Drug-induced liver injury (DILI) is a main cause of drug development failures and market withdrawals. Despite the availability of a large panel of in vitro and animal models to screen for DILI risk, its detection remains problematic as it does not occur in each person. Indeed, idiosyncratic DILI (iDILI) is usually dose-independent and occurs in only a small fraction of subjects. Furthermore, there is evidence that most iDILI cases are immune-mediated, suggesting that its occurrence may also depend on the age and sex of the subject as these non-genetic factors influence the immune cell landscape (Uetrecht, 2019; Huang et al., 2021; Allison et al., 2023).

Troglitazone (TGZ) is an antidiabetic drug that was removed from the market because of severe cases of liver failure. During its development, no hepatotoxicity was detected. Nevertheless, once on the market, severe cases of liver injury leading to liver transplant or death were observed (Funk et al., 2001; Chojkier, 2005). This evidence suggests that iDILI is a difficult-to-detect event at early stages of drug development presumably due to the poor predictivity of available models. Indeed, existing in vitro models, including primary human hepatocytes (PHHs), iPSC-derived liver organoids, and animal models, failed to predict the potential of TGZ to induce severe liver injury in a larger patient population (Segovia-Zafra et al., 2021; Stern et al., 2023). The mechanism of liver injury caused by troglitazone remains unclear. However, data show that troglitazone upregulates cytochrome P450 CYP3A4 expression and is metabolized into reactive metabolites that cause hepatocyte damage and release of damage-associated molecular patterns (DAMPs), which then trigger macrophage inflammasome activation and liver injury (Sahi et al., 2000; Kato and Uetrecht, 2017; Mak et al., 2018). These results suggest that TGZ is an immune-mediated DILI compound. TGZ-mediated iDILI was not predicted accurately during drug discovery. Consequently, TGZ was removed from the market 3 years after its approval due to liver failure. It is, therefore, obvious that the lack of a good predictive model for iDILI risk constitutes a bottleneck for drug development.

We recently developed a subject-dependent educated spheroid system that can detect the DILI risk of difficult-to-detect molecules (Cherradi et al., 2023). Unlike the cell reprogramming technique, our cell educating technology does not modify the cell fate but induces a specific behavior in the cells owing to the composition of the serum supplemented in the cell culture. These cells are called educated cells throughout the study. In the present work, we used this technology to add educated autologous monocyte-derived macrophages and dendritic cells (DCs) into our hepatic spheroids and demonstrated that our system can detect TGZ-mediated iDILI at therapeutic doses. Moreover, we confirmed that our system can detect immune-mediated iDILI of diclofenac. Our easy-to-handle and quick-to-set-up educated spheroid system opens perspectives for high-throughput screening (HTS) of the iDILI risk of new compounds and for its use as a companion test to identify patients at risk for iDILI prior to the initiation of the treatment protocol.

Idiosyncratic DILI is a multifactorial event that relies on host factors, including age and sex, and intrinsic biological environments, making it difficult to detect during drug development. In humans, its unpredictable occurrence constitutes a drawback for FDA-approved drug use. Indeed, a large number of drugs were removed from the market because of iDILI, although they were considered safe during their development. Troglitazone belongs to one of these cases. We presented here the first easy-to-handle and quick-to-generate educated spheroid system that is capable of detecting the iDILI risk of TGZ and diclofenac in a large group of people, opening perspectives for a more reliable way to assess iDILI risk at early stages of drug development. Our system also paves the way for a clinically personalized approach to identify patients at risk for iDILI prior to treatments.

Studies that use existing preclinical models to analyze the DILI risk of a drug are generally performed at concentrations that are higher than one magnitude above the clinical concentrations (i.e., up to 100× C_max) (Proctor et al., 2017). Generally, if cell death occurs before 100× C_max, a drug is considered a DILI-risk compound. This binary analysis is applicable to conclude on a safety or a risk feature of a molecule, but it cannot address the issue of iDILI. More importantly, these elevated concentrations that are tested in vitro are out of the therapeutic window of the drug, making it impossible to predict how safe that molecule is when it is used to treat patients. Indeed, testing a drug at concentrations that are higher than one order of magnitude above clinically relevant concentrations is potentially problematic to predict whether this molecule is safe at therapeutic doses because the mechanisms of toxicity may differ between concentrations at C_max and at higher C_max. ASA, a non-iDILI partner of diclofenac (Granitzny et al., 2017), is believed to be safe. However, it has been shown in a case report of a 41-year-old woman that at a high dose, ASA induced liver injury (Laster and Satoskar, 2014). Our data showed that a 19-year-old female displayed cell death at concentrations above C_max, confirming that ASA could be toxic at higher doses. In healthy volunteers, troglitazone, given per os at therapeutic doses of 200–600 mg/day, reaches peak plasma drug concentrations of 0.9–2.8 mg/L approximately 2–3 h later (Plosker and Faulds, 1999), corresponding to 0.3× C_max and 1× C_max, respectively. Our data clearly demonstrated that TGZ induced an iDILI risk in 5 subjects out of a cohort of 72 within the therapeutic window. Unfortunately, we could not examine whether the five individuals that were identified with iDILI risk have true iDILI susceptibility as the study was conducted on healthy individuals and as TGZ was withdrawn by the FDA from the market in 2000. Nevertheless, a retrospective study showed that TGZ caused liver failure in 94 out of the 1.92 million treated patients, and less than 5% of them developed liver injury with elevated ALT levels (Graham et al., 2003). Our analysis of TGZ-mediated iDILI in a cohort of 72 healthy donors revealed that only 5 subjects out of the 72 (6.9%) showed toxicity at concentrations below C_max, confirming the proportion of iDILI detected in clinical trials. Interestingly, the US Government Accountability Office (GAO) published a letter on 19 January 2001 reporting that among the drugs that were withdrawn, TGZ had a greater liver injury risk for women. Our data showed that among the five patients with iDILI, three were females and two were males, confirming that females are at a higher risk for iDILI than males. To the best of our knowledge, the educated spheroid model is a unique preclinical model that is capable of detecting cell toxicity at therapeutic doses, suggesting that this system could be used as a companion test to identify a patient who could be at risk for iDILI occurrence. Our data encourage us to initiate a prospective study on patients treated with drugs other than TGZ, as TGZ was withdrawn in 2000, to confirm the accuracy of this personalized model with the aim of its use in a personalized approach for iDILI risk.

Presently, studies on immune cell functions were conducted with differentiated macrophages and DCs that were prepared under non-physiological conditions (Villar et al., 2023). This approach is not relevant as the differentiation of these cells in humans depends on host factors and environments that are not reproduced in vitro (Kenna and Uetrecht, 2018). The OneSmartDiff customizable medium offers the possibility to simultaneously differentiate monocytes into macrophages and DCs mimicking the immune cell populations that are present in clinical samples. Kawase et al. previously reported that NSAID-induced liver toxicity is dependent on the differentiation status of macrophages (Kawase et al., 2022). The OneSmartDiff medium could be prepared for each individual; it is, therefore, possible to generate simultaneously educated immune cell populations and educated hepatic cells to reproduce the biological conditions for each subject. Our results show that the number of monocyte-derived macrophages and DCs was higher in males than that in females, and it was also higher in young adults (18 years) than in older adults (53/52 years old), which are in line with previous works, where it was reported that the number of macrophages is significantly lower in adults than in children and young adults (Ogawa et al., 2000), as well as in females than in males (Chen et al., 2021). Moreover, our data confirmed that geriatric conditions brought from the customizable OneSmartDiff media that have been prepared from depleted sera from older adults somehow prevent macrophage differentiation (Mahbub et al., 2012). Finally, our data revealed that educated immune cells may trigger an inter-individual difference in the cytokinome landscape that could modulate iDILI risk occurrence. Further studies are needed to explore the link between the cytokine profile and iDILI risk. All together, we believe that the integration of all biological information from the same person within a spheroid makes it possible to assess iDILI risk in a cohort of subjects. The same strategy could also be used to generate macrophages and DCs under pathological conditions to analyze their functions in a manner that is closer to what occurs in patients.

It is well known that some host factors, including age and sex, could increase the risk of iDILI occurrence in a drug-specific way (Yamashita et al., 2017). Unfortunately, these parameters are, so far, under-considered in preclinical models. Attempts were made using animal models. However, data were not satisfactory to reliably anticipate iDILI risk in humans (Atkins et al., 2020). Our data clearly demonstrated that our system of educated spheroids, which is person-dependent, could address the issue and offers a way to analyze the iDILI risk in individuals for whom a medication is dedicated.

Primary human hepatocytes are considered a gold standard for DILI analysis. However, their limited availability, the requirement of invasive procedures to collect, and their quick dedifferentiation when cultured in vitro make them unsuitable for HTS. Alternatives were proposed with human hepatoma cell lines, such as HepG2 or HepaRG. HepaRG cells are closer to PHHs as they express greater levels of phase I and phase II drug-metabolizing enzymes than HepG2. Nevertheless, it has been reported that HepG2 spheroids could identify 58% of DILI-positive molecules, while HepaRG spheroids could identify only 47% of these compounds on a panel of 150 DILI drugs (Basharat et al., 2020). We are aware that the expression of drug-metabolizing enzymes is lower in HepG2 cells than in PHHs. Nevertheless, we previously published data showing that educated HepG2 spheroids have increased basal CYP activity, which was further enhanced upon exposure to drugs (Cherradi et al., 2023). Moreover, transcriptomic profiling of educated spheroids revealed that the expression of some phase I and phase II drug-metabolizing enzymes is comparable to the liver tissue (unpublished data). Our data suggest that HepG2 cells could be used as alternative cells to PHHs to assess iDILI in our system. The model could be strengthened further by adding endothelial cells as these cells are potential targets of DILI, and a larger panel of iDILI and non-iDILI drugs should be tested with the system. Experimentally, this is a massive amount of work as each drug must be tested on cohorts with a large sample size. Finally, further analysis of the metabolism, for instance, by measuring the induction of drug-metabolizing enzymes and their activities, should be done to better understand the mechanism of iDILI of these compounds.

In summary, we present here an educated spheroid system containing autologous educated macrophages and DCs that can detect the iDILI risk of TGZ and diclofenac at therapeutic concentrations. As the system does not require primary cells and is easy and quick to set up, it could be used for HTS of new molecules supporting drug-development pipelines of pharmaceutical companies.