Prolongation of the QT interval and the associated risk of Torsade de Pointes (TdP) is recognized as a significant cardiovascular risk in drug development. At the cellular level, the slowing of cardiac repolarization is usually a consequence of inhibition of the rapid component of the delayed rectifier potassium current (I_Kr). This current is conveyed by the human ether-à-go-go-related gene, also known as hERG, encoding the K_V11.1 potassium channel (Sanguinetti et al., 1995). Electrophysiological assessment of the hERG ion channel is now routine practice within drug development and uses recombinant cell lines engineered to express a functional human hERG channels. These transfected cell lines have progressively replaced former assays based on animal cells such as Xenopus oocytes, considered less sensitive.

The introduction of hERG screening hugely improved identification of compounds that delay ventricular repolarization to the point that the risk of approving drugs with the potential to induce TdP in patients was effectively eliminated (International Council for Harmonisation (ICH), 2005a, 2005b). However, the high sensitivity of this approach came with low specificity and drugs could be labelled as torsadogenic even if they actually posed no TdP risk. This was attributed to the result of focusing on only one ion channel governing ventricular repolarization (Simpson et al., 2025). Indeed, the cardiac action potential is governed by activity of many ion channels, and cardiotoxicity itself can be caused by other factors (e.g. structural damage). This led to the development of the CiPA initiative which aimed to develop a new paradigm for assessing proarrhythmic risk—utilizing new technologies and an expanded understanding of cardiotoxicity beyond hERG block. CiPA consists of three non-clinical pillars: i) assessment of drug effects on the critical human ventricular ion channel currents; ii) in silico integration of the ion channel effects to determine the net effects on the cardiac action potential; iii) a check for discrepancies in fully integrated biological systems (in vitro human stem cell-derived cardiomyocyte models) (Sager et al., 2014; Gintant et al., 2016). These non-clinical pillars are complemented by robust electrocardiogram (ECG) monitoring in Phase 1 clinical trials (Figure 2).

In 2022, the ICH E14/S7B guidance was revised through a Q&A process and now integrates the 3 CiPA pillars as optional ‘follow-up assays’ (International Council for Harmonisation (ICH), 2022). Despite the fact that CiPA shaped the ICH S7B Q&As, there are no publicly available statistics on real-world regulatory acceptance rates of CiPA assays or published cases where CiPA data were decisive in approving or rejected a drug. Such studies are rather considered as informative to complement core studies and contribute to the integrated non-clinical/clinical risk assessment and allow sponsors to possibly obtain a waiver for the clinical TQT study, under two possible scenarios.

A cardiovascular in vivo safety pharmacology evaluation in a relevant animal species is an essential component of the non-clinical safety assessment of new drug candidates and is still a mandatory core assay of the ICH S7B guidance. Although the primary focus of such study is on QTc prolongation, it also evaluates drug-induced functional effects on other ECG parameters (PR interval, QRS duration), as well as on hemodynamics (arterial blood pressure (BP), and heart rate), thereby fulfilling the ICH S7A requirements. Figure 4 shows the refinement over time of the in vivo methodologies used in safety pharmacology studies across the three vital organ systems. Before the advent of implantable telemetry devices for laboratory animals in the 1990’s, most cardiovascular studies relied on anesthetized, physically-restrained, acutely instrumented animals or employed invasive and stressful techniques in conscious animals, such as tethered systems. While anesthesia eliminates pain and suffering, all anesthetic agents can significantly alter cardiovascular parameters, potentially confounding the pharmacodynamic response to the test drug itself (e.g., alpha-chloralose-induced tachycardia and increase in pulmonary pressures, isoflurane-induced vasodilation, ketamine-induced tachycardia and hypertension, or barbiturates-induced cardio-depressant effects) (Nagayama et al., 2024). Similarly, stress associated with handling or physical restraint is known to activate the sympathetic nervous system, leading to elevated, non-physiological BP and heart rate values (Balcombe et al., 2004; Crestani, 2016). Moreover, restraint-based methods generally exhibit lower sensitivity and reproducibility compared to telemetry in freely moving animals (Henderson et al., 2025; Valentin et al., 2025a).

Over the years, substantial progress has been made in the miniaturization and refinement of telemetry devices. Today, cardiovascular telemetry can be performed even in small species such as mice and allows maintaining groups or at least pair-housing during data acquisition, aligning with animal welfare legislation and best practices (Skinner et al., 2019; Prior and Holbrook, 2021) (Figure 4). Providers have developed dual pressure telemetry implants that allow simultaneously measurement of high quality left ventricular pressure (LVP) and arterial BP from the same animal. This can be used in various species (rodents and non-rodents) for routine safety pharmacology (small molecules) and issue resolution (small and large molecules) studies. This technology allows to optimize animal use, reduce colony size, labor, housing and cost, and reduce compound needs. It is the best alternative to performing two separate telemetry studies (LVP & BP) for myocardial contractility and blood pressure assessment. The revised ICH E14/S7B Questions & Answers (Q&As) regulatory document (International Council for Harmonisation (ICH), 2022) considers the conduct of cardiovascular studies in telemetered non-rodents as the gold standard method for integrated QTc prolongation risk assessment supporting clinical TQT waiver requests.

Recent publications from industry and CRO consortiums (Rossman et al., 2023; Vargas et al., 2023) highlight the importance of defining optimal study practices to enhance the reliability of these studies and improve overall study sensitivity. Such good practices concerning the design, the execution and the analysis/reporting of the study, together with a translational approach (such as concentration-QT modeling), may improve confidence in nonclinical data for regulatory decision-making regarding the proarrhythmic liability assessment of new drugs. Best practices lead to better data, with less animals, and increasing their value for clinical QTc risk assessment.

The classical cross-over design applied in stand-alone cardiovascular telemetry studies exemplifies the application of the Reduction and Refinement principles in safety pharmacology. Indeed, instead of using a parallel dose design with typically 16 animals (e.g., 4 groups of n=4 animals: control and 3 dose levels), a Latin square design requires only 4 animals to achieve comparable or even better scientific outcomes. Taking into account that around 360 small molecule drugs were approved by FDA between 2010 and 2024, applying the cross-over design in the cardiovascular safety pharmacology study has saved more than 4000 non-rodents. This approach not only reduces the number of animals needed but also enhances statistical power and reproducibility, as each animal serves as its own control. Finally, animals can be reused across several studies, provided that an appropriate washout period is applied between treatments, no adverse effects have occurred, and all cardiovascular parameters have returned to baseline. A study performed with 6 reference drugs previously evaluated in a prospective clinical QT study (Darpo et al., 2015) demonstrated sufficient sensitivity with only 4 telemetered dogs in detecting minimal QTc changes (<10 msec), at concentrations close to human clinical (therapeutic) exposure (Darpo et al., 2015; Bétat et al., 2024).

In addition to the stand-alone safety pharmacology study, which evaluates the acute effects of a drug on the cardiovascular system, it is also useful to assess its impact after repeat administration. Indeed, drug-induced cardiovascular effects could be aggravated, remain similar, or disappear after chronic administration. For this purpose, ECGs are often collected during repeat-dose toxicity studies from chemically or physically-restrained animals over a short timeframe. However, manual restraint and the use of anesthesia can induce cardiovascular changes that confound interpretation of ECG measurements, and capturing only a few cardiac cycles provides a limited and potentially non-representative sample of cardiac function. These limitations can be mitigated by using a non-invasive jacket-based ECG system. Jacketed External Telemetry (JET) enables continuous measurement of ECG and heart rate without requiring surgery, reducing stress-related artefacts and improving data quality. It nevertheless needs an acclimation period for the animals, which is critical for the JET data quality and sensitivity to detect drug-induced changes in cardiac conduction and repolarization (Derakhchan et al., 2011a). JET-derived PR, QT, QTc intervals, QRS duration, and heart rate correlated well with those derived from PCT telemetry transmitter (Derakhchan et al., 2014). JET ECG is a great example for Refinement and a preferred method compared to restraint-based ECG because high-density ECG sampling can be collected in unstressed conscious large animals, in short- and long-term studies of 1–7 months duration (Vargas et al., 2010; Chui et al., 2011; Derakhchan et al., 2011b, 2011a). The JET BP add-on acquires accurate and continuous BP data from a minimally invasive implant, and JET RIP (Respiratory Inductive Plethysmography) can continuously monitor respiratory rate and depth with the jacket abdominal and rib cage bands.

In certain circumstances, it might be required to assess additional parameters outlined in the follow-up studies section of the ICH S7A guidance—such as cardiac output, myocardial contractility, and vascular resistance—for a deeper understanding and mechanistic investigation of drug-induced changes in these cardiovascular parameters. This may, therefore, necessitate additional studies. However, recent progress has focused on extracting these additional parameters directly from the cardiovascular telemetry study itself. This advancement—supported by in silico modeling—has the potential to replace some follow-up investigations, thereby further aligning with the 3R principles by both reducing the need for supplementary studies and refining the existing evaluation through the integration of advanced derived cardiovascular endpoints (Champeroux et al., 2024). Pulse contour analysis of arterial waveforms enables estimation of stroke volume, cardiac output, and systemic vascular resistance. Central aortic pressure can be approximated using the N-point moving average (NPMA) method for beat-to-beat modeling, improving accuracy and cross-species applicability (Champeroux et al., 2024). Generalized transfer functions have been developed for dogs to enhance waveform reconstruction (Hotek et al., 2023). Modeling pulsatile arterial hemodynamics and ventricular-arterial coupling has advanced further. Analyses using BP and synthesized aortic flow evaluate pulse wave propagation, wave reflection, arterial impedance, cardiac efficiency, and wave power. Synthesized flow waveforms have shown high concordance with measured flow, providing mechanistic insights into cardiovascular dynamics and drug effects. While promising, routine implementation remains challenging, and integration with AI/ML could enhance usability (Hotek et al., 2023). High-Frequency Autonomic Modulation (HFAM) evaluates autonomic control of heart rate through high-frequency oscillations. The HFAM index differentiates parasympathetic dominance (S1), co-activation of sympathetic and parasympathetic systems (S2), and sympathetic predominance (S3). This approach detects compensatory autonomic responses that may be missed by standard BP analyses (Champéroux et al., 2018, 2022).

Overall, refinements in telemetry technologies, combined with in silico modeling, advanced hemodynamic analyses, and HFAM, expand cardiovascular endpoints, enhance mechanistic insights, and reduce animal use. While full replacement of in vivo models is currently limited, these approaches improve the translational value of preclinical studies to the clinical setting.

In contrast with the cardiovascular safety assessment of pharmaceuticals which is regulated through two dedicated guidelines, there is no or little regulatory framework on CNS non-clinical safety assessment, except for drug abuse liability. Although CNS is one of the three vital systems that form part of the core battery of safety pharmacology studies, the ICH S7A guidance provides sparse information to drug developers with regard to models or studies to focus on. Typically, the most common approach to assess potential CNS side effects is the historical Irwin test (Irwin, 1968), adapted as ‘functional observational battery’ (FOB) for safety assessment purposes (Baird et al., 1997; Redfern et al., 2005; Gauvin and Baird, 2008; Himmel, 2008). This neurobehavioral test is usually conducted in rodents but has also been adapted to larger species, and aims to identify a wide range of autonomic, neuromuscular, sensorimotor, and behavioral effects after administration of a test compound.

The Irwin/FOB relies on highly experienced operators and should be performed blinded to ensure that all endpoints are detected in an objective manner. However, it is still possible that some CNS-related signs may be overlooked, especially if they are sporadic and subtle. The assessment is generally conducted in a sequential order, with the animal first observed in its home cage, then placed in a new environment to assess its exploratory behavior and finally handled by the operator to evaluate some reflexes and record some parameters such as body temperature. These observations can be repeated at different timepoints after drug administration. Despite its inherent subjectivity and interlaboratory variability, and its poor predictive value for some side effects such as headache, dizziness or somnolence/fatigue (Mead et al., 2016), the FOB remains an interesting safety pharmacology tool to characterize the neurobehavioral profile of a lead compound or drug candidate, in particular at early stage of a project (Redfern et al., 2019).

The recent progress in AI/ML has triggered various initiatives to refine neurobehavioral assessments by recording behavioral and motor activity parameters directly from the home cage in a continuous manner (24/7). Data are analyzed by various algorithms able to detect different behaviors (supported rearing, grooming, tremors, stereotypies, sleeping, walking, …) (Chen et al., 2023; Chindemi et al., 2025). This avoids the repeated stress of the animals due to handling and allows to detect potential CNS-related signs occurring during the night. A great example of these ‘smart cages’ is the newly developed automated VivaMARS activity system rack, with each rack having 10 sensor boxes, and each sensor box consists of over 180 infrared beams, for up to 30 animals per session. These systems can be used in the setting of general toxicology studies, or eventually in pharmacology studies, avoiding the need to conduct stand-alone Irwin studies, thereby contributing to the Reduction and Refinement principles (Beck et al., 2025). The only challenges are the associated cost of such sophisticated equipment and the huge amount of generated data that need to be recorded, stored, and analyzed.

On the top of the classical, standard Irwin/FOB that is part of the core battery of safety pharmacology studies and already covers a wide range of potential CNS side effects, some pharmaceutical companies may decide to develop their own CNS safety derisking strategy to focus on side effects that could constitute a stopper in a project such as seizures, or for a particular patient population (e.g., hypertension in diabetic patients, hypotension in Parkinsonian patients,…).

Seizure is a particularly serious side effect defined as uncontrolled electrical activity in the brain. Seizures may not necessarily lead to ‘clinical’ convulsion (Pitkänen et al., 2006) and can only be detected using electroencephalogram (EEG) recordings, meaning they will be missed using the workflow that is reliant on observed behavioral alterations. This allows a significant loophole as approximately 75% of clinical compounds with seizure did not show convulsions in nonclinical safety studies in a retrospective Japanese study (Nagayama, 2015). As proposed by Easter et al. (2009) (Easter et al., 2009), the use of a stepwise cascade of assays is a useful approach to derisk seizure liability from early stages of discovery projects. This enables the early elimination of seizurogenic compounds based on in silico and/or in vitro results, thereby avoiding animal testing and preventing unnecessary harm.

For some CNS liabilities such as suicidality or drug abuse potential, it is generally considered that there is little or no alternative to in vivo models. Suicide or depression is a complex behavior which is impossible to fully reproduce in animals. Some behavioral endpoints in some animal species have however been proposed as predictive markers of depression, such as anhedonia, aggressivity, and helplessness in rodents, or lack of peer-rearing in primates. They are considered as refined approaches compared to historical ‘despair’ tests like the forced swim or tail suspension tests, which are highly controversial due to the perceived severity for the animals (Sewell et al., 2021; Mohseni and Rafaiee, 2025). However, these behavioral traits are still associated with significant challenges and drawbacks (Planchez et al., 2019). Some novel approaches have emerged such as proteomics or micro-RNAs but remain anecdotical at this stage (Wang and Dwivedi, 2025).

From a regulatory perspective, abuse liability studies are expected for any new CNS-active drug and are generally conducted before starting Phase 3 clinical trials. Abuse liability studies in animals are very heavy, expensive, ethically questionable, and time-consuming. However, both EMA and FDA agencies infer that the decision on abuse liability assessment is dependent on the level of risk or concern that is determined case-by-case for each drug. Therefore, it is possible to waive the animal studies (Derakhchan et al., 2023), after evaluation of the dossier by the regulatory agencies and based on the results of the ‘eight-factor analysis’. This includes for example in silico assessment (chemical structure similarity with known drugs of abuse), in vitro pharmacology profiling data (affinity/activity on targets involved in abuse liability), PK data (brain penetration, rapid absorption, fast onset), Irwin/FOB data (e.g., head twitches in mice), general toxicology data (withdrawal signs during recovery phase of repeat dose toxicity studies) (Heal et al., 2025).

Currently, in vitro assays for the detection of neurotoxicity are not widely used prior to performing animal studies in drug development. The neurite outgrowth assay that can be applied to rodent primary neurons or hiPSC-derived neurons is a useful tool to assess neuronal cell viability. Various endpoints can be measured, such as neurite length, cell body area, branch points and number of cell clusters (Bagheri et al., 2020; Zhang et al., 2022). This allows for a dynamic monitoring with automated image analysis.

But the next generation of in vitro assays in the neuroscience field is the human brain organoids (‘microbrain’ or ‘brain-on-chip’), which are 3D structures derived from hiPSCs reflecting human embryonic brain organization. They contain different cell types interacting together (neurons, glia) and have the advantage over animal-based models to better represent human brain function (Kim and Chang, 2023). Although their primary applications are for modelling neurodevelopmental or neurodegenerative disorders, they could also be used for assessing drug effects and hazard identification (Fan et al., 2022), or for personalized medicine if based on patient-derived cells (Giorgi et al., 2024). However, a recent EMA review of organ-on-chips used for medicines safety assessment showed that, while liver, kidney and heart are the most frequently modeled systems, CNS is less commonly used, both as single or multi-organ systems (Portela et al., 2026). This may be due to the current limitations and translational challenges of these models, namely their high variability and limited reproducibility, their structural and functional immaturity, and the lack of vascular and immune components (Park and Sun, 2025). Engineering strategies try to overcome these challenges (Fan et al., 2025).

In contrast with in silico proarrhythmia models that have been extensively developed over the past 10–12 years due to the CiPA paradigm and the integration of such models in the revised ICH S7B guidance, in silico tools to predict CNS side effects are less common and most efforts have focused on blood-brain barrier simulation models. In the framework of the former IMI2-NeuroDeRisk consortium mentioned earlier, a toolbox containing around 60 different in silico models based on various neurotoxicophores associated with seizure liability, suicidality, and peripheral neuropathies was established (Bryant et al., 2022). Some additional models linked to blood-brain barrier (for OATP2 and HCO antiporter) were also added. Any new chemical structure can be uploaded into the toolbox and tested against all models or a pre-selection of it. A heatmap is generated, which flags structure similarities with one or several of the toxicophores.

Traditional respiratory assessments in rats—such as whole-body plethysmography (WBP) and head-out plethysmography—are limited by confinement, stress induction, and short recording durations, which can confound data interpretation (Coggins et al., 1981; Adler et al., 2004; Bohannon et al., 2016; Fares et al., 2019). To overcome these limitations, the DECRO system was developed: a miniaturized telemetry platform enabling continuous, non-invasive recording of respiration, ECG, and activity in freely moving, socially housed rats (Flenet et al., 2020; Fares et al., 2022). This innovation allows 24-hour and repeated recordings under home-cage conditions, capturing expected pharmacological effects (e.g., baclofen-induced respiratory depression) with high sensitivity and reproducibility, while also promoting Reduction and Refinement (Fares et al., 2022). Importantly, the system enables simultaneous respiratory and CNS functional assessments (Irwin test/FOB), demonstrated with baclofen, caffeine, and clonidine. This integration eliminates the need for separate studies, achieving a 100% reduction in animal use for dual endpoints (Fares and Champéroux, 2023). The inclusion of heart rate and locomotor activity further enhances interpretation of CNS, respiratory, and autonomic effects, and offers integrated insights into how drugs impact all these vital systems concurrently. Jacketed telemetry in rats offers a scientifically robust, ethically superior, and regulatory-compliant approach for modern safety pharmacology.

A few advancing alternatives to in vivo respiratory studies are described hereafter. Figure 5 illustrates that complex in vitro systems such as lung organoids are generally associated with higher costs compared to simple 2D cell models but have higher physiological relevance.

Traditional 2D monolayers in vitro models are useful for early screening, and the submerged cultures of the ciliated cells are used to check for structural and functional abnormalities. Nowadays there are numerous available cell lines with standardized and well-defined characteristics. Klein et al. (2013) published a list of the common lung and endothelial cell lines used for mono- and co-culture in vitro lung systems (Klein et al., 2013). As an alternative cell source, there are human pluripotent stems cells (hPSCs), which include embryonic stem cells (ESCs) and induced pluripotent stem cells (iPSCs). These 2D models are practical and inexpensive; however, they are only useful for yes/no approaches and do not allow for the details of response that have been reported as observed in 3D model systems, and pertinent to what would occur in vivo.

The Air-Liquid Interface (ALI) system was introduced by Witcuttj et al. in 1988 (Whitcutt et al., 1988). These ALI cultures offer a physiologically relevant in vitro model for safety pharmacology, particularly for inhalation toxicity studies, by allowing cells to differentiate (cilia, mucus, surfactant) with improved cell morphology and function, and by exposing them directly to aerosols/gases. Key advantages include: 1) a high physiological relevance by often using primary human cells, precision cut lung slices (PCLS), or human-derived bronchial epithelial cells (HBECs) for better mimicking the human lung than submerged cultures; 2) Direct exposure and accurate dosing, as the test articles can be applied directly to the apical surface, simulating in vivo inhalation exposure to aerosols, nanoparticles, and gases; 3) act as an intermediate step between standard in vitro assays and in vivo studies,. While disadvantages of these ALI cultures include high technical complexity as these require specialized instrumentation, high cost, high variability, labor-intensive, limited contribution of immune system, no circulation, low-throughput, and requiring 3–4 weeks for differentiation.

The precision cut lung slices (PCLS) are an ex vivo method, offering a 3D model that preserves native lung architecture and cell-type diversity (epithelium, smooth muscle, macrophages), using tissue already obtained and reliably stored. These PCLS can be obtained from either murine tissue or human tissue (healthy or diseased), reducing animal use while enhancing translational accuracy, and have been known to constrict when exposed to various stimuli, but their responses reduce over time (Neuhaus et al., 2017). These sections however only give a “snapshot” of what the lung looks like at the exact moment of fixation (Liu et al., 2019), and lack intact blood flow, circulating immune cells, and neural innervation. This would therefore only give an indication of what a potential “whole system” response may be and thus can be considered as a scientific ‘middle ground’ between established in vitro models and in vivo models.

Lung organoids are derived by self-organized primary lineages, such as ESC, iPSC, or organ stem/progenitor derivatives that develop into complex airway or alveolar tissues (Lancaster and Knoblich, 2014), offering a more integrated model and bridging the gap between traditional 2D cell cultures and animal models, particularly for identifying tissue-specific safety assessments. The limited number of cells needed to create an organoid further makes them potentially favorable, due to shortage of established primary cell lines. However, they are often used as complementary tools rather than a complete replacement for in vivo studies due to the lack of systemic, vascular, and immune components.

Lung-on-a-chip (LoC) technology was developed by the Wyss Institute founding director, Dr. Don Ingber at Harvard University (Huh et al., 2012). In 2013, he received the NC3Rs 3Rs Prize for his innovative LoC, a microdevice lined by human cells that recapitulates complex functions of the living lung. Indeed, this micro-engineered and microfluidic device is closely mimicking human lung tissue architecture, featuring alveolar-capillary barriers with epithelial and endothelial cells. This can simulate the physical forces of breathing (stretching), fluid flow (shear stress), and air exposure, enabling dynamic assessment of inhaled substances. It is a better replicate structure and function of the human lungs, including fluidics at a much smaller scale. These chips can be customized using patient-derived cells (e.g., in cystic fibrosis models) to test drug responses for specific individuals. While LoC technologies are revolutionizing safety pharmacology with their human-like and dynamic environments, they face limitations in scalability and cost that need to be addressed for broader adoption.

In summary, respiratory in vitro platforms are generally used for mechanistic understanding, but more rarely for drug safety screening and for clinical decision−making. Moreover, quantitative metrics on sensitivity/specificity of these assays are rarely (if ever) reported.

Recent developments include 2D and 3D cell culture systems, organoids, and organ-on-chip technologies. Organoids, derived from intestinal stem cells, self-organize into multicellular structures mimicking villus and crypt domains, enabling the study of epithelial responses to drugs. These models can be tailored to represent healthy or diseased states, enhancing their translational relevance (Peters et al., 2020).

Microphysiological systems (MPS), such as gut-on-a-chip platforms, incorporate fluid flow and mechanical strain to simulate peristalsis and nutrient transport. These systems support co-culture of epithelial, immune, and microbial cells, allowing for integrated assessment of drug-induced GI toxicity, including inflammation, barrier disruption, and microbiome interactions (Neto et al., 2025).

In vitro digestive simulation models also play a role in evaluating drug absorption and metabolism. These range from static single-compartment setups to dynamic multi-compartment systems that mimic physiological digestion processes, offering reproducible and ethically sound alternatives to animal testing (Li et al., 2024).

Within safety pharmacology, in vitro GI models are primarily used to identify and rank compounds with GI liability, with diarrheagenic risk emerging as a key application (Peters et al., 2020; Debad et al., 2025). Human 3D GI microtissue assays that measure transepithelial electrical resistance (TEER) as a readout of tight junction integrity can discriminate drugs that cause clinical diarrhea from those that do not, including cases where short-term animal studies failed to predict risk. TEER and related barrier endpoints therefore serve as important mechanistic surrogates for epithelial injury (Peters et al., 2019).

High-throughput 2D assays using primary human intestinal stem cell–derived monolayers extend this approach by integrating TEER with cell viability (e.g., nuclear counts) and proliferation (e.g., EdU incorporation) across multiple donors. These multi-donor designs enhance population-level relevance and improve prediction accuracy for clinical diarrhea (Pike et al., 2025). Gut-on-chip platforms and organoid–immune co-cultures provide additional mechanistic depth, allowing assessment of cytokine release, immune cell recruitment, and microbiota-modulated injury—features especially relevant for inflammation-driven GI toxicity (Ashammakhi et al., 2020; Macedo et al., 2023; Ofori‐Kwafo et al., 2025). Collectively, these in vitro systems support compound triage, medicinal chemistry optimization, and dose/regimen refinement by generating human-relevant GI safety data early in development, while reducing reliance on exploratory animal studies.

Advanced primary or organoid-derived intestinal monolayers, often combined with region-specific media and transporter/enzyme profiling, are increasingly used to refine segmental absorption estimates, assess drug–drug interactions at intestinal transporters, and characterize metabolism by CYPs and other enzymes (Fedi et al., 2021; Debad et al., 2025). Static and dynamic in vitro digestion models simulate dissolution, solubilization, and precipitation of oral formulations under physiologic pH and bile conditions, providing inputs for absorption modeling and supporting biopharmaceutics risk assessments, particularly for poorly soluble compounds (Xavier et al., 2023). Integration with endothelial cells, hepatocytes, and biliary-like cells in multi-organ MPS enables evaluation of sequential intestinal absorption, hepatic metabolism, and biliary excretion, capturing first-pass and enterohepatic processes in a more physiologically connected manner. Collectively, these DMPK-oriented in vitro methods complement GI safety-focused assays by describing how drugs and formulations behave along the GI tract, thereby strengthening mechanistic understanding of exposure–response relationships and informing dose selection, formulation design, and clinical study planning.

Building on these advances, several opportunities could increase the regulatory relevance of in vitro GI models as complements—or, in some contexts, alternatives—to the in vivo endpoints described in ICH S7A. Expanding beyond the current focus on small intestine and colon toward robust, standardized models of the stomach, duodenum, jejunum, and ileum would improve coverage of region-specific physiology, including differential pH, enzyme expression, bile exposure, and microbial load.

Currently, in vitro efforts emphasize GI injury and diarrheagenic risk, increasingly addressed through TEER-based barrier assays and cytotoxicity measures. Future work could focus on defining and qualifying formal surrogate endpoints for the broader set of S7A parameters, including gastric and bile secretion, transit time, and ileal contraction. Dynamic digestion models that approximate gastric secretion and pH, coupled to intestinal epithelia, could be further developed and standardized as functional analogues of classical in vivo secretory measurements. Similarly, MPS platforms that introduce cyclic strain and fluid flow to emulate peristalsis could be refined to better capture neuromuscular activity and motility patterns, enabling more direct mechanistic assessment of ileal contraction and transit.

As the field matures through systematic benchmarking of in vitro endpoints against clinical data and in vivo S7A-type measures these models have the potential to more directly address guideline expectations, support risk-based interpretation of GI findings, and ultimately reduce reliance on dedicated animal studies in GI safety pharmacology.

In parallel to the use of these in vitro systems and in alignment with the Refinement principle, minimally-invasive systems have been developed to measure gastrointestinal motility and other parameters such as gastric pH or intra-abdominal pressure. The SmartPill™ is a multisensory ingestible capsule well-known in clinical setting to monitor transit time, pH, pressure and temperature. It has also been successively used in minipigs (Knibbe et al., 2023) and dogs both for veterinary medicine purposes (Warrit et al., 2017) or preclinical research (Koziolek et al., 2019). Another example of wireless system is the Bravo™ pill used for esophageal pH test in patients. It has also been applied in various animal species (rats, dogs, monkeys) (Rudholm et al., 2008; Kook et al., 2014) for ambulatory esophageal pH monitoring or to assess the influence of pH on drug absorption and pharmacokinetics. A recent technology developed in rats is an ingestible capsule for non-invasive optical gut stimulation (ICOPS) (Elsherif et al., 2025). This small capsule of less than 2 cm length and 0.22 g weight is equipped with micro-LED lights and electronics for wireless powering based on the inductive coupling and is safely excreted in feces within 20h. It allows optogenetic stimulation of the enteric nervous system to better understand the gut-brain axis. However, such sophisticated technique has never been applied in safety pharmacology so far.

When conducted, in most cases, in vivo renal safety pharmacology studies mostly include the measurement of the glomerular filtration rate (GFR) in rodents, which requires the use of a tracer to determine either its excretion rate in urine or its elimination kinetics from plasma. Urine sampling generally requires singly housing the animals in metabolic cages, while blood sampling necessitates handling and restraint, which are stressful (Passini et al., 2025).

Refinement approaches may include automated blood sampling techniques, which allow blood collection from freely moving rodents even during night period (Han et al., 2013), or microsampling, which reduces the volume of blood required to just a few microliters. Such systems can of course be used not only for renal function assessment but for any studies requiring repeated blood sampling (PK, pharmacology, etc…) (Hopper, 2016; Prior et al., 2025). Another example of refinement is the use of hydrophobic sand for urine collection in rodents. This method proved to be efficient in maintaining volume and integrity of urine samples, while decreasing stress related to single housing in metabolic cages and the risk of injury (Hoffman et al., 2018) and can be applied for urine collection in all toxicology studies. An innovative 3D-printed two-compartmental device was also designed to allow urine collection from two home-caged mice keeping social interactions (Douté et al., 2024).

Over the last decade, significant progress has been made in the development of human-relevant kidney models that may reduce the need for animal testing. Two-dimensional renal proximal tubule epithelial cell cultures (RPTEC) enable high-throughput screening of nephrotoxicity by measuring mechanistic endpoints such as viability, oxidative stress, and specific injury biomarkers including KIM-1 and NGAL. However, these cultures lack complex architecture and transporter expression, limiting physiological relevance. To address these gaps, 3D spheroids and organoids derived from either human proximal tubule cells (Sakolish et al., 2023; Arakawa et al., 2024), urine-derived stem cells (Guo et al., 2020; Yu et al., 2023), or pluripotent stem cells have been developed, demonstrating improved expression of drug transporters (e.g., OAT1) and enhanced sensitivity to nephrotoxic compounds like cisplatin and amphotericin B, when assessed via ATP depletion, KIM-1 measurement, CYP induction, and structural injury markers. Furthermore, a ‘kidney-on-a-chip’ technology was developed, which incorporates perfusable architecture, co-cultures (e.g., RPTECs and endothelial cells), and functional readouts such as glucose reabsorption and TEER (Huang et al., 2024; Guimaraes et al., 2025). These platforms improve transporter activity, injury biomarker detection, and organ-level responses compared to static 2D cultures. This tiered strategy—from 2D assays to 3D and microfluidic systems—enables more human-relevant, mechanistic assessment of nephrotoxicity while reducing reliance on animal models. However, significant challenges remain, including immaturity, poor vascularization, or absence of urinary excretion, before being able to routinely use such in vitro systems for safety testing (Xi and Song, 2025).