Control Compounds for Preclinical Drug-Induced Liver Injury Assessment: Consensus-driven systematic review by the ProEuroDILI Network

doi:10.21203/rs.3.rs-3578817/v1

Download PDF

Systematic Review

Control Compounds for Preclinical Drug-Induced Liver Injury Assessment: Consensus-driven systematic review by the ProEuroDILI Network

https://doi.org/10.21203/rs.3.rs-3578817/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 30 Apr, 2024

Read the published version in Journal of Hepatology →

Version 1

posted

You are reading this latest preprint version

Idiosyncratic drug-induced liver injury (DILI) is a complex and unpredictable event caused by different drugs, herbal, and dietary supplements. The early identification of human hepatotoxicity at the preclinical stages remains a major challenge, in which the selection of validated in vitro systems and test drugs has a significant impact. This systematic review aims to analyse the compounds used in hepatotoxicity assays and establish a unified list of DILI positive and negative control drugs for validation of in vitro models of DILI, supported by literature and clinical evidence and endorsed by a committee of experts from the COST Action ProEuroDILI Network (CA 17112).

This systematic review was performed in accordance with the 2020 PRISMA guidelines. Original research articles focused on investigating DILI occurrence using in vitro human models performing at least one hepatotoxicity assay with positive and negative control compounds were included. A modified version of the "Toxicological Data Reliability Assessment Tool" (ToxRTool) was used to assess the bias of the included studies.

A total of 2,936 studies were retrieved from the different databases. Of these, 51 met the inclusion criteria, with 30 categorized as reliable without restrictions. Diclofenac and buspirone were the most commonly used DILI-positive and DILI-negative control drugs, respectively. Although there was a broad consensus on the positive compounds, the selection of negative control compounds remained less clear. Regarding the models used, the 2D monoculture of primary human hepatocytes (PHHs) was the favoured choice. However, there was no clear consensus on the drug concentrations. Short acute exposure times were mostly utilised, and cytotoxicity was the preferred endpoint.

The extensive analysis of included studies highlighted the lack of agreement on appropriate control compounds for the in vitro assessment of DILI. Therefore, following comprehensive in vitro and clinical data analysis together with input from the expert committee, an evidence-based consensus-driven list of 10 positive and negative drugs is proposed for validating new in vitro models for improving preclinical drug safety testing regimes.

Clinical Pharmacology

Gastroenterology & Hepatology

Toxicology

Drug-induced liver injury (DILI) encompasses liver damage caused by conventional or over-the-counter medicines together with herbal and dietary supplements. DILI can be classified as intrinsic, which is dose-dependent, predictable, and manifests quickly after drug exposure; or idiosyncratic, which is complex, unexpected, and dependent on host and environmental factors¹. Moreover, the mechanisms of liver injury can be divided into at least five main categories: reactive metabolites, mitochondrial dysfunction, transporter inhibition, lysosomal impairment, and immune-mediated toxicity^2,3. DILI constitutes one of the leading causes of drug attrition in clinical trials, use restriction, or withdrawal from the market^4–6.

Failure to predict hepatotoxicity in the drug development process is mainly due to the lack of human-relevant preclinical in vitro models as well as interspecies differences with animal models, resulting in poor preclinical to clinical translation. This is compounded by the multifactorial nature of DILI pathophysiology^7–9. The development of more sophisticated human predictive in vitro models, and technologies including in silico approaches has thus become a priority in pharma and basic research to address hepatotoxicity risk, both in an accurate and accelerated fashion in the drug development process¹⁰.

Predictive in vitro models for hepatotoxicity assessment must present relevance not only at a physiological level but also at pharmacological and pathological levels. A tiered approach¹¹ contemplating not only the in vitro human models selected but also their phenotypic characterization as well as pharmacological and toxicological functionality, is needed to validate new testing systems¹². Here, the lack of consensus about the selection of i) the in vitro human liver model, ii) the critical endpoints to analyse, iii) the number and type of control compounds to test, the concentration, and the time of exposure contribute significantly towards the observed heterogeneity and lack of reproducibility in results obtained in different studies^9,12. Several studies have also highlighted how critical a standardised set of prototypic hepatotoxic compounds with diverse toxicity mechanisms is to validate proof-of-concept studies¹³; whereas different pharmaceutical companies have demonstrated interest in applying, for example, standardised microphysiological systems for drug risk assessment¹⁴.

When developing a new in vitro liver model, the study requirements determine its characteristics, in which the human cell type and tissue architecture are important initial considerations. Hepatic cell types include cell lines (HepG2 or HepaRG), primary cells, (PHHs, non-parenchymal liver cells), as well as stem cell-derived hepatocyte-like cells (HLCs). Cell choice depends on factors such as availability (primary cells), but also the strengths and limitations of each, such as metabolic competence, genetic stability, and inter-individual variability. Tissue architecture is also a critical feature of the liver microenvironment, being conventional two-dimensional (2D) ‘flat’ monolayer cultures the least representative, whilst more complex 3D configurations such as spheroids or organoid cultures and microfluidic platforms closer mimic the in vivo organ. Moreover, the shift from single-cell systems to multicellular systems further enhances the relevance of the model, mimicking the in vivo multicellular interactions^11,15. At a later stage, to assess the relevance of the chosen model for the study requirements, different endpoints can be applied, ranging from measurements of cell death up to more functional and mechanistic pre-cell death endpoints, as DILI involves multiple mechanisms^16,17.

As for the prototypic hepatotoxic compounds selection, learning from drugs that either failed or were approved is an asset to test the expected hepatic response and the known mechanisms of DILI in a new in vitro system. Herein, we conducted a systematic review to summarise both study design and control compounds used in studies utilising predictive in vitro human DILI models. A deeper analysis on the drugs mostly used as positive and negative controls for DILI in the literature was also performed to develop a unified list of DILI-positive and DILI-negative control drugs. These findings were supported by a stringent literature-based review, validated with clinical evidence and endorsed by a consensus committee of experts from the ProEuroDILI Network (CA 17112).

Study Design and Search Strategy

This systematic review was conducted and reported following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines¹⁸. The protocol for the systematic review was registered in the Open Science Framework (OSF) Registries platform (osf.io/yp7g6). There were no deviations from the registered protocol.

Eligible literature published up to June 1st, 2022, was identified through a search in PubMed, Embase, Web of Science (WoS), and Scopus, with no language restrictions. The search strategy was designed based on identifying three terms: DILI, in vitro models, and predictivity. According to this strategy, the search comprised the following terms and Boolean operators: “DRUG*” AND (“LIVER INJURY” OR “HEPATOTOX*”) AND (“PREDICT” OR “IN VITRO” OR “TEST*”) AND (“SPECIFICITY” AND “SENSITIVITY”). To retrieve additional studies eligible for inclusion, references cited by the included studies, narrative or systematic reviews, and meta-analyses identified throughout the literature search were manually reviewed. The retrieved literature was managed using the Rayyan online tool¹⁹.

Inclusion and exclusion criteria

Published studies that fulfilled the following criteria were included:

To be a peer-reviewed original article.

To study the onset of DILI in preclinical stages.

To report at least one hepatotoxicity assay using in vitro human models, aiming to classify at least one drug in each of the following categories: DILI concern (DILI+) or no DILI concern (DILI-).

To report data about the model's predictive power, with sensitivity and specificity values (either quantitative or qualitative).

Studies conducted using in vivo models, and reviews, editorials, letters, commentaries, conference abstracts, and other reports with no relevant data were excluded. If the full text could not be accessed, it was searched via inter-library loan or the corresponding authors were contacted to request a copy. If the study could not be retrieved, it was finally excluded.

Study selection

The literature search was conducted by four independent researchers (ASZ, GMC, ABG, and DDS), who screened the title and abstract and retrieved and reviewed the full text of the relevant studies identified. Any disagreements were solved by discussion, whilst a 5th independent researcher (MVP) was consulted if a consensus needed to be reached.

Data extraction

Five researchers (ASZ, MVP, GMC, ABG, and DDS) independently extracted data from each article, and discrepancies were resolved by discussion until a consensus was reached. The following data were extracted from each of the included studies: full name of the first author, year of publication, the model(s) used to perform the hepatotoxicity assay, drugs tested and their DILI categorization, drug concentration(s), toxicity endpoint(s) measured, and specificity and sensitivity values. For the analysis, all the drugs used were classified in the categories DILI + and DILI-. To do so, in the studies where a binary categorization was not used, only the negative controls used for the hepatotoxicity assays and the model predictivity estimation were indexed as DILI-. The positive controls were classified as DILI + regardless of their severity category. Corresponding authors were contacted to obtain further information if required.

Quality assessment

Based on the software-based Toxicological data Reliability assessment Tool (ToxRTool)²⁰ and following published recommendations²¹, a refined tool named “Modified ToxRTool” was generated to assess the risk of bias of the included articles. This tool provides comprehensive criteria for determining the reliability of toxicological studies. Studies were evaluated in five domains: i) “Test substance identification”; ii) “Test system characterization”; iii) “Study design description”; iv) “Study results documentation”; and v) “Plausibility of study design and data”. Each domain item was scored with 0, 0.5, or 1 point, following the recommendations of Segal D. et al.²¹. After the evaluation, categories of reliability proposed by Klimisch et al.²², i.e., code 1 (reliable without restrictions), code 2 (reliable with restrictions), code 3 (not reliable), and code 4 (not assignable), were assigned to each domain (see also Figs. 1 and 2). Ten researchers (ASZ, MVP, ASS, GMC, ABG, DDS, IAA, HN, IM, and SG) independently conducted the quality assessments. Three independent researchers evaluated each study. The mean score for each category was calculated based on the assessment of the three researchers. Studies with a score of 15–18 points were classified as reliable without restrictions, being useful for the analysis. Studies with a score of 11–14 points were classified as reliable with restrictions, being potentially useful for the analysis. Studies with a score of < 11 points were classified as not reliable and were not considered further.

Data analysis

Data on the different models, drugs, conditions used to perform the toxicity assays, and sensitivity and specificity values were analysed. Other aspects of the toxicological assays were scored, such as the number of times each model and specific spatial configurations (2D, 3D) were used, how long the models were exposed to the different drugs, or which concentrations were tested. However, the heterogeneity in the study design concerning cell types, model configurations, and the number of DILI + and DILI- compounds analysed, along with the wide range in exposure times and concentrations, prevented conducting a meta-analysis.

A list showing all drugs and the number of articles where they appeared as positive or negative controls was created (Fig. 6).

Drug analysis

The drugs most commonly used as positive and negative controls in the literature were selected for examination. At least one hundred drugs from each category, DILI + and DILI-, were extensively analysed (Supplementary document 2). If there were additional drugs with the same number of occurrences in the articles beyond the hundredth drug for positive and negative controls, such drugs were included in the analysis, expanding the list as necessary. The classification of the drugs based on the pharmacological group was extracted from the ATC/DDD index 2023²³.

Clinical use cases

For the number of clinical cases reported per each drug, four different databases were analyzed: the Spanish DILI Registry²⁴ with 980 cases, the Pro-Euro-DILI Registry²⁵ with 246 cases, the DILI Network (DILIN)²⁶ with 899 cases and the LATINDILI Network²⁷ with 480 cases. The DrugBank database²⁸ was used to ascertain if a drug was withdrawn and the reason for withdrawal.. The Liver Toxicity Knowledge Base (LTKB)²⁹ served for several purposes of the analysis: the severity class, drug label, and DILI concern were extracted from the DILIRank list³⁰. The classification as DILI + was extracted from the DILI List (DILIst)³¹. Many of the toxicity properties (mitochondrial liability and reactive metabolite formation), pharmacokinetic properties (half-life, lipophilicity, plasma protein binding, enterohepatic circulation and hepatic metabolism), and the Biopharmaceutics Drug Disposition Classification System (BDDCS) class were also extracted from the LKTB. The injury type of DILI and the toxicity mechanism(s) of the drug were extracted from the Liver Toxicity (LiverTox) Database³². Physicochemical properties were extracted from DrugBank²⁸. From this same database, the metabolic pathway and the enzymes implied were annotated.

Establishment of a unified list of 10 DILI + and 10 DILI– control compounds

After revising the different control drugs used for DILI research on in vitro human models, the great heterogeneity of in vitro model systems, procedures and endpoints between different studies prevented compiling a thorough and unified list of DILI control drugs, based only on the literature analysis. Therefore, clinical data from DILI cases caused by the different drugs analysed, as well as physicochemical, pharmacokinetic, and toxicological characteristics were also taken into account. Moreover, the contribution of a panel of DILI experts, members of Pro-Euro DILI Network (COST-Action 17–112) Working Groups 1 (“In depth-phenotyping of DILI”) and 3 (“Preclinical evaluation of DILI”), invited to discuss the establishment of the control DILI compounds list, was crucial. To be included, drugs were required to be amply explored in both the selected studies and clinical databases and represent all major types of liver injury phenotypes and drug metabolism. After the selection of a potential list of DILI + and DILI- drugs based on the reviewed articles and the cases enrolled in the different DILI registries, experts convened to deliberate on the appropriateness of these drugs, drawing on their extensive knowledge in the clinical and preclinical fields of DILI. In instances of no consensus, the drug under question was replaced by another possessing similar characteristics that met the necessary criteria for inclusion in the list (as explained in the Discussion section).

Literature search

The search strategy retrieved 2,936 studies from four databases (PubMed, Embase, WoS, and Scopus). Of them, 1,341 were duplicate records. After screening the title and abstract, 1,923 studies were excluded, as they did not meet the inclusion criteria. The main reasons for exclusion were: i) The study of a condition other than DILI (75%), ii) not studying hepatotoxicity in in vitro human models (17.1%), and iii) not being an original article (7.9%). Full text of the remaining 125 articles were assessed for eligibility, being 67 studies excluded mainly due to: i) lack of DILI studies in in vitro human models, and ii) absence of hepatotoxicity assays. Furthermore, eight articles^33–40 that appeared to meet the inclusion criteria were excluded due to the detection of a bias in the selection of different drugs. In the hepatotoxicity assays, instead of selecting drugs with the potential to cause general DILI, the different articles focused on specific DILI phenotypes, such as cholestatic, mitochondrial-induced DILI or drug-induced steatosis. Finally, 51 articles were included for in-depth systematic review (Fig. 1).

Study characteristics

Characteristics such as the type of in vitro human model, the study context, the DILI control drugs used, including the concentrations and time of exposure, along with the different endpoints studied and respective predictivity of the system, are essential when assessing the relevance of a given in vitro liver model. A summary of the main characteristics of the 51 articles that fully met the inclusion criteria is shown in Table 1.

Table 1

Key features of studies included in the systematic review.
			Drugs used				Treatment
Study ID	DILI categorization	DILI categories	DILI + drugs used	DILI- drugs used	Total	Human model/s used	Drug concentration	Time of exposure	Endpoint/s measured
Albrecht, W. et al., 2019 ⁵⁹	Binary	DILI positive/DILI negative	14	14	28	PHHs	Variable range depending on the drug (includes Cmax and cytotoxic concentrations)	24 h, 48 h and 7 days	Cell viability, gene expression
Albrecht, W. et al., 2019 ⁵⁹	Binary	DILI positive/DILI negative	14	14	28	HepG2		24 h, 48 h and 7 days	Cell viability, glutathione-related endpoints, gene expression
Aleo, MD. et al., 2017 ⁸¹	Binary	DILI positive/DILI negative	70	55	125	PHHs	Range between 0 and 300 µM (includes Cmax)	2 h	Transporter-related endpoints, cholestasis-related endpoints
Atienzar, FA. et al., 2014 ⁸²	Binary	DILI positive/DILI negative	40	11	51	PHHs	Multiples of Cmax (12.5, 25, 50 and 100*Cmax)	5 days	Cell viability, glutathione-related endpoints
Atienzar, FA. et al., 2014 ⁸²	Binary	DILI positive/DILI negative	40	11	51	HepG2	Multiples of Cmax (12.5, 25, 50 and 100*Cmax)	5 days	Cell viability, glutathione-related endpoints
Basharat, A. et al., 2020 ⁸³	Binary	DILI positive/DILI negative	95	46	141	HepG2 spheroids	Variable range depending on the drug	4 days	Cell viability
Basharat, A. et al., 2020 ⁸³	Binary	DILI positive/DILI negative	95	46	141	HepaRG spheroids	Variable range depending on the drug	4 days	Cell viability
Bell, CC. et al., 2018 ⁸⁴	Binary	DILI positive/DILI negative	5	1	6	PHHs	Variable range depending on the drug	72 h, 7 days and 14 days	Cell viability, reactive metabolites-related endpoints
Bell, CC. et al., 2018 ⁸⁴	Binary	DILI positive/DILI negative	5	1	6	PHHs spheroids	Variable range depending on the drug	72 h, 7 days and 14 days	Cell viability, reactive metabolites-related endpoints
Boon, R. et al., 2020 ⁶⁰	Binary	DILI positive/DILI negative	9	7	16	HepG2	Variable range depending on the drug (includes EC50/Cmax)	24 h and 72 h	Cell viability
						HepaRG
						HC6X
						HC3X
						HepatoPac system
						HLCs (iPSCs-derived)
Burkard, A. et al., 2012 ⁷³	Tertiary	Cytotoxic/Moderately-Cytotoxic/Non-cytotoxic	8	12	20	Upcyte hepatocytes	Variable range depending on the drug	72 h	Cell viability
Burkard, A. et al., 2012 ⁷³	Tertiary	Cytotoxic/Moderately-Cytotoxic/Non-cytotoxic	8	12	20	HepG2	Variable range depending on the drug	72 h	Cell viability
Eckstrum, K. et al., 2022 ⁴⁶	Binary	DILI positive/DILI negative	4	3	7	PHHs	Variable range depending on the drug (from non-toxic to extremely toxic concentrations)	72 h	Cell viability, apoptosis detection, apoptosis detection, mitochondrial-related endpoints, albumin secretion, urea secretion
						LSECs
						Liver chip
Garside, H. et al., 2014 ⁸⁵	Binary	DILI positive/DILI negative	108	36	144	PHHs	Variable range depending on the drug	4 h, 24 h and 48 h	Apoptosis detection, ROS-related endpoints, nuclear and cell cycle-related endpoints, lipid-related endpoints, mitochondrial-related endpoints, cellular stress-related endpoints
Garside, H. et al., 2014 ⁸⁵	Binary	DILI positive/DILI negative	108	36	144	HepG2	Variable range depending on the drug	4 h, 24 h and 48 h
Gerets, HH. et al., 2012 ⁸⁶	Binary	DILI positive/DILI negative	16	5	21	PHHs	Variable range depending on the drug	4 h, 24 h, 48 h and 72 h	Cell viability
						HepG2
						HepaRG
Harada, K. et al., 2021 ⁴⁵	Binary	DILI positive/DILI negative	55	21	76	PHHs	100 µM (optimized for evaluating RM formation)	24 h	Cell viability, glutathione-related endpoints
						HepG2
						HepaRG
						PXB cells
						HepaRG - PXB cells coculture
Hirashima, R. et al., 2017 ⁸⁷	Binary	DILI positive/DILI negative	24	7	31	HaCaT	100 µM	6 h and 48 h	Gene expression
Hussain, F. et al., 2020 ⁸⁸	Binary	DILI positive/DILI negative	59	38	97	PHHs	Range between 2 and 1000 µM	16 h	Cell viability, nuclear and cell cycle-related endpoints, lipid-related endpoints, metabolizing enzymes-related endpoints, phenotypic features, mitochondrial-related endpoints, albumin secretion, urea secretion
						HepG2
						HepaRG
Kawaguchi, M. et al., 2020 ⁸⁹	Tertiary	Most DILI Concern/Less DILI Concern/No DILI Concern	45	9	54	PHHs	Variable range depending on the endpoint	24 h	Cholestasis-related endpoints, lipid-related endpoints, mitochondrial-related endpoints
Kawaguchi, M. et al., 2020 ⁸⁹	Tertiary	Most DILI Concern/Less DILI Concern/No DILI Concern	45	9	54	HepG2	Variable range depending on the endpoint	24 h and 72 h
Khetani, SR. et al., 2013 ⁹⁰	Binary	DILI positive/DILI negative	35	10	45	PHHs	Multiples of Cmax (1, 30, 60 and 100*Cmax)	14 days	Cell viability, glutathione-related endpoints, albumin secretion, urea secretion
Kohara, H. et al., 2020 ⁹¹	Tertiary	Cholestatic or mixed/Hepatocellular/No DILI	37	8	45	PHHs	NA (allows to calculate the ratio between IC50 and Cmax)	30 minutes	Cell viability, cholestasis-related endpoints, transcriptomics
						HepaRG
						PXB cells
Li, F. et al., 2020 ⁹²	Quinary	Quinary (Severe DILI, High DILI Concern, Low DILI Concern, Enzyme Elevations and No DILI)	62	38	100	PHHs	Multiples of Cmax (10, 30 and 50*Cmax)	8 days	Cell viability, transcriptomics, albumin secretion, inflammation-related endpoints
Li, F. et al., 2020 ⁹²	Quinary		62	38	100	PHHs spheroids	Multiples of Cmax (10, 30 and 50*Cmax)	8 days
Lin Z., Will Y., 2012 ⁹³	Binary	DILI positive/DILI negative	109	72	181	HepG2	Range between 0.3 and 300 µM (includes Cmax)	72 h	Cell viability
Luo, Y. et al., 2012 ⁹⁴	Binary	DILI positive/DILI negative	58	31	89	HepG2	Range between 0 and 300 µM (includes or gets close to Cmax)	24 h	Cell viability, apoptosis detection
Maiuri, AR. et al., 2017 ⁹⁵	Binary	DILI positive/DILI negative	14	10	24	HepG2	Range between 0 and 100*Cmax	24 h	Cell viability
Menecozzi, M. et al., 2015 ⁹⁶	Binary	DILI positive/DILI negative	5	1	6	PHHs	Variable range depending on the drug	30 minutes, 2 h and 5 h	Cell viability, ROS-related endpoints, endoplasmic reticulum-related endpoints, mitochondrial-related endpoints, calcium-related endpoints
Norona, LM. et al., 2020 ⁴²	Binary	DILI positive/DILI negative	85	74	159	PHHs	Range between 0 and 100 µM	24 h	Cell viability, gene expression, cholestasis-related endpoints
Norona, LM. et al., 2020 ⁴²	Binary	DILI positive/DILI negative	85	74	159	HEK293T (Human Embryonary Kidney cell line)	Range between 0 and 100 µM	24 h
Novik, EI., et al., 2017 ⁹⁷	Binary	DILI positive/DILI negative	12	7	9	PHHs	Variable range depending on the drug (includes Cmax and cytotoxic concentrations)	24 h	Cell viability, cholestasis-related endpoints, reactive metabolites-related endpoints, metabolizing enzymes-related endpoints
Novik, EI., et al., 2017 ⁹⁷	Binary	DILI positive/DILI negative	12	7	9	PHHs-NPSCs coculture		24 h, 7 days and 14 days
O'Brien, PJ. et al., 2006 ⁹⁸	Binary	DILI positive/DILI negative	147	95	242	HepG2	Variable range depending on the drug (includes Cmax and cytotoxic concentrations)	3 days	Cell viability, apoptosis detection, glutathione-related endpoints, nuclear and cell cycle-related endpoints, membrane-related endpoints, mitochondrial-related endpoints, calcium-related endpoints
Oda, S. et al., 2016 ⁶⁹	Binary	Most DILI-Concern/No or Less DILI concern	53	43	96	HL-60 (treated with conditioned media from HepG2 and differentiated HepaRG)	Range between 10 and 100 µM	24 h	Cell viability, gene expression
Oda, S. et al., 2021 ⁹⁹	Binary	DILI positive/DILI negative	27	50	77	HepG2	Multiples of Cmax (1, 2.5, 5, 10, 25, 50, 100*Cmax)	24 h	Cell viability, gene expression, transcriptomics
						PBMCs
						PBMCs/HepG2 coculture system
Ott, LM. et al., 2017 ¹⁰⁰	Binary	DILI positive/DILI negative	10	2	12	HepaRG	Variable range depending on the drug (includes Cmax)	24 h, 3 days and 7 days	Cell viability, metabolizing enzymes-related endpoints
Ott, LM. et al., 2017 ¹⁰⁰	Binary	DILI positive/DILI negative	10	2	12	HepaRG spheroids	Variable range depending on the drug (includes Cmax)	24 h, 3 days and 7 days	Cell viability, metabolizing enzymes-related endpoints
Persson, M. et al., 2013 ¹⁰¹	Tertiary	Severely hepatotoxic/Moderately hepatotoxic/Non-hepatotoxic	67	35	102	HepG2	Range between 0.01 and 100 µM	24 h and 72 h	Cell viability, membrane-related endpoints, mitochondrial-related endpoints, lysosomal-related endpoints
Proctor, WR. et al., 2017 ¹⁰²	Binary	DILI positive/DILI negative	69	41	110	PHHs	Range between 0 and 100*Cmax	48 h	Cell viability
Proctor, WR. et al., 2017 ¹⁰²	Binary	DILI positive/DILI negative	69	41	110	hLiMT (spheroid human liver microtissues)	Range between 0 and 100*Cmax	5 days, 6 days and 14 days	Cell viability
Rana, P. et al., 2019 ⁴⁴	Binary	DILI positive/DILI negative	73	60	133	HepG2	Multiples of Cmax (1, 10, 30 and 100*Cmax)	24 h	Mitochondrial-related endpoints
Rose, S. et al., 2022 ¹⁰³	Tertiary	Severe DILI/High DILI/Negative control	22	5	27	Hepoid-HepaRG (HepaRG spheroids)	Multiples of Cmax (1, 5, 20, 50 and 100*Cmax)	14 days	Cell viability, apoptosis detection, cholestasis-related endpoints, nuclear and cell cycle-related endpoints, transcriptomics, lipid-related endpoints, reactive metabolites-related endpoints, albumin secretion, urea secretion
Schadt, S. et al., 2015 ⁴³	Binary	High DILI/Low DILI	54	64	118	PHHs	Variable range depending on the drug (highest concentration was 200 µM)	48 h	Cell viability, glutathione-related endpoints, cholestasis-related endpoints, metabolizing enzymes-related endpoints
						HLM (Human Liver Microsomes)	10 and 50 µM	NA
						Human BSEP expressing vesicles	Variable range depending on the drug	NA
Shah, F. et al., 2015 ¹⁰⁴	Binary	DILI positive/DILI negative	70	55	125	HepG2	Range between 0.3 and 300 µM	72 h	Cell viability
						THLE (Transformed human liver epithelial)	Range between 0.3 and 300 µM	72 h	Cell viability
						Human BSEP expressing vesicles	Range up to 100 µM	NA	Cholestasis-related endpoints
Shimizu, Y. et al., 2021 ¹⁰⁵	Binary	DILI positive/DILI negative	266	92	358	Human liver supersomes expressing CYP450 enzymes	10 µM	NA	Metabolizing enzymes-related endpoints
Shinozawa, T. et al., 2021 ¹⁰⁶	Binary	DILI positive/DILI negative	206	32	238	Human liver organoids (HLO)	Range between 0.1 and 100 µM	24 h	Cell viability, cholestasis-related endpoints, mitochondrial-related endpoints
Thompson, RA. et al., 2012 ¹⁰⁷	Binary	DILI positive/DILI negative	27	9	36	THLE which do not express P450s	Range between 0 and 300 µM	24 h	Cell viability, transporter-related endpoints
						THLE which selectively expresses P450 3A4	Range between 0 and 300 µM
						HepG2	Range between 0 and 250 µM
						Human BSEP expressing vesicles	Range between 0 and 1000 µM	5 min	Cholestasis-related endpoints
Tolosa, L. et al., 2012 ¹⁰⁸	Binary	DILI positive/DILI negative	66	12	78	HepG2	Range between 1 and 1000 µM	3 h and 24 h	Cell viability, lipid-related endpoints, mitochondrial-related endpoints, calcium-related endpoints
Tomida, T. et al., 2015 ¹⁰⁹	Binary	DILI positive/DILI negative	17	15	32	HepaRG	Multiples of Cmax (1.6, 6.3, 25, and 100*Cmax).	24 h	Cell viability, apoptosis detection, glutathione-related endpoints, lipid-related endpoints, albumin secretion
Tomida, T. et al., 2017 ¹¹⁰	Binary	DILI positive/DILI negative	23	15	38	HepaRG	Multiples of Cmax (6.25, 12.5, 25, 50 and 100*Cmax)	24 h	Cell viability, apoptosis detection, glutathione-related endpoints, albumin secretion
Vorrink, SU. et al., 2018 ¹¹¹	Binary	DILI positive/DILI negative	70	54	124	PHHs spheroids	Multiples of Cmax (1, 5 and 20*Cmax)	14 days	Cell viability
Ware RB. et al., 2015 ¹¹²	Binary	DILI positive/DILI negative	37	10	47	iMPCCs (iPSC-derived HLCs and 3T3-J2 fibroblasts)	Multiples of Cmax (25 and 100*Cmax)	6 days	Metabolizing enzymes-related endpoints, phenotypic features, albumin secretion, urea secretion
Ware RB. et al., 2019 ¹¹³	Binary	DILI positive/DILI negative	6	3	9	PHHs-MPCCs (3T3-J2 fibroblasts)	Multiples of Cmax (25 and 100*Cmax)	6 days	Cell viability, phenotypic features, albumin secretion, urea secretion
Ware RB. et al., 2021 ¹¹⁴	Binary	DILI positive/DILI negative	37	10	47	HepaRG	Multiples of Cmax (25 and 100*Cmax)	6 days	Cell viability, metabolizing enzymes-related endpoints, albumin secretion
Ware RB. et al., 2021 ¹¹⁴	Binary	DILI positive/DILI negative	37	10	47	HepaRG-MPCCs (3T3-J2 fibroblasts)	Multiples of Cmax (25 and 100*Cmax)	6 days
Williams DP. et al., 2020 ¹¹⁵	Tertiary	High DILI/Medium DILI/Low DILI	63	33	96	HepG2	Range between 0 and 250 µM	24 h	Mitochondrial-related endpoints
						HepG2/C3A spheroids		4 days	Cell viability
						THP-1		48 h	Cell viability
						Human BSEP expressing vesicles	Range between 0 and 1000 µM	NA	Cholestasis-related endpoints
Xu, JJ. et al., 2008 ¹¹⁶	Binary	DILI positive/DILI negative	200	144	344	PHHs	100*Cmax	24 h	Cell viability, glutathione-related endpoints, metabolizing enzymes-related endpoints
Xu, J. et al., 2018 ¹¹⁷	Binary	DILI positive/DILI negative	18	16	34	HepG2	Multiples of Cmax (1.6, 6.4, 25 and 100*Cmax)	24 h	Cell viability, glutathione-related endpoints
Xu, J. et al., 2018 ¹¹⁷	Binary	DILI positive/DILI negative	18	16	34	HepaRG	Multiples of Cmax (1.6, 6.4, 25 and 100*Cmax)	24 h	Cell viability, glutathione-related endpoints
Xu, Q. et al., 2019 ⁴¹	Binary	DILI positive/DILI negative	65	55	120	HepG2	Range between 1 and 1000 µM	Overnight	Cell viability, reactive metabolites-related endpoints
Yamaoka T, Kitamura Y., 2015 ¹¹⁸	Binary	IDR positive/IDR negative	7	3	10	HLM (Human Liver Microsomes)	10 µM	1 h	Glutathione-related endpoints
Yu, KN. et al., 2018 ¹¹⁹	Binary	DILI positive/DILI negative	16	6	22	Hep3B	Range between 0 and 3000 µM	24 h	Cell viability
Yu, KN. et al., 2018 ¹¹⁹	Binary	DILI positive/DILI negative	16	6	22	3D Hep3B cells on a DataChip/MetaChip	Range between 0 and 3000 µM	24 h	Cell viability
Yucha, RW. et al., 2017 ¹²⁰	Tertiary	Major DILI/Minor DILI/No DILI	22	27	49	PHHs	Range between 0 and 100 µM (includes cytotoxic concentrations)	1 h	Cholestasis-related endpoints
Yucha, RW. et al., 2017 ¹²⁰	Tertiary	Major DILI/Minor DILI/No DILI	22	27	49	Human BSEP expressing vesicles		20 min	Cholestasis-related endpoints
Zhang, J. et al., 2016 ¹²¹	Binary	General-DILI/No-DILI and Severe DILI/Non-severe DILI (Two classifications)	117	35	152	PHHs	Range up to 250 µM	NA	Cell viability, apoptosis detection, glutathione-related endpoints, ROS-related endpoints
3D, three-dimensional; BSEP, bile salt export pump; C_max, maximum drug concentration in blood; CYP450, cytochromes P450; DILI, drug-induced liver injury; EC50, half maximal effective concentration; HLC, hepatocyte-like cells; HLM, human liver microsomes; HLO, human liver organoid; IC50, Half maximal inhibitory concentration; iPSC, induced pluripotent stem cells; hLiMT, human liver microtissue; LSEC, liver sinusoidal endothelial cells; MPCCs, micropatterned cocultures; NA, not available; NPSC, non-parenchymal stromal cells; PBMCs, peripheral blood mononuclear cells;; PHH, primary human hepatocytes; PXB, fresh human hepatocytes isolated from chimeric mice with highly humanised liver (PXB-mouse); RM, reactive metabolites; ROS, reactive oxygen species; THLE, transformed human liver epithelial.

Reliability of the studies

The quality and reliability of all 51 publications assessed using the Modified ToxRTool are shown in Fig. 2. Most studies were considered reliable without restrictions regarding test substance identification (69%), test system characterization (84%), study design description (81%), results documentation (81%), and plausibility of the study design and data (100%). Based on these parameters, most of the studies were categorised as reliable without restrictions (59%). The remaining studies were categorised as reliable with restrictions (41%), being potentially useful. Importantly, no article was classified as not reliable regarding this categorization.

DILI categorization

Among the 51 articles included, 43 studies (84%) classified the drugs using a simple binary categorization, namely DILI-positive (DILI+) versus DILI-negative (DILI-). In contrast, 7 studies (14%) used a tertiary categorization, e.g., Most-, Less- or No-DILI-concern, and only 1 (2%) used a quinary categorization, i.e., classifying as Severe-, High-, Low- or No-DILI-concern and Enzyme elevations (Table 1).

DILI control drugs

The models’ applicability depends on the number of tested DILI + and DILI- controls. A list of all the drugs used in the 51 articles is presented in Supplementary Information 1 (SI1). Specifically, the minimum number of drugs tested in the included studies was 4 DILI + and 1 DILI- and a maximum of 266 DILI + and 144 DILI-, with an average of 55 DILI + and 30 DILI- drugs tested. Nevertheless, most articles tested up to 50 DILI+ (53%) and DILI- (78%) drugs.

Supplementary Information 2 summarizes the characteristics of the 104 DILI + and 123 DILI- drugs most commonly used in the 51 studies analysed. Diclofenac is undoubtedly the most widely investigated DILI + control drug (45 studies; 88%), followed by troglitazone (41 studies; 80%) and flutamide (36 studies; 71%). Regarding DILI- controls, buspirone (25 studies; 49%), dexamethasone (21 studies; 41%), and diphenhydramine (18 studies; 35%) were ranked as the most tested compounds. Drugs such as chlorpromazine, imipramine, and simvastatin, which are widely used as DILI+ (53%, 49%, and 35% of studies, respectively), were also used as DILI- (2%, 6%, and 6% of studies, respectively). However, the biggest difference is seen in the DILI- control drugs, e.g., with acetylsalicylic acid, fluoxetine, or warfarin being widely used as DILI- controls (33%, 25%, and 22% of studies, respectively); but also, to similar extents as DILI+ (22%, 20% and 16% of the studies, respectively). Interestingly, the LTKB considers all these drugs Less-DILI-Concern²⁹. In fact, within the included articles, 89% of the drugs categorized as Less-DILI-Concern and 64% categorized as Ambiguous-DILI-Concern were used as both DILI + and DILI- controls. A priori, this heterogeneity in the classification stems from the fact that some articles consider drugs belonging to the Less-DILI-Concern category as DILI-, despite being classified as potentially hepatotoxic.

Notably, the LTKB gathers diverse datasets for DILI assessment and prediction. We therefore evaluated the distribution of DILI labelling and severity among the drugs studied (Fig. 3). Out of 104 DILI + control drugs analysed, 91% present labelling related to DILI, in which 49 drugs are included in the “Warnings and precautions'' section (47%), 19 drugs exhibited “Box warning” (18%), 14 drugs were “Withdrawn” (13%) and 13 drugs were in the “Adverse Reactions” section (13%). Likewise, among the study indicated DILI + controls, 66 drugs were categorised as Most-DILI-Concern (63%) and 29 were categorised as Less-DILI-Concern (28%). On the other hand, within the 123 DILI- control drugs, only 39% present LTKB labels, with 27 drugs evidencing “Adverse Reactions” (22%) and 13 drugs “Warnings and precautions” (11%). Accordingly, 59 drugs are categorised as No-DILI-Concern (48%), 33 as Less-DILI-Concern (27%), and 13 as Ambiguous-DILI-Concern (11%). Interestingly, two drugs used as DILI- (2%), levofloxacin and atorvastatin, present Most-DILI-Concern and were used both as DILI + and DILI- controls.

DILI Registries

When examining cases of hepatotoxicity for the analysed drugs across various DILI registries, the Spanish DILI Registry has a higher number of reported cases for DILI + drugs (53%) compared to DILI- drugs (20%). Conversely, the Pro-Euro-DILI Registry has the fewest reported cases for both DILI+ (29%) and DILI- (19%) drugs. Nevertheless, these differences could be explained, in part, by the different numbers of total cases between registries.

Within the 104 DILI+, 91 of the DILI + drugs (88%) undergo phase I metabolism, and 41 (39%) undergo phase II metabolism. Moreover, 38% and 73% of DILI + compounds lead to the formation of reactive metabolites or present mitochondrial liability, respectively. Finally, hepatocellular (64%) and cholestatic (38%) injury are the most prevalent types of liver damage caused by the DILI + drugs, followed by mixed injury (30%). Hepatic enzyme elevations were only reported on 2% of DILI + drugs. Within the 123 DILI- drugs, 83 drugs are metabolised by phase I enzymes (67%) and 33 by phase II enzymes (27%), however, only a minority are described to lead to the formation of reactive metabolites (16 drugs; 13%) or present mitochondrial liability (40 drugs; 33%).

In vitro human models

Factors such as the cell type(s) and the number of different model formats used in each study constitute essential features when evaluating the model’s predictivity. Additionally, models can be divided into cellular and cell-free systems (Table 2), although cell-free systems, such as human liver microsomes, and human bile salt export pump (BSEP)-expressing vesicles, were utilised in only 17% of the articles.

Table 2

Types and format of *in vitro* human models used in the studies outlined in Table 1.
Type of model		Models	Culture conditions	References
CELLULAR MODELS	Primary cultures	PHHs	2D monoculture	^{42,43,45,46,59,81,82,84–86,88−92,96,97,102,116,120,121}
			2D coculture with NPSCs	⁹⁷
			2D coculture with 3T3-J2 fibroblasts	¹¹³
			HepatoPac coculture	⁶⁰
			3D spheroids	^84,92,111
			3D hLiMT	¹⁰²
			3D Liver chip	⁴⁶
		Upcyte hepatocytes	2D monoculture	⁷³
		PXB cells	2D monoculture	^45,91
		PBMCs	2D monoculture	⁹⁹
	Cell lines	HepG2	2D monoculture	^{41,44,45,59,60,73,82,85,86,88,89,93–95,98,99,101,104,107,108,115,117}
			2D coculture with PBMCs	⁹⁹
			3D spheroids	^83,115
		HepaRG	2D monoculture	^{45,60,86,88,91,100,109,110,114,117}
			2D coculture with PXB cells	⁴⁵
			2D coculture with 3T3-J2 fibroblasts	¹¹⁴
			3D spheroids	^83,100,103
		Hep3B	2D monoculture	¹¹⁹
		Hep3B	3D DataChip/MetaChip	¹¹⁹
		HL-60	2D monoculture	⁶⁹
		LSECs	2D monoculture	⁴⁶
		LSECs	3D Liver chip	⁴⁶
		HaCaT	2D monoculture	⁸⁷
		HEK293T	2D monoculture	⁴²
		HC3X	2D monoculture	⁶⁰
		HC6X	2D monoculture	⁶⁰
		THLE	2D monoculture	^104,107
		THP-1	2D monoculture	¹¹⁵
	Induced hepatocytes	iPSC-derived HLCs	2D monoculture	⁶⁰
			2D coculture with 3T3-J2 fibroblasts	¹¹²
			3D organoids	¹⁰⁶
NON-CELLULAR MODELS	Liver vesicles	BSEP-expressing vesicles	NA	^{43,104,107,115,120}
		Human Liver Supersomes	NA	¹⁰⁵
		Human Liver Microsomes	NA	^43,118
2D, two-dimensional; 3D, three-dimensional; BSEP, bile salt export pump; C_max, maximum drug concentration in blood; HLC, hepatocyte-like cells; hLiMT, human liver microtissue; iPSC, induced pluripotent stem cells; LSEC, liver sinusoidal endothelial cells; NA, not available; NPSC, non-parenchymal stromal cells; PBMCs, peripheral blood mononuclear cells; PHH, primary human hepatocytes; PXB, fresh human hepatocytes isolated from chimeric mice with highly humanised liver (PXB-mouse); THLE, transformed human liver epithelial.

As expected, human primary hepatocytes (PHHs) represent the cell model used in most of the included studies (59%), followed by cancer cell lines such as HepG2 and HepaRG (47% and 29%, respectively). Interestingly, technologies such as stem cell-derived models are emerging as potential alternatives to primary or tumoral cells, and about 6% of the included studies used HLCs differentiated from induced pluripotent stem cells (iPSCs). However, given the costs associated with this technology and the fact that HLCs often show a mixed (adult/foetal phenotype) phenotype, stem cell technologies have not been widely adopted by the pharmaceutical industry thus far. Other important parameters in assessing drug metabolic and toxicity readouts are the cells' morphology and spatial arrangement in vitro, which is influenced by the cell culture conditions and improved in 3D systems. Interestingly, 83% of the included studies that tested the drugs in cell cultures used two-dimensional (2D) conditions, while 15% tested under three-dimensional (3D) conditions (Fig. 4). Moreover, the addition of cells other than hepatocytes, e.g., non-parenchymal cells (e.g., liver sinusoidal endothelial cells) or stromal cells (e.g., mesenchymal cells, fibroblasts), can increase the complexity and relevance of the model and this strategy was used in ~ 20% of these studies (Fig. 4).

Finally, comparing different cell culture systems is essential to understand the relevance of each model. Indeed, the choice of the cellular components and model should be context-specific. For example, hepatocyte-endothelial cell co-culture model would be able to address the role of vascular biology in DILI; whilst cell choice and the most appropriate DILI + or DILI- drug for the study will of course affect readout including drug metabolism and toxicity pathways. Notably, 47% of the studies test the different drugs in only one human in vitro model, 45% use 2 to 3 models, and 8% test the drugs in 4 to 6 models (Table 1, Fig. 3).

Drug concentration and time of exposure

When deciding which drug concentration to use for in vitro hepatotoxicity testing, one concept that appears very frequently in the studies is that of C_max. It usually refers to the mean maximum active and therapeutic plasma concentration after a single administration at the recommended therapeutic dose in humans. Herein, about 49% of the performed hepatotoxicity assays were based on multiples of the C_max for each drug. However, as seen in Supplementary Document 3, the values assumed as C_max for the same drug in different articles still show a great variation. Nevertheless, the DILI + drugs with the most concordant C_max values were trazodone, in which 8 out of the 10 studies that used this drug assumed 5.1 µM as C_max, followed by glafenine (7 out of 10 studies; 1.9 µM) and nimesulide (15 out of 19 studies; 21.1 µM). For DILI- drugs, felodipine, propafenone, metergoline, and prednisone were the most consistent, with all studies reporting C_max values of 0.03 µM, 0.001 µM, 0.07 µM, and 0.07 µM, respectively. On the other hand, drugs such as lumiracoxib and ibuprofen showed no concordance at all, with all studies using these drugs showing different C_max values. The remaining 51% of studies report different concentration ranges, mainly based on the cytotoxicity observed in the in vitro model.

Time of exposure and suitability of the models must be established for acute and/or chronic exposures. This is particularly important given the short in vitro lifespan of PHHs. The majority of studies assessed the effect of acute exposure (≤ 72 h; 71%), while 20% assessed chronic effects (> 72 h), and 10% assessed both. Furthermore, most studies assessed only one time point (65%), whereas 2 and 3 time points represented 18% and 12% of the articles, respectively, with the remaining 6% using 4 time points.

Endpoints evaluated

In addition to the relevance of the culture system and the choice of drug, the use of multiplexing endpoints to assess multiple mechanisms of toxicity is of paramount importance. Cytotoxicity is a fundamental parameter for identifying hepatotoxicity, with 82% of the included studies assessing at least one cell viability endpoint (Fig. 5A). Several studies also included other functional parameters, namely biotransformation activity (20%), albumin (20%), or urea secretion (12%), and clinical/mechanistic endpoints, such as cholestasis (24%), mitochondrial- (22%), lipid- (12%), or inflammation-related (2%) assays. Overall, only 27% of studies relied only on one type of endpoint, of which 64% were based on cytotoxicity assays, while 59% evaluated between 2 and 4 endpoints (Fig. 5B). A detailed classification of the endpoints investigated in each article is provided in Supplementary Information 4.

Predictive capacity

As seen in Supplementary Information 5, the predictive power of the models was shown in quantitative terms in 88% of the studies. The predictive capacity of a model was defined in most cases by sensitivity (% of DILI + drugs correctly classified) and specificity (% of DILI- drugs correctly classified). Several articles present other values such as accuracy (% of total correctly classified drugs), positive likelihood ratio (PLR, ratio between the % of correctly classified DILI + drugs and incorrectly classified DILI + drugs), negative likelihood ratio (NLR, the ratio between the % of correctly classified DILI- drugs and incorrectly classified DILI- drugs), positive predictive value (PPV, the % of cases classified as DILI + that are confirmed as DILI+) and negative predictive value (NPV, the % of cases classified as DILI- that genuinely are DILI-).

Extensive variability was found when analysing all data concerning the predictive capacity of the different models. First, the number of drugs used to determine the predictive ability of a model varied among all the articles analysed, with some using different numbers of drugs depending on: the model⁴¹, the model and the endpoint^42,43, or the cutoff used⁴⁴. In most cases (52%), the sensitivity and specificity values were determined by a cutoff calculated for one or more endpoints in the different models or conditions used. In 20% of cases, the sensitivity and specificity are referred to specific endpoints for one or more models and conditions, but the threshold used is not mentioned. The predictive values in 3 studies (6%) refer to a hierarchical clustering analysis of all endpoints and models used. Harada et al.⁴⁵ calculated ROC curves for all the models and fixed the specificity at 85.7%. Only Eckstrum et al.⁴⁶ calculated different predictive values for each drug tested in every test performed and every model used.

Proposed Control Drugs: 10 DILI + and 10 DILI– compounds

After extensive analysis of all drugs examined in this study, including physicochemical, pharmacokinetic and pharmacodynamic characteristics, mode of action (MoA), etc., a provisional list of 10 DILI + and 10 DILI- drugs was established to be used to validate current and future in vitro DILI systems (Fig. 6, Table 3, and Table 4).

Table 3

Summary of the different metabolic pathways, mechanisms of toxicity, and pharmacological therapeutic class of the 10 DILI + control compounds.
Positive drug	Metabolic pathway	Phase I enzymes metabolizing the compound	Phase II enzymes metabolizing the compound	Toxicity mechanism(s)	LTKB severity class	LKTB DILI concern	Drug class (ATC/DDD Index 2022)
Diclofenac	- Oxidative metabolism to hydroxy metabolites - Conjugation to glucuronic acid, sulfate, and taurine. -Primary metabolite: 4'-hydroxy diclofenac	CYP2C9, CYP2C19,CYP1A2,CYP2C8,CYP3A4, CYP2B6, CYP2C18, CYP2D6, CYP2A6, CYP1A1, CYP1B1, CYP2C18, CYP2E1, CYP2J2, CYP3A5.	UGT2B7, UGT1A3, UGT1A9, UGT2B4, COMT, GSTP1	- Multifactorial -Immuno-allergic component. - Linkage with allelic variants of UGT2B7, CYP2C8, and ABCC2.	8	Most-DILI-Concern	M01AB05
Troglitazone	-It induces P450 CYP3A4 in vivo. -Metabolites detected in plasma and urine: Sulfate conjugate, quinone, and glucuronide metabolites.	CYP3A4, CYP2C9 CYP2C8	UGT1A3 UGT1A4 UGT1A6 UGT1A7 UGT1A8 UGT1A9 UGT1A10 UGT2B7 UGT2B15 UGT1A1	-Its alfa-tocopherol side chain can be metabolized to a quinolone-like metabolite, which may produce aberrant metabolism and hepatotoxicity. -Allergic and immune reactivity are rare.	8	Most-DILI-Concern	A10BG01
Amiodarone	-Primary metabolite: desethylamiodarone -3'-hydroxylamiodarone has been identified in mammals, but its clinical significance is unknown.	CYP3A4, CYP2C8, CYP2D6, CYP2C9, CYP1A2	No data found	-Direct damage to lipid bilayers and disturbance of lysosomal and/or mitochondrial function. - Potent inhibitor of phospholipase A. -Mitochondrial injury	8	Most-DILI-Concern	C01BD01
Ketoconazole	-Hepatic metabolism. - It may also undergo N-deacetylation, alkyl oxidation, N-oxidation, aromatic hydroxylation, or hydroxylation	CYP3A4, CYP2C8, CYP2D6, CYP2C9, CYP1A2, CYP2C19, CYP1A1, CYP3A5.	UGT1A1, UGT1A7, UGT2B7	-Unknown cause of hepatotoxicity. - It may correlate with the ability of ketoconazole to inhibit mammalian sterol synthesis. - Potent inhibitor of human CYP3A4	8	Most-DILI-Concern	J02AB02
Tamoxifen	-Tamoxifen can suffer hydroxylation, N-oxidation, N-dealkylation, or 4-hydroxylation.	CYP2D6, CYP3A4, CYP3A5, CYP3A7, CYP2C9, CYP2C19, CYP2B6, CYP1A1, CYP1B1, CYP2C8, CYP2A6, CYP2E1, CYP1A2, FMO1, FMO3.	UGT1A10, UGT1A8, UGT2B17, UGT2B15, SULT1A1, SULT2A1, SULT1E1.	-Reactive metabolites. -Formation of DNA adducts -Associated with hepatic steatosis. -Liver toxicity involves mitochondrial dysfunction (through uncoupling of mitochondrial respiration and disruption of the inner mitochondrial membrane)	8	Most-DILI-Concern	L02BA01
Chlorpromazine	-Extensively metabolized in the liver and kidneys. -In urine, 20% of chlorpromazine and its metabolites are excreted unconjugated. The remaining 80% consists of conjugated metabolites. -The major metabolites are the monoglucuronide of N-dedimethylchlorpromazine and 7-hydroxychlorpromazine.	CYP2D6, CYP1A2, CYP2E1, CYP3A4, BCHE	No data found	- Hypersensitivity-related liver injury. -Toxic intermediates. -Toxicity mechanism includes the loss of cell tight junctions, membrane integrity, and damage to efflux transport proteins. -Cholestatic/mixed pattern of injury.	2	Less-DILI-Concern	N05AA01
Isoniazid	-Primarily hepatic metabolism. -Isoniazid is acetylated by N-acetyltransferase to N-acetylisoniazid; it is then biotransformed to isonicotinic acid and monoacetylhydrazine. -The rate of acetylation is genetically determined.	CYP2A6, CYP2C8, CYP2E1, CYP1A2, CYP2C9, CYP2D6, CYP2C19, CYP3A4	NAT2	-Toxic intermediate of its metabolism. -Rates of injury may be somewhat higher in patients with a slow acetylation status -The injury may be immune-mediated.	8	Most-DILI-Concern	J04AC01
Valproate/valproic acid	-Most drug is metabolized to glucuronide conjugates -Another large portion is metabolized through mitochondrial β-oxidation -The remainder of metabolism occurs through oxidation, hydroxylation, and dehydrogenation	CYP2A6, CYP2B6, CYP2C9, CYP3A5, CYP1A2, CYP2C19, CYP3A4.	PTGS1, UGT1A4, UGT1A3, UGT1A8, UGT1A1, UGT1A10, UGT1A6, UGT1A9, UGT2B7, UGT2B15.	-Mitochondrial toxicity -Diminishment of tissue carnitine levels, hyperammonemia, and microvesicular steatosis. -Valproate hepatotoxicity is more common in patients having mutations in gamma polymerase	8	Most-DILI-Concern	N03AG01
Imipramine	-Nearly exclusively metabolized by the liver. -Less than 5% of orally administered imipramine is excreted unchanged.	CYP2C19, CYP2D6, CYP1A2, CYP3A4, CYP3A7, CYP2B6, CYP2C18, CYP2E1	O-glucuronidation	-Liver injury mechanism unknown. -Production of an intermediate metabolism may trigger a hypersensitivity reaction.	3	Less-DILI-Concern	N06AA02
Danazol	-Hepatic metabolism -Principal metabolites: ethisterone and 17-hydroxymethylethisterone.	CYP3A4, CYP19A1	No data found	-The selective inactivation of the CYP3A4 enzyme by its metabolite may contribute toward toxicity. -Danazol is associated with a variable rate of serum enzyme elevations which are usually asymptomatic and self-limited. -Drug interaction between danazol and cyclosporin may have adverse effects.	8	Most-DILI-Concern	G03XA01

Table 4

Summary of metabolism, mechanisms of toxicity, and pharmacological class of the 10 DILI- control compounds.
Negative drug	Metabolic pathway	Phase I enzymes metabolizing the compound	Phase II enzymes metabolizing the compound	Toxicity mechanism(s)	Drug class (ATC/DDD Index 2022)
Diphenhydramine	Diphenhydramine undergoes rapid and extensive first-pass metabolism.	CYP2D6, CYP1A2, CYP2C9, CYP2C19	PTGS1	Liver injury or liver abnormalities not reported.	D04AA32 /R06AA02
Isoproterenol	-Isoproterenol is predominantly metabolized to glucuronide conjugates. -Isoproterenol can also be O-methylated by catechol O-methyltransferase to the metabolite 3-O-methylisoprenaline, which can also be further glucuronidated.	CYP1A1, CYP1B1, BCHE	COMT	-Liver injury or liver abnormalities not reported. -Isoproterenol therapy can generate free radicals, which cause oxidative stress leading to progressive mitochondrial damage	R03CB01
Caffeine	-Caffeine metabolism occurs mainly in the liver. -The products of caffeine metabolism include paraxanthine, theobromine, and theophylline. -Minor metabolites ser excreted in urine as urates after additional metabolism.	CYP1A2,CYP3A4,CYP2E1,CYP2C8, CYP2C9, CYP1A1, CYP1B1, CYP2D6	NAT2	Liver injury or liver abnormalities not reported.	N06BC01
Primidone	Primidone is metabolized to phenobarbitol and phenylethylmalonamide.	CYP2C9, CYP2C19, CYP2E1, CYP3A4, CYP1A2	UGT1A1	-Primidone can induce CYP450 enzyme activities. -Primidone can interfere with porphyrin metabolism, causing worsening of porphyria.	N03AA03
Streptomycin	Aminoglycosides are not metabolized and are excreted unchanged in the urine primarily by glomerular filtration.	None	None	-Intravenous and intramuscular therapy with streptomycin have been linked to mild and asymptomatic elevations in serum alkaline phosphatase. -Only isolated case reports of acute liver injury with jaundice have been associated with streptomycin therapy and always in combination with other antituberculosis medications	J01GA01
Oxybutynin	-Oxybutynin is heavily metabolized by the liver. -Primary active metabolite: N-desethyloxybutynin. -Hepatic biotransformation also produces its major inactive metabolite, phenylcyclohexylglycolic acid.	CYP3A4, CYP2C8, CYP2D6, CYP3A5		-Oxybutynin has not been linked to episodes of liver injury. - It can cause significant drug-drug interactions.	G04BD04
Lidocaine	-Biotransformation includes oxidative N-dealkylation, ring hydroxylation, cleavage of the amide linkage, and conjugation. -The primary metabolite in urine is a conjugate of 4-hydroxy-2,6-dimethylaniline.	CYP3A4, CYP2D6, CYP3A5, CYP3A7, CYP1A2, CYP2C9, CYP2C8, CYP2A6, CYP2B6	No data found	Liver injury or liver abnormalities not reported.	N01BB02
Loperamide	-Loperamide is almost completely e metabolized by the liver. -Oxidative N-demethylation is the main metabolic pathway.	CYP3A4, CYP2C8, CYP2B6, CYP2D6	No data found	Liver injury or liver abnormalities not reported.	A07DA03
Pyridostigmine	-Pyridostigmine is hydrolyzed by cholinesterases systemically, and by microsomal enzymes in the liver. -The primary hydrolysis product is 3-hydroxy-N-methyl-pyridinium	ACHE, BCHE, CYP3A4	No data found	Liver injury or liver abnormalities not reported.	N07AA02
Tolterodine	Tolterodine metabolites: 5-Hydroxymethyl tolterodine and N-Dealkylated tolterodine.	CYP2C9, CYP2C19, CYP2D6, CYP3A4	No data found	-Liver injury or liver abnormalities due to tolterodine have rarely been reported. -Cases reported may be an organ reaction of a drug-induced hypersensitivity syndrome	G04BD07

DILI positive (DILI+) control drugs in rank order

Figure 6 lists the 10 DILI + and 10 DILI- drugs selected as positive and negative control compounds to validate new in vitro models. The 10 DILI + drugs are the following:

1. Diclofenac, ranked first as the most frequently used DILI+ control drug. Diclofenac is a nonsteroidal anti-inflammatory (NSAID) benzeneacetic acid compound that undergoes hepatic metabolism. It represents one of the most prescribed NSAIDs used as an anti-inflammatory and antirheumatic drug. Diclofenac is used as a DILI+ control in 88% of studies (45/51 positive occurrences and none using it as a negative control). Furthermore, this drug is highly represented in all the clinical DILI registries analysed. It is one of the drugs within its pharmacological group (NSAIDs) with the majority of clinical cases reported in the DILIN²⁶ and LATINDILI Network²⁷ registries.

2. Troglitazone occurred in 41 studies as DILI+ and none as DILI-. It is a thiazolidinedione drug developed for type II diabetes, metabolised by the liver. There are no recorded cases of this drug in clinical databases since it was soon withdrawn from the market due to its association with serious hepatotoxicity including acute liver failure cases leading to death/liver transplantation.

3. Amiodarone is a class III antiarrhythmic also metabolised by the liver. It ranks third on the DILI+ drugs list, with 35 positive occurrences and none as negative. Clinical cases of DILI due to this iodine-rich benzofuran have been documented in both the Spanish DILI Registry²⁴ and the DILIN²⁶.

4. Ketoconazole and acetaminophen (APAP) ranked equally in 4^th place with 33 occurrences as DILI+ and none as DILI-. Ketoconazole is an imidazole antifungal agent, metabolised by the liver, with one appearance in the Spanish DILI Registry²⁴, one in the DILIN²⁶, and two in the LATINDILI Network²⁷. Acetaminophen is a widely used analgesic and antipyretic medicine, being commonly recommended as the first choice for mild to moderate pain management. It has relatively minor effects on the activity of tissue cyclooxygenase. The process of acetaminophen-induced liver damage has been extensively studied in both humans and animal models. When acetaminophen is metabolised, a small portion of it goes through the cytochrome P450 system, forming intermediates, particularly N-acetyl-p-benzoquinoneimine, which can be toxic. If high doses of acetaminophen overwhelm this metabolic pathway, the reactive intermediate accumulates and binds to intracellular macromolecules, ultimately leading to cell injury, often through programmed necrosis^47,48. Therefore, despite being utilised in a comparable number of studies to ketoconazole, APAP was excluded from the list due to its propensity to induce intrinsic DILI (dose-dependent), although eventually it might cause acute liver failure (ALF) via idiosyncratic DILI mechanisms⁴⁹.

5. Tamoxifen is an antiestrogenic tertiary amino compound, appearing in 29 of the studies as a DILI+ control, with no use as DILI-. This drug is clinically documented in all analysed databases except the ProEuro-DILI Registry. It undergoes extensive metabolism in the liver through various pathways involving phase I and phase II enzymes.

6. Chlorpromazine, with 27 incidences as DILI+ and one as DILI-, is an aliphatic side-chain phenothiazine used as an antipsychotic, with 4 reported cases of idiosyncratic hepatotoxicity in the Spanish DILI Registry and 3 in the LATINDILI Network. This drug undergoes extensive metabolism in the liver and kidneys, primarily through CYP450 isoenzymes.

7. Isoniazid was used in 27 articles as DILI+ and one as DILI-. Isoniazid is a synthetic derivative of nicotinic acid extensively used in the treatment of tuberculosis, and there are many registered clinical cases in all analysed databases, notably with over 5% of all DILIN cases attributed to this drug. Isoniazid is metabolised in the liver by various phase I and II enzymes.

8. Valproic acid, an antiepileptic and synthetic derivative of propylpentanoic acid, was used 26 times as DILI+ and none as DILI-. Hepatotoxicity cases have been reported in all analysed clinical registries. Several phase I and II enzymes metabolise this drug in the liver.

9. Imipramine was also selected as a DILI+ control in 25 studies, and 3 as DILI-. This synthetic tricyclic derivative is used as an antidepressant and has a few cases reported in the Spanish DILI Registry and the DILIN. This drug is metabolised almost exclusively in the liver by phase I and II enzymes.

10. Danazol. This synthetic androgen with anti-gonadotropic and anti-estrogenic activity was used in 15 studies as DILI+ and none as DILI-, with two incidences reported in the Spanish DILI Registry and one in the Pro-Euro-DILI Registry. This drug is primarily metabolised in the liver, and is known to cause cholestatic liver injury.

DILI negative (DILI-) control drugs in rank order

Next, we elaborated a list of DILI- drugs, as follows:

1. Diphenhydramine, with 18 incidences as DILI- and none as DILI+, was the most frequently used DILI negative control. It is an antihistamine drug that undergoes phase I and II metabolism.

2. Isoproterenol, an adrenergic agent used to treat airway obstructive diseases was used in 17 studies as DILI-. Several phase I (CYP1A1, CYP1B1, BCHE) and II (COMT) enzymes metabolise this drug in the liver.

3. Caffeine ranks third on the DILI- drugs list, with 15 negative occurrences. It is a drug primarily metabolised in the liver by several CYP450 enzymes.

4. Primidone was the following selected drug, with 14 occurrences as DILI-. It is an antiepileptic drug metabolised in the liver by several phase I and II enzymes.

5. Streptomycin, with 14 negative occurrences too, was included next. It is an aminoglycoside antibiotic used to treat tuberculosis. It is not metabolised in the liver and is excreted in the urine.

6. Oxybutynin is a urological agent used to treat urinary incontinence. It ranks sixth on the DILI- drugs list, with 13 occurrences as DILI-. It is extensively metabolised by the CYP3A4 enzyme in the liver and intestinal wall, but it is also metabolised by other liver-phase I enzymes.

7. Lidocaine is a drug used as a local anaesthetic, and presented 8 occurrences as DILI-. It has an extensive metabolism by many phase I enzymes belonging to the CYP450 family.

8. Loperamide is an antipropulsive synthetic piperidine derivative used as an antidiarrheal, that has 7 occurrences as DILI-. It has an almost complete hepatic metabolism.

9. Pyridostigmine is an anticholinesterase used as a parasympathomimetic, with six occurrences as DILI-. It is also metabolised by the liver.

10. Tolterodine is a drug used to treat urinary incontinence, with six occurrences as DILI- too. This drug was selected due to its extensive metabolism by phase I enzymes.

Of note, given the selected drugs, several of their specific features were further explored, including metabolic pathways, mechanisms of toxicity, and pharmacological therapeutic class. In Tables 3 and 4, the phase I and phase II enzymes involved in the metabolism of these drugs can be observed. Virtually, all the phase I enzymes involved in the DILI+ drugs' biotransformation (excluding CYP2C18, CYP2J2M, FMO1, and FMO3) are also involved in DILI- drug metabolism. However, this is not the case for phase II enzymes, as the analysed DILI- control drugs are minimally subjected to phase II metabolism, including UGT, COMT, and GSTP enzyme families. Regarding the concentrations tested for each drug, the utilisation of C_max has been previously discussed in this article. In Figure 7, a simplified distribution of C_max concentrations for these 20 drugs is shown.

Early prediction of human toxicity in the drug development process remains a major challenge. Human in vitro models are becoming increasingly important because of their advantages over animal-based preclinical methods. However, the development of more physiologically relevant liver models and careful selection of control DILI + and DILI- drugs are requisites to better predict DILI liability of new drug candidates. This systematic review addresses two critical challenges: i) The lack of a standardised panel of training compounds (DILI-positive and DILI-negative drugs), which would allow appropriate and more robust validation of human in vitro liver models for hepatotoxicity studies; and ii) The wide heterogeneity of methodologies, procedures and endpoints between different studies. To our knowledge, this is the first systematic review that combines not only evidence from in vitro human liver models but also data from clinical DILI registries with expert consensus input from the Pro-Euro DILI Network. Taken together, we define an appropriate set of reference drugs (DILI + and DILI- control compounds) to assess the ability of current in vitro models to predict DILI.

Current models vary significantly in hepatic cell type and culture complexity, with no single one alone being able to recapitulate the multifactorial mechanisms of human DILI. Instead, exploiting well-defined, context-specific, in vitro human liver models with distinct characteristics is advocated to flag potential adverse outcomes and better predict DILI in the preclinical phase of drug development. In this systematic review, we confirm that primary human hepatocytes (PHHs) were most utilised in liver toxicity studies. Industry standard human hepatic cell lines such as HepG2 and HepaRG, were widely used, whilst stem cell-derived HLCs are under development. PHHs are considered the gold standard due to functional relevance but have limitations including availability, dedifferentiation in culture/reduced life-span, inter-donor variability, and cost^50,51. HepG2 cells are a practical cost-effective alternative but have low biotransformation potential, particularly in phase I enzymes. Human hepatic HepaRG cells have largely intact drug metabolic machinery⁵² and significantly outperform standard HepG2/C3A cells for many pharmaceutical and therapeutic applications⁵³. Genetic manipulation of these cell lines to reinstate specific CYP450 enzymes for metabolism-related toxicity assessment is feasible, however unregulated expression differs from PHHs^54,55. Moreover, cell-free systems are extensively employed including subcellular fractions, such as microsomes⁵⁶ or human metabolic enzyme-expressing vesicles^57,58. Whilst these models help identify important attributes of drug metabolic enzymes, they lack full cellular components for toxicity responses. In contrast, several HLC-based models can offer stable genotypes and metabolic activity for phases I and II, enabling extended testing protocols and patient-specific studies, despite incomplete maturation¹¹.

Some of the studies we analysed explored hepatocyte-based co-cultures with, for example, non-parenchymal cells, fibroblasts, or PBMCs. Micropatterned co-cultures improved PHHs, iPSC-derived HLCs, and HepaRG phenotype for long-term toxicity assessment, compared with monocultures. Thus, cell-cell/cell-matrix interactions improve physiological relevance and allow the study of paracrine communication, vital for assessing the immune system effects on DILI.

Diverse cell culture complexities can be harnessed to suit specific purposes. Such context-specific applications of human in vitro models include 2D monolayer cell systems that facilitate high-throughput drug biotransformation and cytotoxicity screening. However, limited in vivo-like tissue architecture reduces predictive value. In contrast, complex 3D liver systems better emulate tissue complexity, spatial co-location of heterotypic cells which further enhance cell-cell and cell-matrix interactions, maintaining in vivo-like phenotypes for weeks, and enabling chronic toxicity studies. Considering the temporal nature of DILI onset and development, especially in idiosyncratic DILI, long-term exposure with various dosing regimens are essential parameters to consider, along with drug dose and exposure duration.

Drug concentrations in cell-based systems must reflect in vivo conditions as closely as possible and consider potential varying mechanisms of toxicity with different concentrations of the same drug. Indeed, we observed substantial variation and lack of consensus in drug concentration for toxicity studies. Furthermore, many studies consider in vitro testing concentrations by multiplying in vivo maximum plasma concentration (≤ 100xC_max); although some studies assume distinct C_max values for the same drug (Supplementary Information 3). For example, APAP concentration used in in vitro studies often ranges between 136 µM and 165 µM, corresponding to therapeutic levels, while Albrecht et al.⁵⁹ employed 1.21 mM, the known APAP hepatotoxicity-inducing concentration in vivo.

A comprehensive and tiered approach necessitates development and application of diverse multi-cellular and context-specific human models for a more nuanced understanding of DILI's complex mechanisms^10–12. Thus, diverse endpoints are essential to enhance model reliability. However, the late-stage manifestation of cell viability markers such as cellular ATP content and LDH release requires pre-cytotoxic endpoints to identify potential DILI candidates. As an example, Boon et al.⁶⁰ demonstrated that cell viability assays combined with a cutoff of 10–20 times each drug's C_max yielded high specificity (100%) but lower sensitivity (11–67%), possibly due to pre-cell death mechanisms.

Nevertheless, functional and mechanistic readouts were also tested in several of the included studies. Functional changes can be assessed by parameters like biotransformation activity, albumin, and urea production. In contrast, mechanistic signals encompass mitochondrial membrane potential, generation of oxidative and nitrosative species, endoplasmic reticulum stress, lipid accumulation and biotransformation, covering different hepatotoxicity mechanistic parameters such as mitochondrial dysfunction, oxidative stress, steatosis, and reactive metabolite formation. These readouts are studied through parallel in vitro assays or automated multiparametric high-content screening methods including cell-based phenotypic screening which can aid in predicting MoA as well as cellular toxicities⁶¹. Indeed, DITOP⁶² (Drug-induced toxicity related protein database, which collects literature-reported drug-induced toxicity-related proteins) could assist in the early triage of phenotypic hits which present network profiles with potentially high prediction of human toxicity. Ultimately, it is very challenging to compare all the predictive power data from each model across different studies, since the methodology used to obtain these values is extremely heterogeneous.

This systematic review has also raised paradoxical classifications of various drugs. Acetylsalicylic acid, a commonly used NSAID, constitutes a drug categorised as Less-DILI-Concern by the LTKB and presents clinical DILI cases reported within the Spanish DILI Registry²⁴. Interestingly, in the reviewed studies, this drug is widely used as both a DILI- and DILI + control. Other drugs such as fluoxetine, warfarin, alendronic acid, entacapone, or metformin also share this problematic feature. This clearly raises concerns regarding whether an Ambiguous-/ or Less-DILI-Concern drug should be used as a DILI- control compound. When defining a panel of control drugs, it is essential to have different categories within DILI + drugs to cover not only severe but also mild hepatotoxicity effects, i.e., to have drugs categorised as Most-DILI-Concern and Less-DILI-Concern. However, when defining true DILI- controls, the drugs should not have any propensity to cause DILI, i.e., only No-DILI-Concern drugs should be included. Moreover, DILI- compounds must not bear any clinical cases within the DILI registries.

The initial aim of our review of literature-reported DILI studies was to propose a consensus list of DILI + and DILI- drugs to validate human in vitro DILI models. The list evolved after conducting an extensive analysis of all the drugs and model systems examined in this study. This included data from clinical cases of DILI caused by the same drugs under analysis in this systematic review, along with their physicochemical, pharmacokinetic, and toxicological characteristics. Additionally, a panel of experts in the field of DILI (COST-Action 17–112 “Pro-Euro DILI Network”) was convened to provide expertise and critical input on the proposed list.

DILI + drugs were the initial focus, identifying ten drugs fulfilling the established conditions. Firstly, the drugs were required to have been significantly utilised in both the selected studies and clinical databases. Additionally, it was crucial for the drugs chosen to accurately represent all major types of liver injury phenotypes (hepatocellular, cholestatic, or mixed) and drug metabolism. Next, we discuss aspects of drug metabolism, mechanisms of toxicity, and pharmacological properties of the selected DILI control compounds (see Fig. 6).

DILI + Control Drugs

Diclofenac, the most utilised DILI + control drug, is mainly metabolised by CYP2C9 and conjugated by UGT2B7, leading to the formation of reactive metabolites that can contribute to DILI. Thus, hepatotoxicity from diclofenac may be metabolism-dependent. Toxicity induced by diclofenac is primarily hepatocellular, through reactive metabolites formation and mitochondrial dysfunction^63,64, with a significant immune-allergic component. Strikingly, although troglitazone showed low hepatotoxicity in preclinical animal models, it was shown to be highly hepatotoxic in humans⁶⁵, highlighting the low predictivity of most animal models of DILI. Like diclofenac, it has been rated 8 in the LTKB severity classification and it is categorised as Most-DILI-Concern. This drug is also metabolised in the liver by several phase I enzymes (mainly CYP3A4) and several phase II enzymes belonging to the UGT family. Its toxicity is apparently due to the production of a highly reactive quinolone-like metabolite, bile acid accumulation by BSEP inhibition, and mitochondrial dysfunction⁶⁵, leading to a hepatocellular pattern of DILI.

After troglitazone, flutamide, although initially ranked 3rd as a DILI + control drug, was not included in the final list due to an inadequate understanding of its mechanism of hepatotoxicity. Amiodarone is a cationic amphiphilic drug metabolised in the liver predominantly through CYP3A4 and CYP2C8 enzyme metabolism. The main mechanism of hepatotoxicity is believed to be directly related to the accumulation of this drug in lysosomes. This leads to subsequent inhibition of the phospholipase A enzyme, damage to the lipid bilayers, accumulation of lipid material in the lysosomes, and mitochondrial dysfunction¹². All three types of damage, cholestatic, mixed, and hepatocellular, are observed.

Oral formulations of Ketoconazole have been withdrawn in most countries, but can be used with severe restrictions in the US⁶⁶, whilst it is still also administered topically. Ketoconazole is metabolised in the liver by several phase I and II enzymes, although no highly reactive metabolites are known. The toxicity of this drug appears predominantly hepatocellular - caused by inhibition of sterol synthesis and/or CYP3A4, altering the serum levels of several drugs that are metabolised via the CYP450 system, increasing the toxicity of these compounds.

Tamoxifen was selected as a positive control compound for induction of hepatic steatosis, and is extensively metabolised by phase I and II enzymes. The idiosyncratic reaction leading to liver injury is not associated with the drug's estrogenic effects but rather with the formation of genotoxic reactive metabolites that lead to hepatic steatosis⁶⁷ and can cause mitochondrial dysfunction⁶⁸. The type of injury caused by tamoxifen can manifest as cholestatic, mixed, or hepatocellular.

Chlorpromazine (ranked #6) was included as a DILI + control compound, although it had one occurrence as DILI-, in Oda et al.⁶⁹. The authors considered chlorpromazine a DILI- drug since it is annotated as a Less-DILI-Concern compound. However, chlorpromazine produces mixed or cholestatic liver injury after causing a hypersensitivity reaction. Its extensive metabolism in the liver and kidneys also produces toxic metabolic intermediates and induces damage to efflux transport proteins⁷⁰. Chlorpromazine is considered a Less-DILI-Concern compound, probably because of its low prescription frequency in favour of other new generation atypical antipsychotic agents, which are rarely associated with clinically apparent liver injury⁷¹. We consider it important to include chlorpromazine due to its well-established hepatotoxicity in the literature, with many DILI cases presenting acute cholestatic hepatitis⁷².

Similarly, Isoniazid (ranked #7) also had one occurrence as a DILI- control, in Burkard⁷³. The authors stated that although isoniazid presents clinically relevant liver toxicity, albeit cases are rare and occur through indirect mechanisms. However, it is well known that the development of hepatocellular DILI after isoniazid treatment is due to the accumulation of a toxic metabolic intermediate, caused by low acetylation status or abnormalities (sometimes genetic) of CYP2E1, or due to the immunomodulatory production of anti-isoniazid/anti-CYP450 antibodies⁷⁴.

The next DILI + compound, valproic acid, is principally metabolised by CYP2A6. Valproate typically induces hepatocellular injury, likely due to mitochondrial toxicity resulting from the inhibition of β-oxidation and a decrease in tissue carnitine levels and depletion of GSH and an increase in mitochondrial cholesterol trafficking^75,76. DILI caused by valproate can manifest with hyperammonemia and microvesicular steatosis⁷⁷.

Imipramine, being a drug with Less-DILI-Concern, was considered by some authors a DILI- compound, although it is predominantly used as a positive control in the literature. We included imipramine as a DILI + control since it can induce idiosyncratic DILI. This drug is extensively metabolised by CYP1A2, CYP3A4, and CYP2C19 enzymes. Although the precise mechanism of imipramine-induced DILI is unclear, it is believed that metabolic intermediates trigger a hypersensitivity reaction, resulting in three types of clinical injury, namely cholestatic, mixed, and hepatocellular⁷⁸. Like amitriptyline, imipramine inhibits acid sphingomyelinase and hence disrupts intracellular cholesterol trafficking, which affects mitochondrial antioxidant defense and impairs mitophagy⁷⁹. New human-relevant multicellular predictive models are required to be able to distinguish between DILI subtypes.

Finally, we sought a DILI + compound that predominantly causes a cholestatic pattern of liver injury. The first drug fulfilling these criteria was the immunosuppressive cyclosporin A. However, it was excluded from the list because there are no registered cases in the analysed clinical databases despite having 19 occurrences as DILI + and none as DILI- in the included studies. In contrast, the cholestatic danazol was selected as fitting the criteria. Danazol is primarily metabolised in the liver by CYP3A4 and CYP19A1, and is thought to induce cholestasis via malfunction of bile salt transporter proteins⁸⁰.

DILI- Control Drugs

For selecting DILI- drugs, a series of pre-established criteria were followed. First, the selected drugs should be metabolised to the greatest extent by the same phase I and Phase II enzymes as the DILI + drugs. Additionally, these drugs should not have reported occurrences as DILI + in the literature, be classified as severity class 0 in the LTKB, and be considered No-DILI-Concern. Diphenhydramine, isoproterenol, and caffeine were the first DILI- selected, with the most occurrences as DILI-, and all undergo phase I and II metabolism. Primidone and streptomycin, with 14 incidences as DILI- were identified. Streptomycin represents the only drug that does not undergo hepatic metabolism, allowing differentiation of potential DILI found in the developed model independently of hepatic metabolism. Notably, flumazenil which undergoes hepatic metabolism, was utilised in 14 studies as a DILI- control, however, there is no available information regarding specific enzymes involved, and therefore it was not included.

Five lower-ranked DILI- compounds were identified: oxybutynin, lidocaine, loperamide, pyridostigmine, and tolterodine. The inclusion of these drugs was principally due to their almost complete metabolism by a range of CYP450 enzymes in the liver.

As we have commented, almost all Phase I enzymes participating in the biotransformation of DILI + drugs are also engaged in DILI- drug metabolism. However, this does not apply to Phase II enzymes, as the evaluated DILI- control drugs undergo minimal phase II metabolism. However, the selection of DILI- drugs with a predominant phase I metabolism in DILI- compounds is the best strategy, since drugs that primarily undergo Phase I metabolism may have more potent active metabolites, while those undergoing Phase II metabolism may have less potent or inactive metabolites. Moreover, drugs that undergo extensive Phase I metabolism may be more likely to produce toxic metabolites, which can lead to adverse effects. On the other hand, drugs that undergo Phase II conjugation reactions are often detoxified and less likely to cause toxicity.

Regarding the recommended concentrations to use for each drug included on the list, a suggestion based purely on bibliographic evidence is given (Table 5). These concentrations were selected based on the most used value as C_max for each one of the drugs. In the case that the same number of articles use different concentrations for the DILI- drugs, the highest value is proposed.

Table 5

Suggested Cmax concentrations of DILI + and DILI- control compounds.
Control Compound	Suggested Cmax (µM)
Diclofenac	8
Troglitazone	6.4
Amiodarone	5.3
Ketoconazole	11.3
Tamoxifen	0.2
Chlorpromazine	0.9
Isoniazid	76.6
Valproate	693.4
Imipramine	0.1
Danazol	0.1
Diphenhydramine	0.34
Isoproterenol	2.4
Caffeine	77.24
Primidone	4.77
Streptomycin	74.5
Oxybutynin	0.02
Lidocaine	36
Pyridostigmine	1.1
Tolterodine	0.04

Given the multifactorial, complex nature of idiosyncratic DILI, no single system has yet emerged as a universal preclinical testing platform. The liver toxicology community needs to develop guidelines to improve preclinical standardisation for in vitro model design, and to build and test a portfolio of well-characterised, context-specific predictive platforms. Moreover, there is a clear and unmet need for consensus on the reference drugs to be used, in order to validate DILI assays, recommendations about the concentrations to test and criteria for interpreting the data. This systematic study proposes an evidence-based, consensus-driven, unified list of 11 positive and 10 negative control drugs to provide benchmarking, continuity and reproducibility in validation of human in vitro models for improving preclinical drug safety testing studies. From an initial corpus of nearly 3,000 literature-based studies, only 51 met the rigorous inclusion criteria to achieve this goal based on PRISMA guidelines. In addition, we report tremendous variation with the types of human in vitro DILI models utilised, for example, across cell choice and culture formats (2D monolayers to 3D multi-cellular cultures) and conditions (medium formulations; ECM configuration). Together with the lack of concordance due to interspecies differences of animal models with human DILI, a paradigm shift is required to develop models with human-relevant biological variation and complexity, which can lead to better understanding of mechanistic and predictive DILI signals. Cross validation of at least 2–3 human in vitro models with both in vivo and clinical data and integration using artificial intelligence (AI) approaches should form part of standard operating procedure for DILI prediction in the future. Applying appropriate DILI + and DILI- control compounds to test these systems would provide added value in terms of validation, benchmarking and improving relevance of the model to evaluate drug toxicity as well as pharmacologic effects.

Acknowledgments

We thank Dr. Minjun Chen for providing the LKTB list.

The list of twenty drugs (DILI+ and DILI-) was thoroughly reviewed, corrected, and validated by an expert committee from the Pro-Euro DILI Network. Their expertise and insights were instrumental in ensuring the accuracy and reliability of the selected drugs. The committee's involvement added a layer of credibility to the list, further enhancing its value in DILI research.

This work was supported by grants from the Instituto de Salud Carlos III, cofounded by Fondo Europeo de Desarrollo Regional (FEDER) (contract numbers: PI21/01248, PI19-00883, PT20/00127), and grants from Consejería de Salud de Andalucía cofounded by FEDER (contract numbers: PEMP-0127-2020, PI-0310-2018) and Agencia Española del Medicamento y Productos Sanitarios. M Villanueva-Paz and I Alvarez-Alvarez hold Sara Borrell research contracts from ISCIII (CD21/00198 and CD20/00083, respectively). H Niu holds a postdoctoral research contract funded by Junta de Andalucía (POSTDOC_21_00780). A Segovia-Zafra holds a Jaume Bosch predoctoral contract funded by CIBERehd. D Di Zeo-Sánchez holds an iPFIS predoctoral contract funded by ISCIII (IFI21/0034). SCReN and CIBERehd are funded by ISCIII. This publication is based on work from COST Action' CA17112 - Prospective European Drug-Induced Liver Injury Network' supported by COST (European Cooperation in Science and Technology); www.cost.eu. We acknowledge the support from grants PID2019-111669RB-100 from Plan Nacional de I+D funded by the Agencia Estatal de Investigación (AEI) and the Fondo Europeo de Desarrollo Regional (FEDER) and from the CIBEREHD; as well as support from AGAUR of the Generalitat de Catalunya SGR-2017-1112, and from the Project 201916/31 funded by Fundació Marató TV3.

Author contributions

Conceptualization- L.N.; Study Design-A.S.Z., M.V.P., M.I.L.; Search Strategy, Study Selection and Data Extraction- A.S.Z., M.V.P., G.M.C., A.B.G, D.D.S.; Quality assessment- A.S.Z., M.V.P., A.S., G.M.C., A.B.G., D.D.S, I.A.A., H.N., S.G., I.M.; Data analysis- A.S.Z., M.V.P., A.S., G.M.C., A.B.G., D.D.S, I.A.A., H.N., L.S.-V., S.G., I.M., R.J.A; Establishment of a unified list of 10 DILI+ and 10 DILI– control compounds- A.S.Z., M.V.P, A.S., L.N., M.I.L, JC.F.-C., FJ.C., J.M.; Writing – Original Draft- A.S.Z., M.V.P., A.S Writing – Review & Editing- R.J.A., M.I.L., JC.F.-C., FJ.C., J.M., L.N.

Competing interests

The authors declare no competing interests.

Materials & Correspondence

Correspondence and material requests should be addressed to:

-M. Isabel Lucena. Departamento de Farmacología, Facultad de Medicina, Universidad de Málaga, Boulevard Louis Pasteur 32, 29010, Málaga, Spain.

E-mail: [email protected]

-José C. Fernández-Checa. Department of Cell Death and Proliferation, Institute of Biomedical Research of Barcelona (IIBB), CSIC, 08036, Barcelona, Spain.

E-mail: [email protected]; [email protected]

Hoofnagle, J. H. & Björnsson, E. S. Drug-Induced Liver Injury - Types and Phenotypes. N Engl J Med 381, 264-273, doi:10.1056/NEJMra1816149 (2019).
Lee, W. M. Drug-Induced Hepatotoxicity. New England Journal of Medicine 349, 474-485, doi:10.1056/NEJMra021844 (2003).
Walker, P. A., Ryder, S., Lavado, A., Dilworth, C. & Riley, R. J. The evolution of strategies to minimise the risk of human drug-induced liver injury (DILI) in drug discovery and development. Arch Toxicol 94, 2559-2585, doi:10.1007/s00204-020-02763-w (2020).
Andrade, R. J. et al. Drug-induced liver injury. Nature Reviews Disease Primers 5, 58, doi:10.1038/s41572-019-0105-0 (2019).
Aithal, G. et al. Case Definition and Phenotype Standardization in Drug-Induced Liver Injury. Clinical Pharmacology & Therapeutics 89, 806-815, doi:https://doi.org/10.1038/clpt.2011.58 (2011).
Watkins, P. Drug Safety Sciences and the Bottleneck in Drug Development. Clinical Pharmacology & Therapeutics 89, 788-790, doi:https://doi.org/10.1038/clpt.2011.63 (2011).
Gerd, A. K.-U. et al. Drug-induced liver injury: recent advances in diagnosis and risk assessment. Gut 66, 1154, doi:10.1136/gutjnl-2016-313369 (2017).
Martignoni, M., Groothuis, G. M. M. & de Kanter, R. Species differences between mouse, rat, dog, monkey and human CYP-mediated drug metabolism, inhibition and induction. Expert Opinion on Drug Metabolism & Toxicology 2, 875-894, doi:10.1517/17425255.2.6.875 (2006).
Olson, H. et al. Concordance of the toxicity of pharmaceuticals in humans and in animals. Regul Toxicol Pharmacol 32, 56-67, doi:10.1006/rtph.2000.1399 (2000).
Fernandez-Checa, J. C. et al. Advanced preclinical models for evaluation of drug-induced liver injury - consensus statement by the European Drug-Induced Liver Injury Network [PRO-EURO-DILI-NET]. J Hepatol 75, 935-959, doi:10.1016/j.jhep.2021.06.021 (2021).
Serras, A. S. et al. A Critical Perspective on 3D Liver Models for Drug Metabolism and Toxicology Studies. Front Cell Dev Biol 9, 626805, doi:10.3389/fcell.2021.626805 (2021).
Weaver, R. J. et al. Managing the challenge of drug-induced liver injury: a roadmap for the development and deployment of preclinical predictive models. Nat Rev Drug Discov 19, 131-148, doi:10.1038/s41573-019-0048-x (2020).
Zhou, Y., Shen, J. X. & Lauschke, V. M. Comprehensive Evaluation of Organotypic and Microphysiological Liver Models for Prediction of Drug-Induced Liver Injury. Frontiers in Pharmacology 10, doi:10.3389/fphar.2019.01093 (2019).
Schofield, C. A. et al. Evaluation of a Three-Dimensional Primary Human Hepatocyte Spheroid Model: Adoption and Industrialization for the Enhanced Detection of Drug-Induced Liver Injury. Chemical Research in Toxicology 34, 2485-2499, doi:10.1021/acs.chemrestox.1c00227 (2021).
Godoy, P. et al. Recent advances in 2D and 3D in vitro systems using primary hepatocytes, alternative hepatocyte sources and non-parenchymal liver cells and their use in investigating mechanisms of hepatotoxicity, cell signaling and ADME. Arch Toxicol 87, 1315-1530, doi:10.1007/s00204-013-1078-5 (2013).
Xu, J. J., Diaz, D. & O'Brien, P. J. Applications of cytotoxicity assays and pre-lethal mechanistic assays for assessment of human hepatotoxicity potential. Chem Biol Interact 150, 115-128, doi:10.1016/j.cbi.2004.09.011 (2004).
Vinken, M. & Hengstler, J. G. Characterization of hepatocyte-based in vitro systems for reliable toxicity testing. Arch Toxicol 92, 2981-2986, doi:10.1007/s00204-018-2297-6 (2018).
Page, M. J. et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 372, n71, doi:10.1136/bmj.n71 (2021).
Ouzzani, M., Hammady, H., Fedorowicz, Z. & Elmagarmid, A. Rayyan—a web and mobile app for systematic reviews. Systematic Reviews 5, 210, doi:10.1186/s13643-016-0384-4 (2016).
Schneider, K. et al. “ToxRTool”, a new tool to assess the reliability of toxicological data. Toxicol Lett 189, 138-144, doi:https://doi.org/10.1016/j.toxlet.2009.05.013 (2009).
Segal, D. et al. Evaluation of the ToxRTool’s ability to rate the reliability of toxicological data for human health hazard assessments. Regulatory Toxicology and Pharmacology 72, 94-101, doi:https://doi.org/10.1016/j.yrtph.2015.03.005 (2015).
Klimisch, H. J., Andreae, M. & Tillmann, U. A Systematic Approach for Evaluating the Quality of Experimental Toxicological and Ecotoxicological Data. Regulatory Toxicology and Pharmacology 25, 1-5, doi:https://doi.org/10.1006/rtph.1996.1076 (1997).
World Health Organization (WHO), C. C. f. D. S. M. Guidelines for ATC classification and DDD assignment, <https://www.whocc.no/atc_ddd_index/> (2010).
Stephens, C. et al. Comprehensive analysis and insights gained from long-term experience of the Spanish DILI Registry. J Hepatol 75, 86-97, doi:10.1016/j.jhep.2021.01.029 (2021).
Björnsson, E. S. et al. A new framework for advancing in drug-induced liver injury research. The Prospective European DILI Registry. Liver Int 43, 115-126, doi:10.1111/liv.15378 (2023).
Chalasani, N. et al. Features and Outcomes of 899 Patients With Drug-Induced Liver Injury: The DILIN Prospective Study. Gastroenterology 148, 1340-1352.e1347, doi:10.1053/j.gastro.2015.03.006 (2015).
Bessone, F., Hernandez, N., Lucena, M. I. & Andrade, R. J. The Latin American DILI Registry Experience: A Successful Ongoing Collaborative Strategic Initiative. Int J Mol Sci 17, 313, doi:10.3390/ijms17030313 (2016).
Wishart, D. S. et al. DrugBank: a comprehensive resource for in silico drug discovery and exploration. Nucleic Acids Research 34, D668-D672, doi:10.1093/nar/gkj067 (2006).
Chen, M. et al. FDA-approved drug labeling for the study of drug-induced liver injury. Drug Discovery Today 16, 697-703, doi:https://doi.org/10.1016/j.drudis.2011.05.007 (2011).
Chen, M. et al. DILIrank: the largest reference drug list ranked by the risk for developing drug-induced liver injury in humans. Drug Discovery Today 21, 648-653, doi:https://doi.org/10.1016/j.drudis.2016.02.015 (2016).
Thakkar, S. et al. Drug-induced liver injury severity and toxicity (DILIst): binary classification of 1279 drugs by human hepatotoxicity. Drug Discovery Today 25, 201-208, doi:https://doi.org/10.1016/j.drudis.2019.09.022 (2020).
LiverTox. Clinical and Research Information on Drug-Induced Liver Injury, <https://www.ncbi.nlm.nih.gov/books/NBK547852/> (
Susukida, T., Sekine, S., Nozaki, M., Tokizono, M. & Ito, K. Prediction of the clinical risk of drug-induced cholestatic liver injury using an in vitro sandwich cultured hepatocyte assay. Drug Metab. Dispos. 43, 1760-1768 (2015).
Aleo, M. D. et al. Moving beyond Binary Predictions of Human Drug-Induced Liver Injury (DILI) toward Contrasting Relative Risk Potential. Chemical Research in Toxicology 33, 223-238 (2020).
Benet, M. et al. A simple transcriptomic signature able to predict drug-induced hepatic steatosis. Arch Toxicol 88, 967-982 (2014).
Eakins, J. et al. A combined in vitro approach to improve the prediction of mitochondrial toxicants. TOXICOLOGY IN VITRO 34, 161-170 (2016).
López-Riera, M., Conde, I., Castell, J. V. & Jover, R. A novel MicroRNA signature for cholestatic drugs in human hepatocytes and its translation into novel circulating biomarkers for drug-induced liver injury patients. Toxicol. Sci. 173, 229-243 (2020).
Pradip, A. et al. High content analysis of human pluripotent stem cell derived hepatocytes reveals drug induced steatosis and phospholipidosis. Stem Cells International 2016 (2016).
Tilmant, K. et al. In vitro screening of cell bioenergetics to assess mitochondrial dysfunction in drug development. Toxicol. Vitro 52, 374-383 (2018).
Tomida, T., Ishimura, M. & Iwaki, M. A cell-based assay using HepaRG cells for predicting drug-induced phospholipidosis. JOURNAL OF TOXICOLOGICAL SCIENCES 42, 641-650 (2017).
Xu, Q. et al. Can Galactose Be Converted to Glucose in HepG2 Cells? Improving the in Vitro Mitochondrial Toxicity Assay for the Assessment of Drug Induced Liver Injury. Chem. Res. Toxicol. 32, 1528-1544 (2019).
Norona, L. M. et al. In vitro assessment of farnesoid X receptor antagonism to predict drug-induced liver injury risk. Arch Toxicol 94, 3185-3200, doi:10.1007/s00204-020-02804-4 (2020).
Schadt, S. et al. Minimizing DILI risk in drug discovery - A screening tool for drug candidates. Toxicol. Vitro 30, 429-437 (2015).
Rana, P., Aleo, M. D., Gosink, M. & Will, Y. Evaluation of in Vitro Mitochondrial Toxicity Assays and Physicochemical Properties for Prediction of Organ Toxicity Using 228 Pharmaceutical Drugs. Chem. Res. Toxicol. 32, 156-167 (2019).
Harada, K., Kohara, H., Yukawa, T., Matsumiya, K. & Shinozawa, T. Cell-based high-throughput screening for the evaluation of reactive metabolite formation potential. Toxicol. Vitro 74 (2021).
Eckstrum, K., Striz, A., Ferguson, M., Zhao, Y. & Sprando, R. Evaluation of the utility of the Beta Human Liver Emulation System (BHLES) for CFSAN's regulatory toxicology program. Food Chem Toxicol 161, 112828 (2022).
Ramachandran, A. & Jaeschke, H. Oxidant Stress and Acetaminophen Hepatotoxicity: Mechanism-Based Drug Development. Antioxid Redox Signal 35, 718-733, doi:10.1089/ars.2021.0102 (2021).
Jaeschke, H. & Ramachandran, A. Acetaminophen-induced apoptosis: Facts versus fiction. J Clin Transl Res 6, 36-47 (2020).
Kaplowitz, N. Idiosyncratic drug hepatotoxicity. Nat Rev Drug Discov 4, 489-499, doi:10.1038/nrd1750 (2005).
Sison-Young, R. L. et al. Comparative Proteomic Characterization of 4 Human Liver-Derived Single Cell Culture Models Reveals Significant Variation in the Capacity for Drug Disposition, Bioactivation, and Detoxication. Toxicol Sci 147, 412-424, doi:10.1093/toxsci/kfv136 (2015).
den Braver-Sewradj, S. P. et al. Inter-donor variability of phase I/phase II metabolism of three reference drugs in cryopreserved primary human hepatocytes in suspension and monolayer. Toxicol In Vitro 33, 71-79, doi:10.1016/j.tiv.2016.02.013 (2016).
Anthérieu, S., Chesné, C., Li, R., Guguen-Guillouzo, C. & Guillouzo, A. Optimization of the HepaRG cell model for drug metabolism and toxicity studies. Toxicol In Vitro 26, 1278-1285, doi:10.1016/j.tiv.2012.05.008 (2012).
Nelson, L. J. et al. Human Hepatic HepaRG Cells Maintain an Organotypic Phenotype with High Intrinsic CYP450 Activity/Metabolism and Significantly Outperform Standard HepG2/C3A Cells for Pharmaceutical and Therapeutic Applications. Basic Clin Pharmacol Toxicol 120, 30-37, doi:10.1111/bcpt.12631 (2017).
Xuan, J., Chen, S., Ning, B., Tolleson, W. H. & Guo, L. Development of HepG2-derived cells expressing cytochrome P450s for assessing metabolism-associated drug-induced liver toxicity. Chemico-Biological Interactions 255, 63-73, doi:https://doi.org/10.1016/j.cbi.2015.10.009 (2016).
Dara, L., Liu, Z. X. & Kaplowitz, N. Mechanisms of adaptation and progression in idiosyncratic drug induced liver injury, clinical implications. Liver Int 36, 158-165, doi:10.1111/liv.12988 (2016).
Choi, J. M. et al. Prediction of Drug-Induced Liver Injury in HepG2 Cells Cultured with Human Liver Microsomes. Chemical Research in Toxicology 28, 872-885, doi:10.1021/tx500504n (2015).
Morgan, R. E. et al. Interference with bile salt export pump function is a susceptibility factor for human liver injury in drug development. Toxicol Sci 118, 485-500, doi:10.1093/toxsci/kfq269 (2010).
Köck, K. et al. Risk Factors for Development of Cholestatic Drug-Induced Liver Injury: Inhibition of Hepatic Basolateral Bile Acid Transporters Multidrug Resistance-Associated Proteins 3 and 4. Drug Metabolism and Disposition 42, 665-674, doi:10.1124/dmd.113.054304 (2014).
Albrecht, W. et al. Prediction of human drug-induced liver injury (DILI) in relation to oral doses and blood concentrations. Arch Toxicol 93, 1609-1637, doi:10.1007/s00204-019-02492-9 (2019).
Boon, R. et al. Amino acid levels determine metabolism and CYP450 function of hepatocytes and hepatoma cell lines. Nat Commun 11, 1393 (2020).
Hughes, R. E., Elliott, R. J. R., Dawson, J. C. & Carragher, N. O. High-content phenotypic and pathway profiling to advance drug discovery in diseases of unmet need. Cell Chem Biol 28, 338-355, doi:10.1016/j.chembiol.2021.02.015 (2021).
Zhang, J. X. et al. DITOP: drug-induced toxicity related protein database. Bioinformatics 23, 1710-1712, doi:10.1093/bioinformatics/btm139 (2007).
Madsen, K. G. et al. Bioactivation of diclofenac in vitro and in vivo: correlation to electrochemical studies. Chem Res Toxicol 21, 1107-1119, doi:10.1021/tx700419d (2008).
Boelsterli, U. A. Diclofenac-induced liver injury: a paradigm of idiosyncratic drug toxicity. Toxicol Appl Pharmacol 192, 307-322, doi:10.1016/s0041-008x(03)00368-5 (2003).
Smith, M. T. Mechanisms of troglitazone hepatotoxicity. Chem Res Toxicol 16, 679-687, doi:10.1021/tx034033e (2003).
Gupta, A. K. & Lyons, D. C. A. The Rise and Fall of Oral Ketoconazole. Journal of Cutaneous Medicine and Surgery 19, 352-357, doi:10.1177/1203475415574970 (2015).
Zhao, L., Krishnan, S., Zhang, Y., Schenkman, J. B. & Rusling, J. F. Differences in metabolite-mediated toxicity of tamoxifen in rodents versus humans elucidated with DNA/microsome electro-optical arrays and nanoreactors. Chem Res Toxicol 22, 341-347, doi:10.1021/tx8004295 (2009).
Ribeiro, M. P., Santos, A. E. & Custodio, J. B. Mitochondria: the gateway for tamoxifen-induced liver injury. Toxicology 323, 10-18, doi:10.1016/j.tox.2014.05.009 (2014).
Oda, S., Matsuo, K., Nakajima, A. & Yokoi, T. A novel cell-based assay for the evaluation of immune- and inflammatory-related gene expression as biomarkers for the risk assessment of drug-induced liver injury. Toxicol. Lett. 241, 60-70 (2016).
Morgan, K. et al. Chlorpromazine toxicity is associated with disruption of cell membrane integrity and initiation of a pro-inflammatory response in the HepaRG hepatic cell line. Biomed Pharmacother 111, 1408-1416, doi:10.1016/j.biopha.2019.01.020 (2019).
Slim, M. et al. Hepatic Safety of Atypical Antipsychotics: Current Evidence and Future Directions. Drug Safety 39, 925-943, doi:10.1007/s40264-016-0436-7 (2016).
Selim, K. & Kaplowitz, N. Hepatotoxicity of psychotropic drugs. Hepatology 29, 1347-1351, doi:10.1002/hep.510290535 (1999).
Burkard, A. et al. Generation of proliferating human hepatocytes using Upcyte® technology: characterisation and applications in induction and cytotoxicity assays. Xenobiotica 42, 939-956 (2012).
Metushi, I., Uetrecht, J. & Phillips, E. Mechanism of isoniazid-induced hepatotoxicity: then and now. Br J Clin Pharmacol 81, 1030-1036, doi:10.1111/bcp.12885 (2016).
Brion, L. et al. Valproic acid alters mitochondrial cholesterol transport in Y1 adrenocortical cells. Toxicology in Vitro 25, 7-12, doi:https://doi.org/10.1016/j.tiv.2010.08.006 (2011).
Torres, S. et al. Endoplasmic Reticulum Stress-Induced Upregulation of STARD1 Promotes Acetaminophen-Induced Acute Liver Failure. Gastroenterology 157, 552-568, doi:10.1053/j.gastro.2019.04.023 (2019).
Ezhilarasan, D. & Mani, U. Valproic acid induced liver injury: An insight into molecular toxicological mechanism. Environ Toxicol Pharmacol 95, 103967, doi:10.1016/j.etap.2022.103967 (2022).
Voican, C. S., Corruble, E., Naveau, S. & Perlemuter, G. Antidepressant-induced liver injury: a review for clinicians. Am J Psychiatry 171, 404-415, doi:10.1176/appi.ajp.2013.13050709 (2014).
Baulies, A. et al. Lysosomal Cholesterol Accumulation Sensitizes To Acetaminophen Hepatotoxicity by Impairing Mitophagy. Sci Rep 5, 18017, doi:10.1038/srep18017 (2015).
Boue, F., Coffin, B. & Delfraissy, J. F. Danazol and cholestatic hepatitis. Ann Intern Med 105, 139-140 (1986).
Aleo, M. D., Shah, F., He, K., Bonin, P. D. & Rodrigues, A. D. Evaluating the Role of Multidrug Resistance Protein 3 (MDR3) Inhibition in Predicting Drug-Induced Liver Injury Using 125 Pharmaceuticals. Chem. Res. Toxicol. 30, 1219-1229 (2017).
Atienzar, F. A. et al. Predictivity of dog co-culture model, primary human hepatocytes and HepG2 cells for the detection of hepatotoxic drugs in humans. Toxicol. Appl. Pharmacol. 275, 44-61 (2014).
Basharat, A., Rollison, H. E., Williams, D. P. & Ivanov, D. P. HepG2 (C3A) spheroids show higher sensitivity compared to HepaRG spheroids for drug-induced liver injury (DILI). Toxicol. Appl. Pharmacol. 408 (2020).
Comparison of Hepatic 2D Sandwich Cultures and 3D Spheroids for Long-term Toxicity Applications: A Multicenter Study v. 2 (2018).
Garside, H. et al. Evaluation of the use of imaging parameters for the detection of compound-induced hepatotoxicity in 384-well cultures of HepG2 cells and cryopreserved primary human hepatocytes. Toxicol. Vitro 28, 171-181 (2014).
Gerets, H. H. et al. Characterization of primary human hepatocytes, HepG2 cells, and HepaRG cells at the mRNA level and CYP activity in response to inducers and their predictivity for the detection of human hepatotoxins. Cell Biol Toxicol 28, 69-87 (2012).
Hirashima, R., Itoh, T., Tukey, R. H. & Fujiwara, R. Prediction of drug-induced liver injury using keratinocytes. J. Appl. Toxicol. 37, 863-872 (2017).
Hussain, F., Basu, S., Heng, J. J. H., Loo, L. H. & Zink, D. Predicting direct hepatocyte toxicity in humans by combining high-throughput imaging of HepaRG cells and machine learning-based phenotypic profiling. Arch. Toxicol. 94, 2749-2767 (2020).
Kawaguchi, M. et al. Mechanism-based integrated assay systems for the prediction of drug-induced liver injury. Toxicol Appl Pharmacol 394, 114958 (2020).
Khetani, S. R. et al. Use of micropatterned cocultures to detect compounds that cause drug-induced liver injury in humans. Toxicol. Sci. 132, 107-117 (2013).
Kohara, H. et al. High-throughput screening to evaluate inhibition of bile acid transporters using human hepatocytes isolated from chimeric mice. Toxicol. Sci. 173, 347-361 (2020).
Li, F., Cao, L., Parikh, S. & Zuo, R. Three-Dimensional Spheroids With Primary Human Liver Cells and Differential Roles of Kupffer Cells in Drug-Induced Liver Injury. J. Pharm. Sci. 109, 1912-1923 (2020).
Lin, Z. & Will, Y. Evaluation of drugs with specific organ toxicities in organ-specific cell lines. Toxicological Sciences 126, 114-127 (2012).
Luo, Y., Rana, P. & Will, Y. Palmitate increases the susceptibility of cells to drug-induced toxicity: An In Vitro method to identify drugs with potential contraindications in patients with metabolic disease. Toxicol. Sci. 129, 346-362 (2012).
Maiuri, A. R. et al. Synergistic Cytotoxicity from Drugs and Cytokines In Vitro as an Approach to Classify Drugs According to Their Potential to Cause Idiosyncratic Hepatotoxicity: A Proof-of-Concept Study. J Pharmacol Exp Ther 362, 459-473 (2017).
Mennecozzi, M., Landesmann, B., Palosaari, T., Harris, G. & Whelan, M. Sex differences in liver toxicity-do female and male human primary hepatocytes react differently to toxicants in vitro? PLoS One 10, e0122786 (2015).
Long-enduring primary hepatocyte-based co-cultures improve prediction of hepatotoxicity (United States, 2017).
O'Brien, P. J. et al. High concordance of drug-induced human hepatotoxicity with in vitro cytotoxicity measured in a novel cell-based model using high content screening. Arch Toxicol 80, 580-604 (2006).
Oda, S. et al. An in vitro coculture system of human peripheral blood mononuclear cells with hepatocellular carcinoma-derived cells for predicting drug-induced liver injury. Arch. Toxicol. 95, 149-168 (2021).
Ott, L. M., Ramachandran, K. & Stehno-Bittel, L. An Automated Multiplexed Hepatotoxicity and CYP Induction Assay Using HepaRG Cells in 2D and 3D. SLAS Discov 22, 614-625 (2017).
Persson, M., Løye, A. F., Mow, T. & Hornberg, J. J. A high content screening assay to predict human drug-induced liver injury during drug discovery. Journal of pharmacological and toxicological methods 68, 302-313 (2013).
Proctor, W. R. et al. Utility of spherical human liver microtissues for prediction of clinical drug-induced liver injury. Arch. Toxicol. 91, 2849-2863 (2017).
Rose, S. et al. DMSO-free highly differentiated HepaRG spheroids for chronic toxicity, liver functions and genotoxicity studies. Arch Toxicol 96, 243-258 (2022).
Shah, F. et al. Setting clinical exposure levels of concern for drug-induced liver injury (DILI) using mechanistic in vitro assays. Toxicol. Sci. 147, 500-514 (2015).
Shimizu, Y. et al. Association of CYP1A1 and CYP1B1 inhibition in in vitro assays with drug-induced liver injury. J. Toxicol. Sci. 46, 167-176 (2021).
Shinozawa, T. et al. High-Fidelity Drug-Induced Liver Injury Screen Using Human Pluripotent Stem Cell-Derived Organoids. Gastroenterology 160, 831-846.e810 (2021).
Thompson, R. A. et al. In vitro approach to assess the potential for risk of idiosyncratic adverse reactions caused by candidate drugs. Chemical research in toxicology 25, 1616-1632 (2012).
Tolosa, L. et al. Development of a multiparametric cell-based protocol to screen and classify the hepatotoxicity potential of drugs. Toxicol Sci 127, 187-198 (2012).
Tomida, T., Okamura, H., Satsukawa, M., Yokoi, T. & Konno, Y. Multiparametric assay using HepaRG cells for predicting drug-induced liver injury. Toxicol. Lett. 236, 16-24 (2015).
Tomida, T., Okamura, H., Yokoi, T. & Konno, Y. A modified multiparametric assay using HepaRG cells for predicting the degree of drug-induced liver injury risk. J. Appl. Toxicol. 37, 382-390 (2017).
Vorrink, S. U., Zhou, Y., Ingelman-Sundberg, M. & Lauschke, V. M. Prediction of drug-induced hepatotoxicity using long-term stable primary hepatic 3D spheroid cultures in chemically defined conditions. Toxicol. Sci. 163, 655-665 (2018).
Ware, B. R., Berger, D. R. & Khetani, S. R. Prediction of drug-induced liver injury in micropatterned co-cultures containing iPSC-derived human hepatocytes. Toxicol. Sci. 145, 252-262 (2015).
Ware, B. R., Brown, G. E., Soldatow, V. Y., LeCluyse, E. L. & Khetani, S. R. Long-Term Engineered Cultures of Primary Mouse Hepatocytes for Strain and Species Comparison Studies During Drug Development. Gene Expr 19, 199-214 (2019).
Ware, B. R., Liu, J. S., Monckton, C. P., Ballinger, K. R. & Khetani, S. R. Micropatterned Coculture with 3T3-J2 Fibroblasts Enhances Hepatic Functions and Drug Screening Utility of HepaRG Cells. Toxicol. Sci. 181, 90-104 (2021).
Williams, D. P., Lazic, S. E., Foster, A. J., Semenova, E. & Morgan, P. Predicting Drug-Induced Liver Injury with Bayesian Machine Learning. Chem. Res. Toxicol. 33, 239-248 (2020).
Xu, J. J. et al. Cellular imaging predictions of clinical drug-induced liver injury. Toxicol Sci 105, 97-105 (2008).
Xu, J., Oda, S. & Yokoi, T. Cell-based assay using glutathione-depleted HepaRG and HepG2 human liver cells for predicting drug-induced liver injury. Toxicol In Vitro 48, 286-301 (2018).
Yamaoka, T. & Kitamura, Y. Characterization of a highly sensitive and selective novel trapping reagent, stable isotope labeled glutathione ethyl ester, for the detection of reactive metabolites. JOURNAL OF PHARMACOLOGICAL AND TOXICOLOGICAL METHODS 76, 83-95 (2015).
Yu, K. N. et al. Prediction of metabolism-induced hepatotoxicity on three-dimensional hepatic cell culture and enzyme microarrays. Arch. Toxicol. 92, 1295-1310 (2018).
Yucha, R. W. et al. In Vitro Drug-Induced Liver Injury Prediction: Criteria Optimization of Efflux Transporter IC50 and Physicochemical Properties. Toxicol Sci 157, 487-499 (2017).
Zhang, J. et al. Evaluation of multiple mechanism-based toxicity endpoints in primary cultured human hepatocytes for the identification of drugs with clinical hepatotoxicity: Results from 152 marketed drugs with known liver injury profiles. Chem.-Biol. Interact. 255, 3-11 (2016).

Download PDF

Journal Publication

published 30 Apr, 2024

Read the published version in Journal of Hepatology →

Version 1

posted

You are reading this latest preprint version

Control Compounds for Preclinical Drug-Induced Liver Injury Assessment: Consensus-driven systematic review by the ProEuroDILI Network

Status:

Journal Publication

Version 1

Abstract

Figures

INTRODUCTION

MATERIAL AND METHODS

Study Design and Search Strategy

Inclusion and exclusion criteria

Study selection

Data extraction

Quality assessment

Data analysis

Drug analysis

Clinical use cases

Establishment of a unified list of 10 DILI + and 10 DILI– control compounds

RESULTS

Literature search

Study characteristics

Reliability of the studies

DILI categorization

DILI control drugs

DILI Registries

Drug concentration and time of exposure

Endpoints evaluated

Predictive capacity

Proposed Control Drugs: 10 DILI + and 10 DILI– compounds

DILI positive (DILI+) control drugs in rank order

DISCUSSION

DILI + Control Drugs

DILI- Control Drugs

CONCLUSION

Declarations

References

Supplementary Files

Status:

Journal Publication

Version 1