The assay combines two complementary technologies. The first is a microelectrode array (MEA) that records the electrical activity of networks of human induced pluripotent stem cell (hiPSC) neurons. These cultures, supplied by Fujifilm Cellular Dynamics, consist of roughly 68 percent glutamatergic neurons, 17 percent GABAergic neurons and 15 percent astrocytes, a carefully controlled ratio designed to mimic key features of human brain circuitry. After 21 to 25 days in culture, the neurons form functional networks whose spikes and bursts can be monitored before and after compound exposure using an Axion Biosystems Maestro instrument. The second component is a panel of human ion channel assays, run on automated patch clamp electrophysiology platforms, that measures how potently a drug inhibits channels long associated with seizure biology, including sodium channels such as NaV1.1, NaV1.2 and NaV1.6, potassium channels such as Kv7.2 and Kv7.3, GABA receptors, NMDA receptors and nicotinic acetylcholine receptors.

To interpret the data, the researchers quantified changes in a defined set of MEA parameters, including mean firing rate, burst frequency, network burst frequency, the number of spikes per burst and the mean inter-spike interval within bursts. Changes were categorized as no change (0 to 29 percent), moderate change (30 to 49 percent) or high change (greater than 50 percent), thresholds grounded in the typical variability of vehicle controls and historical datasets with known seizurogenic drugs. Crucially, the team then compared in vitro effect concentrations with human or animal exposure data, asking whether the concentrations that perturbed neuronal networks in the dish resembled the concentrations actually circulating in patients or test animals.

The benchmark for validation was amoxapine, an antidepressant with a well-documented incidence of clinical seizures. In the MEA assay, amoxapine produced visible disruption of firing patterns at 1 micromolar, marked changes at 3 micromolar and intense changes at 10 micromolar, with burst frequency and network burst frequency rising in a clear concentration-dependent manner. Strikingly, these effects began within the human therapeutic plasma concentration range of 0.57 to 1.91 micromolar, providing direct evidence that the dish-based readout can align with clinically relevant exposures. The ion channel panel reinforced the picture: amoxapine showed high-potency inhibition of nine seizure-related targets, including NaV1.1, Kv1.1, Kv4.2, the GABA alpha1 beta2 gamma2 receptor and the nicotinic alpha2 beta4 receptor, with medium-potency hits at Kv7.2 and NMDA1/1.

The first case study applied the assay during lead optimization, the stage where medicinal chemists select which chemical series to advance. Five lead series, tested at 30 micromolar, produced sharply divergent profiles. Compound A showed no changes in any MEA parameter and was judged the safest lead to progress. Compounds B, C, D and E showed progressively concerning alterations in network burst frequency and inter-spike intervals, with compound E displaying the most severe perturbations. This kind of ranking, the authors argue, allows seizurogenic risk to be designed out of a molecule before animals, resources and years of development time are spent, offering a human-relevant alternative to zebrafish larval assays and rat hippocampal brain slices, both of which have struggled with throughput and translation to patients.

The second case study tackled a problem that had brought an entire project to a standstill: convulsions observed in dogs during the good laboratory practice toxicology studies required before first-in-human trials, with no seizures seen in rats tested in parallel. Metabolic analysis revealed a species difference. Humans and rats shared one predominant metabolite, while dogs produced a different metabolite that was only minor in the other species. When the parent drug, the human and rodent metabolite, and the dog-specific metabolite were run through the assay, the results were decisive. The parent compound and the human-relevant metabolite were essentially clean, while the dog metabolite produced a clear seizurogenic phenotype at 100 micromolar, disrupting burst frequency, spikes per burst and, most notably, the mean inter-spike interval. The dog metabolite also showed a high-potency hit at the NMDA 1/2A receptor, with an IC50 of 19.3 micromolar, whereas the other two compounds had no ion channel hits at all. The conclusion was that the canine convulsions stemmed from a dog-specific metabolite unlikely to pose a comparable risk to humans.

The third case study compared two clinical candidates, X and Y, to guide prioritization. Candidate Y displayed a potent seizurogenic signature at 3 and 10 micromolar, concentrations overlapping its clinical maximum plasma concentration of 5 micromolar, and at higher doses it completely disintegrated synchronous firing, rendering the raster plots unreadable. Candidate X, by contrast, was clean at its clinical Cmax of 1 micromolar, with only scattered changes appearing above 10 micromolar. The data clearly identified X as the lower-risk candidate. In a fourth case study, the team directly tested in vitro to in vivo translation using compound B, for which animal exposure data existed. The compound left MEA parameters untouched up to 10 micromolar, matching the absence of nervous system signs in animals at total plasma concentrations up to 38 micromolar, but produced marked changes at 100 micromolar, mirroring convulsions seen in vivo at plasma concentrations of 148 to 297 micromolar.

The authors are candid about limitations. The case studies come from client work, so the compounds are anonymous and the datasets limited. Changes in MEA parameters could in principle reflect generalized network perturbation rather than seizure liability specifically, although the researchers note that non-specific effects produce a different fingerprint and that their benchmark validation set represents an industry standard for seizure detection. A formal decision framework and refined exposure metrics, including unbound brain concentrations that are difficult to measure, remain works in progress. The work is therefore described as hypothesis-generating, a starting point for defining the context of use that regulators increasingly demand from new approach methodologies.

Nevertheless, the regulatory timing is favorable. A recent FDA Center for Drug Evaluation and Research review has emphasized the potential of new approach methodologies, and proposals to modernize the ICH S7A safety pharmacology guideline call for greater use of secondary pharmacology, integrated risk assessment and weight-of-evidence approaches. The authors propose that the MEA assay could be framed as a tool for assessing whether small molecule candidates alter human neuronal excitability, while the ion channel panel assesses perturbations of neuronal channel function, together supporting compound prioritization and first-time-in-human submissions. By analogy with the hERG cardiac channel assay, which transformed arrhythmia screening two decades ago, a potent ion channel IC50 would serve as an alert for further investigation rather than an automatic stop.

Looking ahead, the team is developing structure-activity relationship models to predict unwanted ion channel activity in silico, potentially allowing predicted metabolites to be evaluated computationally before any laboratory work begins. For projects where brain penetration can be avoided, that remains the simplest route to eliminating seizure risk; for central nervous system programs, where brain exposure is essential, the focus shifts to selectivity between the intended target and off-target channel effects. What the iLAseizure case studies demonstrate, above all, is that human cells in a dish can now speak with surprising authority about a risk that has historically only revealed itself in convulsing animals or, tragically, in patients.

Subject of Research: An in vitro seizure liability assay using human induced pluripotent stem cell neurons and ion channel panels to predict drug-induced seizure risk in drug discovery and development.

Article Title: Implementation of the In Vitro Seizure Liability Assay (iLAseizure) in Drug Discovery and Development: Mechanism of Action Case Studies

Article References: Rockley, K., Roberts, R., Jennings, H., Jones, K., Maizières, M.-A., & Morton, M. (2026). Implementation of the In Vitro Seizure Liability Assay ( iLA seizure ) in Drug Discovery and Development: Mechanism of Action Case Studies. Pharmacology Research & Perspectives, 14(5), Article e70317. https://doi.org/10.1002/prp2.70317

Image Credits: AI Generated

DOI: 10.1002/prp2.70317

Keywords: seizure liability, drug discovery, hiPSC neurons, microelectrode array, ion channels, new approach methodology, safety pharmacology, neurotoxicity, MEA assay, drug-induced seizures, ICH S7A, in vitro to in vivo translation