (MPS) under dynamic perfusion. Each liver microtissue demonstrated stable species-specific functionality (CYP, albumin, urea) for up to 14 days.To evaluate differential hepatotoxic responses across species, we applied a panel of reference compounds known to elicit DILI alongside structural analogues with no known human DILI liability. A daily dosing regimen was applied at a 7-dose concentration range for 4 days. Functional biomarkers (urea, albumin) were assessed in parallel with clinical liver injury markers (LDH, ALT), offering comprehensive readout of hepatocellular health and toxicity. The system successfully captured known species-specific toxicities. For example, nefazodone (DILI Rank 8) induced hepatotoxicity in human and rat with greater sensitivity compared to canine. In human and rat, urea and albumin showed decreases at similar IC50 values. Buspirone (DILI Rank 3) was demonstrated as safe across all species. In canine hepatocytes, neither urea nor LDH indicated toxicity with nefazodone or buspirone treatment. Albumin emerged as a sensitive marker across all three species, with nefazodone inducing dose-dependent decrease at a lower IC50 than other markers. This study demonstrates the value of integrating animal and human liver MPS models for DILI assessment to improve cross-species interpretation. This approach offers a path toward improved human risk prediction by identifying species-specific liabilities earlier in the development pipeline. As regulatory interest in MPS continues to grow, these platforms have potential to better inform, complement, or reduce reliance on traditional models for liver safety risk assessments.
Cardiotoxicity detection still represents a critical step in pre-clinical phases. To obtain human-relevant data, the combined exploitation of relevant technologies such as Organs-on-Chip (OoCs) and human iPSC derived cardiomyocytes (h-iPSC-CMs) holds great promises, also to reduce animal use. Here we present a 3D cardiac model, uHeart, developed within a beating-OoC integrating readouts to detect compounds' cardiotoxicity. The model was exploited in different contexts of uses (COU) demonstrating its ability in detecting drug-induced changes in electrophysiology and contractility.Microtissues are developed by using h-iPSC-CMs and human fibroblasts embedded in fibrin (100 × 106 cells/mL, 75%–25% ratio) and cultured for 7 days under physiological mechanical stimulation (i.e. 10% strain, 1 Hz). uHeart was qualified for QT-prolongation and pro-arrhythmia detection by following the ICH S7B guidelines, using 11 drugs listed in the Comprehensive in vitro Proarrhytmia Assay (CiPA). The model was further exploited for evaluating the effect of 19 reference compounds that affect contractility by acting on calcium homeostasis, on sodium channel and on the sarcomere. RNA-seq analysis was implemented to characterize the microtissue functionality and drug responses.The mechanical stimulation promoted the formation of microtissues exhibiting spontaneous and synchronized beating (coefficient of variation of the beating period-BP < 25%; ~85% yield). The analyses on field potential (FP) (i.e., BP, spike amplitude, FP duration) revealed that the microtissues responded to drugs acting on single and multiple ion channel. Overall, uHeart demonstrates 83.3% sensitivity and 100% specificity in predicting QT-prolongation. Moreover, uHeart showed a superior outcome with respect to 2D models in detecting arrhythmic events and detect toxicities of molecules that resulted false-negative in animal models. The analyses on contractile parameters (e.g., beating period-BP, contraction time-CT, relaxation time-RT, contraction amplitude-CA, contraction velocity-CV) highlighted that uHeart has 65% sensitivity and 100% specificity in predicting inotropic effects. Notably, uHeart outperformed traditional 2D models in detecting the effects of drugs targeting the sarcomere, underscoring the importance of using a 3D model. These results highlight uHeart enhanced relevance and potential for application in more specific COUs.uHeart is a functional human 3D cardiac model designed for cardiotoxicity applications across different COUs, developed in alignment with the 3Rs principles.
QT interval prolongation induced by drugs (ΔQT) is a key biomarker of proarrhythmic potential and continues to be a major hurdle in cardiac safety assessments. Although regulatory guidance emphasizes the need for better non-clinical tools, mechanistic in silico approaches remain an underutilized asset for early-stage risk prediction.This study introduces an integrated framework that combines detailed electrophysiology simulations with AI-powered surrogate models to evaluate drug-induced ΔQT effects in a sex-specific context.We generated male and female virtual populations using 3D cardiac electrophysiological models, incorporating drug interactions via a multi-ion channel block model. Pseudo-ECGs were derived from these simulations to assess QT interval shifts. To facilitate rapid safety screening, Gaussian Process Regression emulators were developed using outputs from over 900 simulations, achieving high-fidelity ΔQT predictions with uncertainty estimation and mean absolute errors under 4 ms.As a proof of concept, the framework was applied to loperamide, a drug associated with abuse-related cardiotoxicity. The emulators enabled extensive dose-response exploration beyond therapeutic levels, identifying sex-dependent ΔQT thresholds associated with arrhythmic risk.Specifically, arrhythmogenic responses emerged at 109 × Cmax for female models and 286 × Cmax for male models (Cmax = 3.98 ng/mL). These findings highlight the differential responses and the associated ΔQT trends across concentrations.Our results illustrate how machine learning–accelerated mechanistic modelling can unlock high-throughput, clinically relevant safety assessments. This approach is particularly valuable in exploring drug overdose scenarios that lie beyond the scope of conventional clinical trials.
Background: Pro-arrhythmic and contractility liabilities are two major concerns in safety pharmacology. Human-based computer cardiac models have matured to deliver credible simulations of contractility, electrophysiology and excitation-contraction coupling, matching experimental and clinical data. Integration of the in silico human-based methodologies into safety pharmacology pipelines requires robust, professionally-developed software alongside validation studies to evaluate credibility and benefits. We present extend the context of use of human-based in silico simulations the simultaneous evaluation of drug-effects in contractility and electrophysiology using the Virtual Assay software. A novel, human-based in silico model of ventricular electromechanical activity was integrated into Virtual Assay software, which includes a user-friendly interface. The credibility of Virtual Assay to predict drug effects on contractility, electrophysiology and calcium transients was evaluated through two independent in-silico drug trials in collaboration with industry partners. For this, experimentally-calibrated populations of human ventricular cell models were generated in Virtual Assay to account for cell-to-cell variability, and drug effects were incorporated through dose-dependent ion channel inhibitions based on IC50 values and Hill coefficients. Firstly, the effects of 28 neutral/negative inotropic and 13 positive inotropic reference compounds were simulated based on experimental data available in the literature. Virtual Assay simulations consistently predicted drug-induced inotropic changes for 25 neutral/negative inotropes and 10 positive inotropes. Quantitative agreement of negative inotropic changes was observed in 86% of tested drugs. Secondly, simulations of 37 compounds were conducted in Virtual Assay and compared with ex vivo rabbit Langendorff experiments for contractility and electrophysiology. I -silico predictions using the human electromechanical model in Virtual Assay achieved quantitative matches with the ex vivo data for 76% of compounds, and qualitative matches for 92%. Moreover, the accuracy of Virtual Assay software simulations for prediction of pro-arrhythmic cardiotoxicity reached close to 90%. Simulations with the Virtual Assay software provide a user-friendly in silico tool for the simultaneous evaluation of electrophysiology and contractility with high accuracy and explainability when compared to experimental data.
Advancements in organ-on-chip (OoC) and microphysiological systems (MPS) have markedly improved preclinical drug assessment by recreating physiologically relevant microenvironments for studying drug-induced cardiac injury and central nervous system (CNS) toxicity. Yet, most current assays still rely on nominal bath concentrations and single biomarkers that fail to reflect true cellular exposure or capture complex injury pathways. Consequently, MPS platforms remain under-utilized in drug development and have not been fully validated or qualified for regulatory decision-making. These shortcomings also impede broad implementation of the 3Rs principle—Replacement, Reduction, and Refinement of animal use—in safety testing. Embedded within the Virtual Testing Center, our DigiTocs workflow has already demonstrated high accuracy for drug-induced liver injury. We seek to extend its applicability to cardiac and CNS safety by combining mechanistic pharmacokinetic (PK) digital-twin modelling with advanced artificial-intelligence (AI) analytics and integrated multi-omic biomarker data in order to (1) refine toxicity assessment in cardiac- and CNS-on-chip systems and (2) generate the evidence needed to qualify and validate these assays for regulatory use while advancing 3Rs goals. Recognizing limitations of nominal concentrations, we developed digital twins to simulate dynamic drug-concentration profiles within the chip environment, accounting for metabolism, non-specific binding, and restricted drug penetration in the cell systems, thereby yielding realistic intracellular and medium concentrations. Second, we applied AI/ML algorithms to integrate these refined exposure metrics with physicochemical properties, multiple toxicity biomarkers, and experimental conditions. This hybrid approach leverages AI/ML's predictive power to identify cardiotoxic and neurotoxic outcomes, improving sensitivity and specificity by incorporating multiple toxicity end-points. Preliminary results show an increase in specificity, reducing misclassification of compounds entering clinical trials and eliminating unnecessary follow-up animal studies. Although our analysis was retrospective, the same workflow can be applied prospectively by simulating human PK profiles and determining clinically relevant Cmax values for on-chip toxicity comparisons. By generating human-relevant safety data earlier, the platform replaces exploratory animal studies and refines those that remain through better dose selection—delivering measurable benefits across all three pillars of the 3Rs, shortening development timelines, lowering costs while accelerating regulatory acceptance of these novel approach methodologies.
Quizartinib is a FLT3 kinase inhibitor used in acute myeloid leukaemia (AML). It prolongs QTcF interval in AML patients by inhibiting the slow delayed rectifier potassium current (IKs)1. This contrasts with the usual IKr (hERG)-mediated mechanism. We investigated the ability of the ten Tusscher 2006 (tT) and O'Hara-Rudy 2011 (OHR) action potential models to predict the clinical QTcF effect of this IKs-selective drug. Ion channel data for quizartinib and its metabolite (AC886), along with clinically relevant, unbound concentrations of both entities formed inputs for Cardiac Safety Simulator™ (CSS) software. Quizartinib: IKsIC50 0.4 uM; IKr 16.4% inhibition at 3 uM. AC886: IKs and IKr 26.9% and 12.0% inhibition at 3 uM, respectively. Data were analyzed using tT or OHR models, and CSS applied a 1D-string of cells algorithm to generate a pseudo-ECG and thereby calculated QTcF. CSS also simulated the effect of hypokalaemia. Data are median (lower; upper 90% confidence interval). Quizartinib doses of 30 and 60 mg/day increase QTcF in AML patients by 15.9 ms (13.5; 18.4) and 23.7 ms (20.6; 26.2), respectively. Using the unbound Cmax exposures of quizartinib and AC886 at these doses, tT-based QTcF modelling predicted similar increases: 13.4 ms (10.2; 16.8) and 17.2 ms (14.1; 20.9), respectively. In contrast, OHR-based QTcF modelling did not predict an effect (1.5 ms (0.7; 1.9) and 1.8 ms (1.2; 2.4) using exposures seen at 30 and 60 mg, respectively).Concentration-QTcF modelling of hypokalaemic AML patients ([K+] < 3.5 mM) indicated a 6.2 ms QTcF baseline increase relative to normokalaemic patients. However, hypokalaemia did not increase the quizartinib effect on QTcF1. tT-based QTcF modelling in CSS predicted a similar outcome in hypokalaemic conditions ([K+] = 3.2 mM): baseline QTcF increased by 7.5 ms (7.2; 8.0); no effect on the quizartinib-induced increase. These data match a previous action potential modelling-based comparison indicating that the tT model is more sensitive than OHR to IKs inhibition. This indicates the need to enhance the OHR model in terms of its ability to predict the clinical effect of IKs blockers.1.Kang D et al. (2021) Cancer Chemother Pharmacol 87: 513–523.
AI applications has established itself as a powerful tool for cell biology research in a new, quantitative dimension. We present five cell research AI applications in combination with a microfluidic device, and evaluate the results. Additionally, new potential avenues for patch clamp research are highlighted. Methodologically, microscopic observations are analyzed with specific AI solutions and compared with results from conventional technology. Applications: 1) stiffness of red blood cells (RBCs) including quantitative subpopulation analysis, 2) shear stress effect (< 3 Pa) on blood cell endothelial cell (EC) adhesion, 3) osmotic fragility of healthy RBCs (potentially in hemolytic anemias), 4) healthy induced RBC aggregate formation (potential tool for inflammation research) and 5) effect of temperature, shear flow and drugs in patch clamp measurements on the same measured cell. Results: 1) AI-RBC stiffness experiments showed that glutaraldehyde (0.1%, 15 min, Hct 2.5%) results in a significant stiffening of 8% compared to untreated healthy RBCs and no subpopulation forming, 2) The mean AI-derived cell count of RBCs adhering to ECs following two minutes of fluid shear at 0.25 Pa demonstrated a 98.7% agreement with visual counting. 3) the reference range for MCF50 (mean corpuscular fragility) for classic osmotic fragility methods was established as 0.40–0.45%, whilst the AI-based range was 0.39–0.45%. The AI approach is suggested for investigating haemolytic anaemia patients, offering unparalleled simplicity. 4) the number of RBC roleaux formation induced by increased plasma with RBCs resulted in numbers identical to those of manually counted ones. 5) patch clamp evaluation studies are still ongoing. They are performed within 20 °C to 45 °C at ±0.2 °C accuracy and at a shear flow of 0.25 Pa around the patched cell. Precise data on critical switching temperatures of TRP channels or enzymes where Q10 values switch are expected. The above examples demonstrate the high potential of similar AI approaches for further applications, e.g., dose-response characteristics, substance wash-in and wash-out times, as well as other in vitro cell studies of sickle cell disease, thalassemia, coronavirus studies (SARS-CoV-2), white blood cell activation, or endothelial responses to mechanical stimuli.
In-person observation of animal behavior and clinical observations (ClinObs) within preclinical safety studies in canines are limited in time, subjective and clinical signs might be influenced by human presence. Video surveillance allows 24/7 monitoring, but analysis is time-consuming. Therefore, we developed an integrated computer vision (AI) model to enable continuous analysis of the canines' activity and clinical behavior via video streams.The goal was to validate our custom-developed computer vision model for continuous monitoring of canines in different toxicology and safety pharmacology studies. We applied the AI model for a 24/7 analysis of locomotive activity, physiological behaviors and ClinObs in two settings; i) in single dose toxicology studies and ii) synchronized with cardiovascular telemetry endpoints under baseline conditions (using Notocord-hem: activity, heart rate, mean, systolic and diastolic blood pressure). In the toxicology studies, we utilized the AI model during three baseline days and on each dosing day – covering both single and group housing conditions during day and night phases. The AI model was able to analyze per individual animal their activity and clinical behavior to evaluate any potential compound-related effects. Quantitative food/drinking consumption allowed to map out detailed timing and duration of food and water intake. The model showed accuracies of ~96% for eating and drinking. The AI model also profiled the ClinObs during day and night to complement the in-life evaluations. For example, mapping the head/body shaking ClinObs versus baseline and the control group showed that these observations were procedure-related in one of the investigated studies, rather than drug-related. For both analyzed animals, there was a strong correlation between the AI and telemetric activity tracking in 24 h baseline conditions (r2 = 0.874). More importantly, our AI model enabled a minute-per-minute overlay of behavioral and ClinObs information to cardiovascular parameters, which allows for a detailed investigation of any potential relationship. We demonstrated the real-life applicability of our model as activity, physiological behavioral and ClinObs tracker which can increase the understanding of the drug candidates' safety profile, refining preclinical in vivo studies (3Rs) and monitoring animal welfare.
