Investigation of the Nrf2 Stress Response at Cellular Level to Detect Compounds Inducing Oxidative and Electrophilic Stress

Abstract

Introduction:

Drug-induced liver injury (DILI) remains a significant reason for drug attrition and withdrawal. DILI can be caused by several mechanisms, one of these is the formation of reactive metabolites (RM) leading to oxidative and electrophilic stress. Since such event activates NF-E2-related factor-2 (Nrf2) and its downstream pathways, it seems relevant to measure Nrf2 activation.

Materials and Methods:

In this article, Nrf2 activation was evaluated using two approaches. First, three human hepatocytes donors and HepG2 cells were exposed to five compounds known to produce RM- and four non-RM forming compounds. The expression of seven genes under the regulation of Nrf2 was measured. An index was calculated taking into account the results of the seven genes. Subsequently, the mean index for the RM-forming and the non-RM-forming compounds was calculated for each donor, and the midpoint between these two values was determined, which represents the donor-specific threshold and the compounds were classified accordingly.

Results and Discussion:

All three donors obtained a sensitivity of 80% (4/5) and a specificity between 75% (3/4) and 100% (4/4). HepG2 cells obtained 80% (4/5) sensitivity, but very poor specificity 0% (0/4), which makes them less suitable, compared to human hepatocytes donors, for screening of electrophilic stress. Second, the Antioxidant-Response-Element-Reporter-Cell Line, AREc32, was exposed to the same set of compounds in addition to 13 RM-forming and 5 non-RM-forming compounds. Using an arbitrary cutoff of 1.5, a sensitivity of 78% (14/18) and a specificity of 88% (7/8) were obtained.

Conclusion:

Acceptable predictive values and easiness of use make the AREC32 assay more suited to investigate Nrf2 activation compared to the first approach.

Introduction

Despite major efforts of pharmaceutical companies, the lack of efficacy (30%) and unacceptable clinical safety and toxicity (30%) are still considered the major factors for discontinuation of clinical candidates nowadays.¹ Consequently, the pharmaceutical industry needs to maximize its effectiveness of preclinical R&D through the development of more predictive in vitro models to detect safety liabilities at early stages of drug development.^2,3

Although parent compounds may be responsible for causing toxicity, drug metabolites can also be associated with safety liabilities.^4,5 Due to this drug metabolite-mediated toxicity, both the pharmaceutical industry and regulatory agencies are still showing increasing concern with the safety testing of drug metabolites. Although metabolites can have beneficial effects, as proven by the therapeutic role of active metabolites in 22% of drugs prescribed in the United States in 2003,⁶ metabolites that accumulate inside the hepatocyte can induce toxic effects.^5,7 For instance, reactive metabolites (RM) exert cell stress through a wide range of mechanisms, such as depletion of glutathione (GSH), binding to enzymes, lipids, nucleic acids, and other cell structures.⁸ If not detoxified properly, these electrophiles can react with electron-rich species, that is, nucleophiles through covalent bond formation.⁴ This may lead to alterations of biological function and disturbance of redox balance, and consequently initiate serious adverse drug reactions^9,10 and lead to idiosyncratic drug toxicity. Furthermore, like parent drugs, RM may specifically inhibit other hepatocellular functions, such as the bile salt efflux pump, in which case the subsequent intracellular accumulation of its substrates may cause secondary toxic hepatocyte damage.⁸

Cells respond to the formation of RM and/or toxic parent compounds in a variety of ways, ranging from activation of survival pathways to eliciting programmed cell death to eliminate damaged cells to maintain cell homeostasis.^11,10 Survival pathways such as the antioxidant defense mechanism against oxidants and electrophiles, the heat shock response against mild heat stress, oxidative stress, or heavy metal stimuli are key defense mechanisms.¹¹ The survival pathways act through the induction of transcription factors to express the genes necessary for adaptation to chemical toxicity.⁹ For example, oxidative stress is known to activate the redox-sensitive transcription factor NF-E2-related factor-2 (Nrf2) to mediate the antioxidant stress response by inducing the expression of a large variety of cytoprotective genes.¹² Under homeostatic or nonstressed conditions, Nrf2 is coupled to the inhibitory factor Kelch-like ECH-associated protein (Keap1) in the cytosol, facilitating ubiquitination and consequent proteasomal degradation of Nrf2. Once activated by chemical inducers or oxidative stress, Nrf2 will loosen from Keap1 and translocate to the nucleus where it induces the expression of phase I genes (e.g., carboxylesterase [CES]) and phase II genes (e.g., glutathione S-transferase [GSTA] and UDP-glucuronosyltransferase [UGTs]) that are under the control of antioxidant response elements (ARE). Finally, this will result in an orchestrated antioxidant and anti-inflammatory response.^13,14 As such, cells are able to protect, using the Keap1-Nrf2-ARE pathway, against electrophilic or oxidative stress-related injuries and inflammatory diseases.¹²

Oxidative stress can be measured in different ways: directly by measuring the reactive oxygen species, which is challenging depending on the level of reactivity, and indirectly by evaluating the damage to proteins/RNA/DNA or by measuring the level of antioxidant enzymes. In a previous study, we measured the antioxidant α-GST levels after compound incubation on HepG2 cells,¹⁵ and demonstrated that α-GST was not a sufficient sensitive marker for tissue damage compared to other traditional markers such as aspartate aminotransferase and glutamate dehydrogenase.

In this study, we used two different in vitro approaches to investigate whether the Nrf2 stress response is a suitable marker for the measurement of electrophilic stress caused by RM. In the first set of experiments, the expression profile of genes under the regulation of Nrf2 was investigated as potential markers of electrophilic stress. Takakusa et al.¹⁶ measured the expression of four genes under the control of Nrf2 and showed that heme oxygenase (HO-1) gene was a good indicator of electrophilic cell stress due to RM in primary human hepatocytes (PHH). In this article, we extend Takakusa's work by¹ studying the expression patterns of seven genes under the regulation of Nrf2²; investigating donor variability by testing three different human hepatocytes donors; and³ comparing the expression patterns of the seven genes in the hepatocarcinoma cell line HepG2, which, although metabolically less competent compared to human hepatocytes,¹⁷ is commonly used for in vitro toxicity studies in the pharmaceutical industry. In the second set of experiments, the ARE reporter cell line, AREc32, was evaluated to measure Nrf2 activation. The AREc32 cell line is a stable transfected MCF7 cell line that contains a luciferase gene construct under the control of ARE and as such, provides a rapid and convenient quantification of Nrf2-mediated activation.¹⁸ Since Nrf2 initiates a cytoprotective reaction, this assay can be used to screen cytoprotective as well as electrophilic or oxidative stress-inducing compounds.^18,19 In both approaches, nine compounds selected from the article of Takakusa et al.¹⁶ were used. Five of these compounds, which are known to produce RM and drug-induced liver injury (DILI) in humans, were withdrawn from the market or received serious warnings. Four other drugs are not associated with the formation of RM in humans. In addition, another set of 13 RM-forming and 5 non-RM-forming compounds was tested in the AREc32 cell line in the second set of experiments. All these 13 hepatotoxic compounds produce RM that consequently leads to the induction of DILI.

The objectives of this study were to determine if Nrf2-mediated gene expression and/or the luminescence reporter cell line AREc32 could be used to distinguish RM-forming compounds inducing oxidative and electrophilic stress from non-RM-forming compounds.

Materials and Methods

Materials and kits

All compounds (Table 1) and reagents were of analytical grade and were purchased from Sigma-Aldrich (Diegem, Belgium). The RNeasy Plus Mini kit was purchased from Qiagen (Venlo, The Netherlands) and the reverse transcription–polymerase chain reaction (RT-PCR) reagents were obtained from Life Technologies (Gent, Belgium). The Luciferase Assay system kit was purchased from Promega (Leiden, The Netherlands).

Table 1.

Overview of the Compounds Used in the Gene Expression and the AREC32 Assay

Name	Forming of RM/OS	Therapeutic indication	Cmax (μM)	Labeling	DILI description	Test
Ticlopidine	Y⁴²	Antithrombotic	7.09	Warning and precautions	Associated with elevations of alkaline phosphatase, bilirubin, and transaminases	Gene expression and AREC32 assay (experiment 1 and 2)
Clozapine	Y⁴²	Antipsychotic	2.36	Warning and precautions	Elevations in hepatic transaminases. Damage to the liver, fulminant liver failure
Diclofenac	Y⁴²	NSAID	6.29	Warning and precautions	Severe hepatic reactions, including liver necrosis, jaundice, fulminant hepatitis with and without jaundice, and liver failure
Acetaminophen	Y⁴²	Analgesic (pain reliever) and antipyretic (fever reducer)	138.93	Not in LTKB	Hepatotoxicity due to acetaminophen overdose
Tienilic Acid	Y⁴²	Diuretic	172.12	Withdrawn	Elevation in hepatic transaminases. Hepatitis of the immunoallergic type⁴³
Caffeine	N	Psychoactive stimulant drug	42	Not in LTKB
Levofloxacin	N	Chemotherapeutic antibiotic	29.63	Not in LTKB	Very low risk for liver injury, but acute hepatitis has been observed⁴⁴
Furosemide	N	Diuretic	16	Adverse Reactions	Intrahepatic cholestatic jaundice
Aspirin	N	Analgesic and antipyretic	1650	Not in LTKB	Hepatotoxicity due to aspirin overdose
Amiodarone	Y⁴²	Antiarrhythmic	60.14	Boxed Warning	Liver injury is common, but is usually mild and evidenced only by abnormal liver enzymes. Overt liver disease can occur, however, and has been fatal in a few cases	AREC32 assay (experiment 2)
Danazol	Y⁴⁵	Antiestrogenic and synthetic steroid, sex hormone modulators, menstruation disorders, reproductive gonadal	0.46	Boxed Warning	Peliosis hepatis and benign hepatic adenoma have been observed with long-term use. This may be silent until complicated by acute, potentially life-threatening intra-abdominal hemorrhage
Furazolidone	Y⁴⁶	Antibiotic	4	Not in LTKB	Can lead to hepatocellular and cholestatic injury⁴⁷
Flutamide	Y⁴²	Antineoplastic, Antiandrogen	6.22	Boxed Warning	There have been postmarketing reports of hospitalization and rarely death due to liver failure in patients taking flutamide
Labetalol	Y⁴⁸	Beta blockers, Antihypertensive, adrenergic	0.41	Warnings and Precautions	Severe hepatocellular injury occurs rarely. The hepatic injury is usually reversible, but hepatic necrosis and death have been reported both for short- and long-term treatment, and may be slowly progressive despite minimal symptomatology
Ketoconazole	Y⁴²	Antifungal	6.59	Boxed Warning	When used orally, ketoconazole has been associated with hepatic toxicity, including some fatalities
Chlorpromazine	Y⁴²	Antipsychotic, Neurological	0.5	Adverse reaction	Chronic jaundice has been reported
Cyclosporin A	Y⁴⁹	Immunosuppressant, antipsoriasis, antiarthritic, antipruritic, ophtalmological, urological, antianemic	1.26	Warnings and Precautions	Hepatotoxicity has been noted in 4% of cases of renal transplantation, 7% of cases of cardiac transplantation, and 4% of cases of liver transplantation. Renal and liver functions should be assessed repeatedly by measurement of BUN, serum creatinine, serum bilirubin, and liver enzymes
Imipramine	Y⁴²	Antidepressant	0.63	Adverse reaction	Jaundice (simulating obstructive); altered liver function
Tamoxifen	Y⁴²	Antineoplastic, Anticancer hormonal	0.51	Warnings and Precautions	Liver effect: changes in liver enzyme levels, and on rare occasions, a spectrum of more severe liver abnormalities, including fatty liver, cholestasis, hepatitis, and hepatic necrosis
Clofibrate	Y⁵⁰	Hypolipemic and antiaterosclerosis	235.27	Warnings and Precautions	Liver function tests abnormal	AREC32 assay (approach 2)
Tacrine	Y⁴²	Cholinergic, Cognition enhancer	0.14	Warnings and Precautions	Liver Injury: Rare cases of liver toxicity associated with jaundice, raised serum bilirubin, pyrexia, hepatitis, and liver failure have been reported
Aflatoxin B1	Y⁵¹	Mycotoxin (not used as a drug)	NA	Not in LTKB	Causes acute hepatotoxicity and liver carcinoma in exposed humans⁵³
Flufenamic acid	N	NSAID	46	Not in LTKB	NSAID are known to cause idiosyncratic liver injury
Betaine HCl	N	Amino Acid supplements	940.04	No DILI concern
Isoproterenol	N	Adrenergic	4.04	No DILI concern
Praziquantel	N	Anthelmintic	1.8	Not in LTKB	Has been reported to cause serum aminotransferase elevations during therapy, but clinically apparent liver injury is very rare
Primidone	N	Antiepileptic	54.98	No DILI concern

For each drug, the chemical structure, the therapeutic indication, the Cmax, the labeling, and DILI description were presented, as well as the potential for the compound to form RM or not. The labeling and DILI description were extracted from the LTKB⁵² or from literature data. The Cmax refers to the maximum concentration of a given compound in human blood. Human Cmax values were found in the literature, as well as in commercial databases. BHA and t-BHQ, both antioxidants, are not displayed in the table.

BHA, butylhydroxyanisol; DILI, drug-induced liver injury; LTKB, Liver Toxicity Knowledge Databases; NSAID, nonsteroidal anti-inflammatory drug; RM, reactive metabolites; t-BHQ, tertiary butyl hydroquinone; Y, yes; N, no; OS, oxidative stress.

Cellular models

The human hepatocellular carcinoma (HepG2) cell line, purchased from the European Collection of Cell Cultures (ECACC, Salisbury, United Kingdom), was maintained as an adherent cell line in Dulbecco's modified Eagle medium (DMEM), supplemented with 10% fetal bovine serum (FBS), 2 mmol/L L-glutamine, and 1× nonessential amino acids at 37°C in a 5% CO₂:95% air humidified atmosphere. Cells were passaged as needed using 0.5% trypsin-EDTA. All the cell culture solutions were purchased from BioWhittaker, Inc. (Walkersville).

Cryopreserved PHH from three different donors were purchased from CellzDirect (division of Invitrogen, Carlsbad) (for donor characteristics see Table 2). Cryopreserved PHH were thawed based upon CellzDirect's standard method as previously described in Gerets et al.¹⁷

Table 2.

Human Hepatocytes Donor Demographics and Characterization

					Functional control provided by the supplier (Activity in pmol/min/10⁶ cells)
Cell preparation	Sex	Age (years)	Ethnicity	Cause of death	CYP1A2	CYP2B6	CYP3A
Hu4237	♂	57	Caucasian	Anoxia	32.7	10.4	17.7
Hu1389	♂	68	Caucasian	NA	66.8	11.6	6.4
Hu1198	♀	65	Caucasian	NA	56.9	10.6	6.8

The table gives information about the three donors of cryopreserved human hepatocytes. For each of the donors, details are described regarding their sex, age, ethnicity, and cause of death if applicable. Control CYP1A2, CYP3A4, and CYP2B6 activity are reported for each donor.

NA, nonapplicable, implies that tissue is taken from living human donors.

The ARE reporter cell line (AREc32) was licensed from Concept Life Sciences (Dundee, Scotland). AREc32 is a cell-based assay reporting on the induction of the ARE on a stable MCF7 cell line background. The cells were transfected with eight copies of the rat GSTA2 ARE cis-element linked to a luciferase gene, such that the induction of the ARE resulted in increased luciferase activity. AREC32 cells were maintained in DMEM/GlutaMAX medium containing 10% FBS, 50 U/mL penicillin, 50 μg/mL streptomycin, and 0.8 mg/mL G418 (geneticin). Cells were subcultured every three to 4 days. All products were purchased from BioWhittaker, Inc.

Optimizations

Primer validation

PHH and HepG2 cells were seeded onto six-well plates at a density of 1,000,000 cells per well. After overnight attachment in the CO₂ incubator, total RNA was extracted using the RNeasy Plus Mini kit, according to the manufacturer's instructions (Qiagen, Valencia). RNA quantity was assessed using the nanodrop spectrophotometer (Isogen, IJsselstein, The Netherlands), and RNA quality was checked with the Agilent™ bioanalyzer 2100 (Agilent Technologies, Massy, France).

Subsequently, the samples were diluted to 250 ng/μL and the RT was performed in duplicate using 2 μg of total RNA and oligo DT primer. The RT was performed for 1 hour at 42°C. Quantitative real-time PCR was performed in triplicate using 4 μL of the 1/10 diluted RT samples, 12 μL of 2× PCR master mix (Thermo Fisher Scientific, Gent, Belgium), 4 μL of forward and reverse primers, and 0.1 μL of ROX (diluted 125 times) (Thermo Fisher scientific, Gent, Belgium). The primers were designed in-house using Beacon Designer software (PREMIER Biosoft, CA) and purchased from Eurogentec (Seraing, Belgium). Table 3 gives an overview of the primers used in this study, including the sequences. Negative, genomic, and mRNA controls were also included (in duplicate) in the validation experiments.

Table 3.

Overview of the Primers Used to Measure Seven Genes Under the Regulation of Nrf2 and Beta-Actin (House Keeping Gene)

Gene name	Abbreviation	Accession number	Sense primer	Antisense primer
Heme oxygenase 1	HMOX1	NM_002133	CTTTATGCCATGTGAATGC	AAGCCAGCCAAGAGAATT
NAD(P)H dehydrogenase, quinone 1	NQO1	NM_001025433	CCACAAGGAATAAATGAGAG	GGTTATATGCCATGATAGTAAT
carboxylesterase 1	CES1	NM_001025194	ACAAACCGAAAGGAGAACATT	TCAAGTCAGCAGGAGTGTAA
UDP-glucuronosyltransferase 1, polypeptide A1	UGT1A1	NM_000463	CATCAGAGACGGAGCATT	GGAACAGCCAGACAAAAG
glutamate-cysteine ligase	GCLC	NM_001498	AATGTCCGAGTTCAATAC	AATCTGGGAAATGAAGTTAT
glutathione S-transferase alpha 1	GSTA1	NM_145740	TAGAAGAAGCAAGGAAGATT	CAGCGAATAGGAGTTGTAT
glutathione S-transferase pi 1	GSTP1	NM_000852	GGAGGCAAGACCTTCATT	ATGGATCAGCAGCAAGTC
beta-actin	b-actin	NM_001101	TGGCATCCACGAAACTACC	GTGTTGGCGTACAGGTCTT

For each SybrGreen primer pair, the gene name, the abbreviation, the gene accession number, and the sequence of the sense and the antisense primer are presented.

QPCR measurements were performed with the Stratagene Mx3000P PCR machine (SA Biosciences, Qiagen, Venlo, The Netherlands) with the following conditions: 15-minute denaturation at 95°C followed by 40 cycles of denaturation at 94°C for 15 seconds, annealing temperature of 55°C for 30 seconds, and extension at 72°C for 30 seconds. Denaturation curves were produced after 40 cycles to check the specificity of the reactions. A pair of primers was considered validated when the efficacy of amplification was comprised between 90% and 110% and with a minimum r² of 0.98.

Real-time cell analyzer experiments

Real-time cell analyzer (RTCA, xCELLigence platform; Roche Diagnostics, Vilvoorde, Belgium) experiments were performed, as described by Atienzar et al.,^20,21 to define the concentration range of the compounds that will be applied to the cells.

Selection of the housekeeping gene

Two different housekeeping genes (β-actin and GAPDH) were tested for their suitability. For that purpose, the cells were incubated with the 10 compounds at three different concentrations (30, 100, and 300 μM) for 24 hours, the total RNA was extracted, and RT-PCR reactions were performed. The most stable housekeeping gene was selected for future use.

Experiment 1: Evaluation of drug effects on PHH donors and HepG2 cells

Three PHH donors and HepG2 cells were incubated with five compounds known to form RM [ticlopidine, clozapine, diclofenac, acetaminophen, and tienilic acid (Table 1)], and four compounds that are not associated with the formation of RM [caffeine, levofloxacin, furosemide, and aspirin (Table 1)] and a technical positive control, butylhydroxyanisol (BHA), at three different concentrations (30, 100, and 300 μM) for 24 hours. The classification of the compounds with regard to their potential to form RM or not has been extracted from the literature and based on the outcome, the compounds have been classified as specified in Table 1.

Twelve-well plates were seeded with 500,000 cells/mL/well. For the PHH, the medium was changed after 6 hours of cell attachment and the compounds were added. The HepG2 cells were allowed to attach overnight and then incubated in triplicate with the compounds for 24 hours. Compound stock solutions were prepared in 100% dimethyl sulfoxide (DMSO), and further dilutions were made in the medium. The final concentration of DMSO added to the cells was 0.1%. As a negative control, 0.1% of DMSO was used. After the 24 hours of compound incubation, the medium was removed, and the cells washed with PBS and lysed with lysis buffer. The total RNA was extracted using the RNeasy Plus Mini kit (Qiagen). RT was performed in duplicate using oligo-dT primers; the triplicates from each tested concentration were pooled and diluted 1/10. Quantitative real-time PCR was performed under the same conditions as described under the primer validation section. PCR were performed in triplicate using seven genes under the control of Nrf2 (NOQ1, CES1, GSTA1, HO1, GCLC, GSTP1, and UGT1A1) and β-actin as the housekeeping gene. Negative, genomic, and mRNA controls were included in duplicate.

Gene expression analysis

Fluorescence emission was detected for each PCR cycle, and the threshold cycle (C_T) values were determined. The C_T value was defined as the actual PCR cycle when the fluorescence signal increased above the background threshold. Average C_T values from triplicate PCR were normalized to average C_T values for housekeeping gene (β-actin) from the same cDNA preparation. The calculations were done as follows: ratio 2^−ΔΔCT and ΔΔC_T = (C_T gene 1 treated − β-actin gene treated) − (C_T gene 1 control − C_T β-actin gene control).

An index, measuring the potential to induce Nrf2 and taking into account the seven investigated genes, was calculated for each compound concentration. To do so, the control condition (0.1% DMSO) was arbitrarily set to 1 and all treated conditions were compared against this control condition. The normalized sum of all ΔΔC_T values for each gene above 1 was counted and divided by the number of genes under investigations (i.e., n = 7). A gene where the normalized ΔΔC_T value was below 1 was not taken into account (i.e., zero score for the calculation of the index). The mean index for the RM-forming and the non-RM-forming compounds was calculated; subsequently, the midpoint between those two values was taken to obtain the donor-specific threshold and the compounds were classified accordingly. If the index for one of the three concentrations of an RM-forming compound was above the donor-specific threshold for the PHH and the batch-specific threshold for the HepG2 cells, the compound was considered positive. When the index was below this specific threshold value at the three concentrations, the compound was considered negative and vice versa for the non-RM-forming compounds.

Experiment 2: Evaluation of drug effects in the AREc32 reporter assay

AREc32 cells were seeded in a 96-well plate at a density of 1.2 × 10⁴ cells per well. The following day, the medium was replaced with a serum-free medium containing the tested compound. The cells were incubated with the previously described 10 compounds and another set of 18 compounds, including 13 hepatotoxic, known to generate RM, and 5 non-RM-forming compounds, and a positive control, as defined in Table 1. The compounds were tested at six concentrations ranging from 3.13 up to 100 μM in a ½ dilution series for 24 hours. On the next day, the cells were washed and the luciferase activity was measured using the Luciferase Assay system kit according to the manufacturer's instructions (Promega, Madison). The control condition was arbitrarily set to one, and the treated conditions were compared against this control condition. If the fold induction versus control for one of the concentrations tested was above 1.5, then the compound was considered positive. Based on the responses obtained with the RM-forming and non-RM-forming compound, the 1.5 threshold has been selected as representative of significant induction. When the fold induction versus control was below 1.5 for any of the concentrations tested, the compound was considered negative. In addition, the ratio of the lowest effective concentration (LOEC) over the Cmax has been calculated for each compound.

Results

Optimizations

Preliminary experiments were performed to validate the primer pairs, select the best housekeeping gene, and to study the cytotoxicity profile of all the compounds used in both approaches. The results (data not shown) are summarized as follows. For all primers, efficacy of amplification was comprised between 90% and 110%, with a minimum r² of 0.98. β-actin was selected as the most stable housekeeping gene in our studies. RTCA experiments allowed to select a range of noncytotoxic concentrations for the main experiments. The following concentrations were selected for the compounds in experiment: 1: 30, 100, and 300 μM. These concentrations were the same as used in the study of Takakusa et al.¹⁶ In experiment 2, the following concentrations were used 3.125, 6.25, 12.5, 25, 50, and 100 μM. The range of concentrations was lower in experiment 2 because some of the compounds induced important cytotoxicity effects, as AREc32 cells were a bit more sensitive compared to PHH and HepG2 cells. Consequently, 100 μM was used as a top concentration.

Experiment 1: Evaluation of drug effects on PHH donors and HepG2 cells

The induction profile of Nrf2-related genes was investigated in three human hepatocyte donors and in HepG2 cells exposed to five drugs known to be associated with the formation of RM, four drugs not associated with the formation of RM, and BHA, a technical positive control. The results are shown for PHH donors 1, 2, and 3 and HepG2 cells in Tables 4 –7, respectively.

Table 4.

Gene Expression Analysis of Donor 1 (Hu4237)

Genes			NQO1	CES1	GSTA1	HMOX1	GCLC	GSTP1	UGT1A1	Index
	Control	μM	1	1	1	1	1	1	1	1.00	Mean index	Midpoint of the mean	Donor-specific threshold of 1.08
RM-forming compounds	BHA (pos ctrl)	30	2.05	1.32	−1.51	4.00	1.49	−1.02	−1.81	1.27	Valid assay
		100	2.29	1.61	−1.76	9.87	2.59	1.21	−1.01	2.51
		300	2.09	1.42	−1.99	13.45	2.01	1.00	−1.30	2.85
	Ticlopidine	30	−1.26	1.18	1.40	2.30	−1.09	−1.33	1.65	0.93	1.70	1.08	✓
		100	1.26	5.91	2.35	19.03	1.73	15.42	2.16	6.84
		300	ND	ND	ND	ND	ND	ND	ND	ND
	Clozapine	30	1.33	1.14	−1.07	2.52	1.55	1.05	−1.01	1.08			✓
		100	2.44	6.22	−1.17	6.28	4.20	6.18	1.23	3.79
		300	ND	ND	ND	ND	ND	ND	ND	ND
	Diclofenac	30	1.75	−1.13	2.74	1.95	1.53	1.19	−1.18	1.31			✓
		100	−1.02	−1.15	1.62	1.82	1.41	−1.08	2.80	1.09
		300	−1.48	1.83	−2.56	3.55	4.53	−1.30	1.42	1.62
	Acetaminophen	30	−1.44	−2.23	1.71	−1.09	−1.12	−1.33	−2.00	0.24			✗
		100	−1.08	−1.79	1.90	1.22	−1.45	−1.76	−2.05	0.45
		300	−1.32	−1.71	2.02	1.03	−1.53	−1.30	−1.63	0.43
	Tienilic acid	30	1.14	1.01	1.55	2.28	1.23	1.11	1.59	1.41			✓
		100	1.34	−1.13	1.62	2.97	1.28	−1.02	2.00	1.31
		300	1.65	−1.08	1.67	4.05	1.20	−1.17	2.53	1.59
Non-RM-forming compounds	Caffeine	30	−1.35	−1.34	1.22	−1.16	1.01	−1.55	−2.09	0.32	0.45		✓
		100	−1.25	−1.58	−1.09	−1.25	1.01	−1.51	−2.65	0.14
		300	−1.67	−1.64	−1.11	−1.13	1.26	−1.74	−2.01	0.18
	Levofloxacin	30	1.42	−1.64	−1.07	1.06	1.74	−1.42	−2.03	0.60			✓
		100	1.00	−1.44	1.07	1.30	1.49	−1.57	−2.23	0.69
		300	1.47	−1.25	−1.19	1.83	1.96	−1.46	−2.57	0.75
	Furosemide	30	1.01	−1.58	−1.16	2.07	1.31	−1.30	−1.88	0.63			✗
		100	1.53	−1.51	−1.42	2.23	1.47	−1.19	−1.95	0.75
		300	1.45	1.27	−2.45	4.43	1.25	−1.16	−1.60	1.20
	Aspirin	30	−1.11	−1.03	−2.00	−1.07	−1.32	−1.73	−2.26	0.00			✓
		100	−1.25	−1.07	−1.38	−1.08	−1.01	−1.95	−2.17	0.00
		300	−1.07	−1.14	−1.16	−1.16	1.16	−1.79	−1.68	0.17

Primary human hepatocytes of donor 1 (Hu4237) were exposed to five drugs known to induce RM formation and four non-RM-forming drugs in human and BHA, a technical positive control, at three different concentrations (30, 100, and 300 μM) in triplicate. Seven genes under the regulation of Nrf2 (i.e., NOQ1, CES1, GSTA1, HMOX1, GCLC, GSTP1, and UGT1A1) were investigated and normalized versus the β-actin housekeeping gene. An index, representing the sum of the positive changes divided by the number of genes, was calculated for each concentration. To determine the donor-specific threshold, the mean index for the RM-forming (1.70) and the non-RM-forming compounds (0.45) was calculated. Subsequently, the midpoint of these two values was taken to obtain the donor-specific threshold. For donor 1, the threshold was 1.08 and the compounds were classified accordingly; for RM-forming compounds, if an index above 1.08 was observed for one of the triplicates, the compound was correctly classified; when the index was below 1.08 for all triplicates, the compound was noncorrectly classified. For non-RM-forming compounds, with an index below 1.08 for all triplicates, the compound was correctly classified; above 1.08, the compound was noncorrectly classified, 1.08-fold inductions are colored in gray shade.

ND, no data available because of toxic effects at the highest concentration; Nrf2, NF-E2-related factor-2.

Table 5.

Gene Expression Analysis of Donor 2 (Hu1389)

Genes			NQO1	CES1	GSTA1	HMOX1	GCLC	GSTP1	UGT1A1	Index
	Control	μM	1	1	1	1	1	1	1	1.00	Mean index	Midpoint of the mean	Donor-specific threshold of 3.52
Positive compounds	BHA (pos ctrl)	30	1.97	−1.75	3.14	−1.32	−1.19	−1.75	1.60	0.96	Valid assay
		100	6.36	1.73	2.20	10.46	1.41	1.71	5.21	4.15
		300	1.10	1.98	3.27	16.36	3.11	1.94	−1.49	3.97
	Ticlopidine	30	45.73	6.40	626.63	7.14	−1.28	4.39	44.12	18.92	5.60	3.52	✓
		100	−1.75	1.33	1.50	4.52	3.82	1.32	−1.72	1.78
		300	2.59	2.16	−1.56	4.11	2.62	2.14	2.25	2.27
	Clozapine	30	20.25	8.02	96.34	44.02	1.78	7.99	20.37	28.40			✓
		100	2.65	2.36	−1.59	11.62	5.13	2.37	2.61	3.82
		300	2.22	1.57	3.42	2.81	1.74	1.56	2.16	2.21
	Diclofenac	30	3.50	8.08	2.24	22.76	5.15	8.03	2.92	7.53			✓
		100	1.60	1.03	2.65	2.44	2.06	−1.27	1.42	1.60
		300	2.41	1.49	2.58	5.02	3.22	1.48	2.09	2.61
	Acetaminophen	30	−1.41	1.01	−2.86	1.83	2.81	1.02	−1.67	0.95			✗
		100	1.20	−1.08	1.53	1.44	1.90	−1.08	1.13	1.03
		300	1.17	1.36	13.13	2.81	3.49	1.35	−1.04	3.33
	Tienilic acid	30	1.85	1.51	4.43	3.93	3.50	1.50	1.76	2.64			✓
		100	2.49	1.44	7.78	4.65	5.72	1.43	2.28	3.68
		300	1.86	1.80	5.33	3.07	7.34	1.79	1.72	3.27
Negative compounds	Caffeine	30	1.01	1.00	4.79	1.15	1.66	1.18	−1.14	1.54	1.44		✓
		100	1.16	−1.01	16.51	2.71	1.77	−1.01	−1.08	3.16
		300	−1.64	−1.15	−1.18	1.25	1.31	−1.15	−2.08	0.37
	Levofloxacin	30	1.49	1.18	9.06	4.48	3.52	1.16	1.48	3.20			✓
		100	−1.16	1.09	5.74	3.51	3.08	1.07	−1.59	2.07
		300	−1.15	−1.09	3.45	2.66	3.48	−1.11	−1.45	1.37
	Furosemide	30	−1.28	−2.94	1.45	−1.11	−1.56	−3.03	−1.79	0.21			✓
		100	1.01	−4.35	7.19	1.04	1.40	−4.00	−1.15	1.52
		300	5.10	1.87	−3.45	3.08	−2.04	1.25	4.54	2.26
	Aspirin	30	2.12	−1.04	−1.79	−2.33	−2.22	−1.06	1.73	0.55			✓
		100	1.86	−1.03	1.93	−3.13	−2.27	−1.05	1.57	0.77
		300	−1.03	−2.33	1.87	−4.76	−2.04	−2.33	−1.56	0.27

Primary human hepatocytes of donor 2 (Hu1389) were exposed to five drugs known to induce RM formation and four non-RM-forming drugs in human and BHA, a technical positive control, at three different concentrations (30, 100, and 300 μM) in triplicate. Seven genes under the regulation of Nrf2 (i.e., NOQ1, CES1, GSTA1, HMOX1, GCLC, GSTP1, and UGT1A1) were investigated and normalized versus the β-actin housekeeping gene. An index, representing the sum of the positive changes divided by the number of genes, was calculated for each concentration. To determine the donor-specific threshold, the mean index for the RM-forming (5.60) and the non-RM-forming compounds (1.44) was calculated. Subsequently, the midpoint of these two values was taken to obtain the donor-specific threshold. For donor 2, the threshold was 3.52 and the compounds were classified accordingly; for RM forming compounds, if an index above 3.52 was observed for one of the triplicates, the compound was correctly classified; when the index was below 3.52 for all triplicates, the compound was noncorrectly classified. For non-RM-forming compounds with an index below 3.52 for all triplicates, the compound was correctly classified; above 3.52, the compound was noncorrectly classified, 3.52-fold inductions are colored in gray shade.

Table 6.

Gene Expression Analysis of Donor 3 (Hu1198)

Genes			NQO1	CES1	HMOX1	GCLC	GSTP1	UGT1A1	Index
	Control	μM	1	1	1	1	1	1	1.00	Mean index	Midpoint of the mean	Donor-specific threshold of 2.27
RM-forming compounds	BHA (pos ctrl)	30	1.55	−1.55	18.32	1.98	2.07	1.38	4.22	Valid assay
		100	1.35	−2.09	3.22	−1.08	−1.13	1.88	1.07
		300	−1.70	−7.70	2.00	−1.53	1.40	2.05	0.91
	Ticlopidine	30	1.59	−26.20	3.46	−1.63	1.81	1.57	1.41	3.01	2.27	✓
		100	−1.17	−146.19	20.28	1.60	9.48	1.60	5.49
		300	ND	ND	ND	ND	ND	ND	ND
	Clozapine	30	1.28	−8.37	4.82	1.49	2.61	1.11	1.88			✓
		100	2.71	ND	9.32	5.70	14.40	1.33	5.58
		300	ND	ND	ND	ND	ND	ND	ND
	Diclofenac	30	3.06	1.36	3.54	1.95	3.68	1.82	2.57			✓
		100	1.71	−4.38	5.84	1.26	5.11	3.39	2.88
		300	−1.46	4.00	46.05	11.90	4.66	2.06	11.45
	Acetaminophen	30	1.26	1.06	1.14	−1.12	2.09	−2.32	0.92			✗
		100	−1.11	−1.34	1.26	−1.10	1.26	−1.75	0.42
		300	−1.25	−2.65	1.02	−1.60	1.60	−1.28	0.44
	Tienilic acid	30	1.12	−1.42	1.83	1.32	1.48	1.42	1.19			✓
		100	2.02	−1.32	4.64	−1.11	2.41	3.00	2.01
		300	1.64	−1.02	9.18	1.19	2.06	3.56	2.94
Non-RM-forming compounds	Caffeine	30	3.24	−2.10	2.45	2.35	3.47	1.16	2.11	1.53		✓
		100	1.76	−1.66	1.63	2.08	1.69	1.06	1.37
		300	3.39	1.03	1.71	2.08	2.99	1.33	2.09
	Levofloxacin	30	1.82	−3.32	3.25	2.60	1.73	−1.18	1.57			✓
		100	2.27	−1.52	4.00	−1.45	1.25	−1.17	1.25
		300	2.09	−2.60	1.85	1.73	1.75	−1.23	1.24
	Furosemide	30	1.68	−3.08	2.72	−1.15	1.10	−1.52	0.92			✓
		100	1.70	−2.82	2.64	1.58	1.20	−1.40	1.19
		300	1.55	−1.55	4.28	1.02	−2.27	1.13	1.33
	Aspirin	30	1.76	−1.82	8.33	2.27	−2.04	1.24	2.27			✗
		100	−1.24	−3.08	1.60	1.08	1.43	−1.59	0.68
		300	−1.20	−1.36	10.07	2.31	1.45	−1.21	2.31

Primary human hepatocytes of donor 3 (Hu1198) were exposed to five drugs known to induce RM formation and four non-RM-forming drugs in human and BHA, a technical positive control, at three different concentrations (30, 100, and 300 μM) in triplicate. Seven genes under the regulation of Nrf2 (i.e., NOQ1, CES1, GSTA1, HMOX1, GCLC, GSTP1, and UGT1A1) were investigated and normalized versus the β-actin housekeeping gene. An index, representing the sum of the positive changes divided by the number of genes, was calculated for each concentration. To determine the donor-specific threshold, the mean index for the RM forming (3.01) and the non-RM-forming compounds (1.53) was calculated. Subsequently, the midpoint of these two values was taken to obtain the donor-specific threshold. For donor 3, the threshold was 2.27 and the compounds were classified accordingly; for RM forming compounds, if an index above 2.27 was observed for one of the triplicates, the compound was correctly classified; when the index was below 2.27 for all triplicates, the compound was noncorrectly classified. For non-RM-forming compounds with an index below 2.27 for all triplicates, the compound was correctly classified; above 2.27, the compound was noncorrectly classified, 2.27-fold inductions are colored in gray shade.

Table 7.

Gene Expression Analysis of HepG2 Cells

Genes			NQO1	CES1	GSTA1	HO-1	GCLC	GSTP1	UGT1A1	Index
	Control	μM	1	1	1	1	1	1	1	1.00	Mean index	Midpoint of the mean	HepG2-specific threshold of 2.28
RM-forming compounds	BHA (pos ctrl)	30	1.76	−1.16	2.59	1.72	12.51	−1.62	−4.54	2.65	Valid assay
		100	2.68	−1.27	6.89	2.69	20.99	−2.08	−2.83	4.75
		300	3.53	−1.33	2.09	1.77	20.32	−1.10	−1.95	3.96
	Ticlopidine	30	−1.04	−1.35	3.46	1.98	2.25	−1.25	−3.83	1.10	2.17	2.28	✓
		100	1.81	−1.33	−1.03	2.38	10.52	5.07	−1.51	2.83
		300	−3.10	−1.64	−4.86	−2.14	31.96	ND	ND	4.57
	Clozapine	30	1.65	1.05	2.83	2.57	6.84	2.64	−4.76	2.51			✓
		100	3.33	2.16	2.16	6.13	ND	5.76	−1.19	2.79
		300	ND	ND	ND	ND	ND	ND	ND	ND
	Diclofenac	30	1.09	−1.08	5.25	2.73	5.45	−1.27	−6.27	2.08			✓
		100	−1.01	−1.29	2.23	2.21	6.61	−1.22	−4.82	1.58
		300	−1.34	−1.05	2.59	3.43	9.48	1.96	−3.72	2.49
	Acetaminophen	30	−1.02	−1.27	2.83	1.59	5.69	−1.02	−7.42	1.44			✗
		100	−1.26	−1.21	3.10	1.86	3.01	1.16	−7.29	1.30
		300	−1.14	−1.13	2.33	1.56	3.24	1.11	−6.80	1.18
	Tienilic acid	30	−1.67	−1.12	6.73	2.04	2.83	−1.39	−6.66	1.66			✓
		100	1.39	−1.07	5.05	2.17	4.93	−1.84	−6.36	1.93
		300	1.83	1.00	4.80	4.55	8.78	−1.10	−4.83	2.99
Non-RM-forming compounds	Caffeine	30	1.74	−1.30	4.32	2.18	12.34	−1.08	−6.34	2.94	2.39		✗
		100	1.42	−1.22	2.89	2.37	8.33	1.22	−5.22	2.32
		300	1.66	−1.52	3.40	2.04	17.92	−1.87	−4.15	3.57
	Levofloxacin	30	1.55	−1.15	4.24	2.17	9.21	1.22	−5.86	2.63			✗
		100	1.65	−1.21	5.77	2.32	7.87	−1.39	−5.28	2.52
		300	1.13	−1.18	1.78	2.38	5.14	1.14	−4.77	1.65
	Furosemide	30	1.15	−1.23	4.98	2.28	4.96	1.34	−6.23	2.10			✗
		100	1.33	−1.15	1.59	2.28	3.20	1.31	−7.20	1.39
		300	1.08	−1.29	6.00	2.46	7.84	−1.57	−7.17	2.48
	Aspirin	30	1.05	−1.11	3.49	2.08	7.95	1.11	−5.59	2.24			✗
		100	1.04	−1.25	4.85	2.12	8.95	1.05	−5.06	2.57
		300	1.35	−1.36	2.42	1.66	10.07	−1.08	−3.69	2.21

HepG2 cells were exposed to five drugs known to induce RM formation and four non-RM-forming drugs in human and BHA, a technical positive control, at three different concentrations (30, 100, and 300 μM) in triplicate. Seven genes under the regulation of Nrf2 (i.e., NOQ1, CES1, GSTA1, HO1, GCLC, GSTP1, and UGT1A1) were investigated and normalized versus the β-actin housekeeping gene. An index, representing the sum of the positive changes divided by the number of genes, was calculated for each concentration. To determine the HepG2 batch-specific threshold, the mean index for the RM-forming- and the non-RM-forming compounds was calculated. Subsequently, the midpoint of these two values was taken to obtain the donor-specific threshold. The threshold was 2.28 and the compounds were classified accordingly; for RM-forming compounds, if an index above 2.28 was observed for one of the triplicates, the compound was correctly classified; when the index was below 2.28 for all triplicates, the compound was noncorrectly classified. For non-RM-forming compounds with an index below 2.28 for all triplicates, the compound was correctly classified; above 2.28, the compound was noncorrectly classified, 2.28-fold inductions are colored in gray shade.

A donor-specific threshold was calculated for each donor. This donor-specific threshold is determined by the midpoint between the mean index for the RM-forming and the non-RM-forming compounds. For donor 1, the donor-specific threshold was 1.08. Four out of five RM = forming compounds were correctly classified, except acetaminophen. Three out of four non-RM-forming compounds were correctly classified. Furosemide was classified as false positive (Table 4). For donor 2, the donor-specific threshold was 3.52; four out of five RM-forming compounds were correctly classified, except acetaminophen. All four non-RM-forming compounds were correctly classified (Table 5). For donor 3, the donor-specific threshold was 2.27. Acetaminophen and aspirin were not correctly classified as an RM-forming compound and non-RM-forming compound, respectively (Table 6). Interestingly, for the three donors, the best predictivity was obtained with donor 2 associated with the highest midpoint (i.e., 3.52). For HepG2 cells, the threshold was calculated to be 2.28. Four out of five compounds were correctly classified, except acetaminophen, but none of the non-RM-forming compounds was correctly classified (Table 7). In each of the PHH donors and in the HepG2 cells, the technical positive control was classified as an RM-forming compound and consequently, the assay was considered valid. Overall, specificity varied in the three PHH donors, ranging from 75% up to 100%, but the sensitivity was equal for all three donors (80%). Acetaminophen was not correctly identified in any of the donors. The attained sensitivity in HepG2 cells was 80%, but specificity was very poor (0%); none of the compounds could be correctly identified (Table 8).

Table 8.

Overview of Sensitivity, Specificity and Predictivity

	Experiment 1				Experiment 2
	PHH donor 1	PHH donor 2	PHH donor 3	HepG2	AREC32
Sensitivity	4/5	4/5	4/5	4/5	14/18
Sensitivity	80%	80%	80%	80%	78%
Specificity	3/4	4/4	3/4	0/4	7/8
Specificity	75%	100%	75%	0%	88%
Predictivity	7/9	8/9	7/9	4/9	22/27
Predictivity	78%	89%	78%	44%	82%

Displayed are the percentages sensitivity, specificity, and predictivity obtained in the gene expression assay and for the Nrf2 activation in the reporter cell line AREC32.

HEPG2, human hepatocellular carcinoma; PHH, primary human hepatocytes.

The responsiveness toward the seven individual genes varied among the donors. In donor 1, the most inducible gene was HO-1 for the RM-forming compounds and UGT1A1 was downregulated in the PHH exposed to the non-RM-forming compounds (Table 4). In donor 2, HO-1 was also the most inducible gene, together with GSTA-1 for the RM-forming compounds. For the non-RM-forming compounds, besides the GSTA1 gene, no other gene reacted (Table 5). In donor 3, HO-1 gene was highly induced not only in the cells exposed to four out of five RM-forming compounds but also in the cells exposed to non-RM-forming compounds (Table 6). In HepG2 cells, a distinction between RM-forming and non-RM-forming compounds could not be made on the basis of gene expression as most of the genes are going in the same direction independent of the compounds used, that is, genes HO-1, GSTA1, and GCLC were induced, while the UGT1A1 gene was strongly repressed after exposure to all compounds (Table 7).

Experiment 2: Evaluation of drug effects in the AREc32 reporter assay

AREc32 cells were incubated with the 10 compounds previously described, as well as with tert-butylhydroquinone (t-BHQ), a prototypical Nrf2 activator used as a technical positive control. Five non-RM-forming compounds and another set of 13 RM-forming compounds were also added in this experiment. Results are displayed in Table 9. t-BHQ increased the luminescence signal in a concentration-dependent manner. The response was considered to be strong with fold induction ratios ranging from 3- to 36-fold in the range between 3.125 and 25 μM. In total, 14 out of the 18 RM-forming compounds were correctly classified (>1.5-fold increase), as well as 8 out of 9 non-RM-forming compounds (<1.5-fold increase) (Table 9). Concentrations responses could be observed for most compounds, although the responses were not as strong as the one obtained with t-BHQ. Nevertheless, the highest responses were obtained for clozapine, furazolidone, and tacrine with fold induction ratios in the range threefold to fivefold at 50 and/or 100 μM. Acetaminophen, flutamide, clofibrate, and aflatoxin B1 were not detected as positives in the assay. Flutamide could be classified as equivocal as their fold induction was very close to the threshold value at the highest concentration. The LOEC/Cmax ratio has been calculated for all compounds; the lowest ratios have been obtained for amiodarone, tienilic acid, and ketoconazole (respectively 0.20, 0.29, and 1.89) (Table 9). The highest ratios were obtained for labetalol, imipramine, tacrine, and danazol (respectively 30.48, 39.68, 44.64, and 54.35) (Table 9).

Table 9.

The Antioxidant-Response-Element-Reporter-Cell Line, AREC32, Results

			Concentration (μM)
Compound	Cmax (μM)	Category	3.125	6.25	12.5	25	50	100	Score	LOEC/Cmax
BHA		Positive controls	1.67	2.66	1.20	1.06	0.94	0.84	Valid
t-BHQ		Positive controls	3.01	8.58	16.03	36.05	29.58	24.82	Valid
Ticlopidine	7.09	RM-forming compound	1.03	1.11	1.27	1.28	1.76	2.85	✓	7.05
Clozapine	2.36		1.05	1.03	1.24	1.42	2.42	4.42	✓	21.19
Diclofenac	6.29		1.08	1.14	1.07	1.18	1.44	1.53	✓	15.90
Acetaminophen	138.93		0.98	1.11	0.99	1.11	1.09	1.14	✗	NA
Tienilic acid	172.12		0.92	1.28	1.36	0.89	1.72	2.00	✓	0.29
Amiodarone	60.14		1.34	1.46	1.73	1.95	NC	NC	✓	0.21
Danazol	0.46		0.43	1.34	1.36	1.53	2.00	1.97	✓	54.35
Furazolidone	4		1.09	1.29	1.61	1.28	1.41	3.72	✓	3.13
Flutamide	6.22		0.96	1.06	1.07	1.10	1.09	1.42	✗	NA
Labetalol	0.41		1.16	1.37	1.68	1.48	1.70	2.37	✓	30.49
Ketoconazole	6.59		0.97	1.18	1.52	1.90	1.95	NC	✓	1.90
Chlorpromazine	0.5		1.00	1.53	2.17	NC	NC	NC	✓	12.50
Cyclosporin A	1.26		0.87	1.07	1.01	1.61	2.49	2.51	✓	19.84
Imipramine	0.63		0.98	0.97	1.00	1.67	2.90	NC	✓	39.68
Tamoxifen	0.51		1.04	1.52	2.45	NC	NC	NC	✓	12.25
Clofibrate	235.27		1.08	1.09	1.02	1.17	1.06	1.05	✗	NA
Tacrine	0.14		1.40	1.65	1.88	3.29	4.98	NC	✓	44.64
Aflatoxin B1	NA		0.96	0.94	0.74	0.79	0.65	0.76	✗	NA
Flufenamic acid	46	Non-RM- forming compound	1.15	1.18	1.08	1.24	1.59	1.93	✗	1.09
Betaine hydrochloride	940.04		1.03	1.05	1.01	1.14	1.05	1.08	✓	NA
Isoproterenol	4.04		1.23	1.35	1.31	1.39	1.26	1.49	✓	NA
Praziquantel	1.8		1.10	1.22	1.20	1.29	1.24	1.31	✓	NA
Primidone	54.98		1.06	1.05	0.93	1.03	0.90	0.83	✓	NA
Aspirin	1650		1.05	1.18	1.07	1.14	0.99	1.10	✓	NA
Caffeine	42		1.18	1.12	1.11	1.15	1.09	1.20	✓	NA
Levofloxacin	29.63		0.94	1.00	0.98	1.05	0.89	0.91	✓	NA
Furosemide	16		1.01	0.92	1.06	1.07	1.04	1.05	✓	NA

The Antioxidant Response Element Reporter cell line AREC32 was exposed to 18 RM-forming compounds, 9 non-RM compounds, and 2 technical positive controls. The fold induction of each compound over the vehicle control at each concentration is listed, as well as the LOEC over the Cmax ratio. The experiment has been performed in triplicate for the 10 compounds used in approach 1 and in duplicate for the remaining compounds, but for clarity, only one of the replicates is shown, 1.5-fold inductions or higher are colored in dark gray shade and below 1.5-fold is colored in white.

LOEC, lowest effective concentration; NC, not tested due to cytotoxic effects.

Discussion

DILI is known as a multistep process, which is initiated by a chemical assault on the liver cells. There are multiple processes underlying DILI, including, for instance, RM formation,² mitochondria impairment,¹⁰ and inhibition of bile transport activities.²² Recently, Atienzar et al.² proposed to apply de-risking strategies early in drug development to reduce drug attritions at preclinical and clinical stages. The immune system has also been identified as an important component in the occurrence of DILI.¹⁰ Beside all these mechanisms, genetics can also play an important role, particularly for idiosyncratic reactions.^3,23 Chemical assault may trigger a biological response, which can either elicit protective or detrimental processes to the liver. There are several stress response pathways such as nuclear factor-like 2 (NFE2L2 or Nrf2), NFE2L1 (Nrf1), p53, heat shock factor, and the unfolded protein response following exposure to different types of cell stress.²⁴ For instance, Nrf2 is a redox-specific transcription factor that is activated upon oxidative/electrophilic stress that will subsequently lead to an orchestrated antioxidant and anti-inflammatory response. By inducing a large variety of cytoprotective genes involved in detoxification, Nrf2 is able to protect against oxidative/electrophilic stress-related injuries and inflammatory diseases.¹⁴ When these protective processes are depleted, electrophilic stress and inflammatory processes can lead to organ injury.

It has to be noted that Nrf2 is also regulated by compounds such as Vitamin A.²⁵ Nevertheless, the exact mechanism of Nrf2 activation by vitamin A is not known. Despite the well-known regulation of Nrf2 by Keap1, there is also evidence that Nrf2 can be regulated independent of Keap1. A study by Li et al.²⁶ showed that the Nrf2 inducer sulforaphane prevents the dissociation of Nrf2 from Keap1, supporting the hypothesis that there are alternative mechanisms of Nrf2 activation that do not rely uniquely on dissociation from Keap1.²⁷ Other mechanisms of Nrf2 activation involve phosphorylation of Nrf2 by signal transduction pathways, involvement of epigenetic factors (e.g., miRNAs), or interaction of Nrf2 with other proteins.²⁷

In this study, we ought to determine whether Nrf2-mediated gene expression could be used to distinguish between RM-forming and non-RM-forming compounds in PHH and in HepG2 cells. Nrf2 activation has the potential to detect compounds inducing oxidative and electrophilic stress in any organ. Nevertheless, since DILI has been widely studied, it was decided to focus on DILI compounds known to cause toxicity through RM formation. In addition, we investigated if the luminescence reporter cell line AREc32 is able to correctly classify RM-forming from non-RM-forming compounds. We previously showed the utility of gene expression analysis to detect events such as phospholipidosis²⁸ and hepatotoxicity.^19,29 In the literature, studies on Nrf2-mediated induction have been carried out in HepG2 cells, PHH (PHH), and rodent hepatocytes.^30–32

The work presented in this study is novel in several ways. First, we investigated PHH donor variability, whereas most published studies used only a single donor (e.g., Takakusa et al.¹⁶). Second, we made the direct comparison between primary cells and the hepatocarcinoma cell line HepG2, which is widely used in toxicological studies within the pharmaceutical industry. Third, a set of seven genes, under the regulation of Nrf2, was investigated in all models for all compounds, while Takakusa et al.¹⁶ only used the four most promising genes. Fourth, the toxicogenomic approach allows to calculate a donor-specific threshold that takes into account the gene expression variability following exposure to negative and positive compounds, whereas most approaches used in the literature use fixed and arbitrary cutoffs. Finally, this article also allows to compare gene expression data with the AREc32 reporter cell line data.

Nrf2 activation can be used to flag compounds as potentially hazardous, but has to be interpreted with other toxicity data to better assess the risk. This assay can be used in early drug discovery to identify potential hazards of compounds triggering electrophilic and oxidative stress. In this study, the concentrations used in both experiments were fixed (30–300 and 3–100 μM for experiments 1 and 2, respectively). Therapeutic information (e.g., Cmax values) was known for most tested compounds and in the majority of cases, the Cmax values were below the tested concentration. For some compounds, the total portal vein concentration can be significantly higher than Cmax. Nevertheless, since this information is rarely available, it was decided to use Cmax data to move from hazard identification to risk assessment with the calculation of safety margins. This exercise has been performed with the AREC32 data by calculating the LOEC over Cmax ratio for all compounds. Ratios below and close to 1 indicate a very high risk of inducing electrophilic and oxidative stress. Higher ratios mean that the risk of electrophilic stress is much lower. In this article, the compounds associated with the lowest LOEC/Cmax ratios are amiodarone, tienilic acid, and ketoconazole. The compounds associated with the highest LOEC/Cmax ratio are labetalol, imipramine, tacrine, and danazol.

In conclusion, the seven compounds (i.e., amiodarone, tienilic acid, ketoconazole, labetalol, imipramine, tacrine, and danazol) have all the potential to induce electrophilic and oxidative stress (hazard identification). Nevertheless, in terms of risk assessment, the last four compounds carry a lower risk due to a higher safety margin compared to amiodarone, tienilic acid, and ketoconazole.

Previously, Takakusa et al.¹⁶ identified HO-1 gene induction in human hepatocytes as a relevant marker of DILI secondary to RM formation. However, measurement of a single gene and/or use of a single donor can lead to misinterpretations. For that reason, in this study, an index was calculated taking into account the fold changes observed for each gene divided by the number of genes tested (i.e., seven in total). This approach has previously been used successfully, to detect phospholipidosis.²⁸ To determine the cutoff, a donor-specific threshold value was calculated for each donor and the HepG2 cells based on the results obtained with a set of RM-forming and non- RM-forming compounds. In our study, in all three PHH donors, the HO-1 gene was upregulated in cells exposed to four out of five RM-forming compounds, except acetaminophen for which no upregulation or downregulation could be observed. However, the HO-1 gene was also upregulated in donor 3 exposed to non-RM-forming compounds. By calculating a donor-specific threshold, data can be compared among different donors and unknown compounds can be assessed. This implies that preliminary testing needs to be performed for each new donor using a set of positive and negative compounds to calculate the donor-specific threshold.

The choice of a classification criterion (i.e., cutoff) may greatly influence the results, particularly for the gene expression data due to the well-known existing variability in primary human hepatocyte donors. If a fixed cutoff had been used, instead of the donor-specific cutoff, lower sensitivity and specificity data were calculated (data not shown).

When applying the donor-specific threshold, sensitivity was 80% for each of the three donors and the HepG2 cells. Specificity ranged between 75% and 100% for the different donors, but was 0% for the HepG2 cells. Variability among donors has been widely reported in the literature. Our data are in agreement with other studies revealing variable gene expression responses when using different PHH donors.^19,33 There are several reasons that can explain this donor variability such as (1) alterations in metabolic activities, (2) polymorphisms in specific genes or gene regulators, (3) altered uptake of the compounds, and/or (4) epigenetic differences due to cell preparations.^34–36 Most of them are difficult to analyze in the PHH donor preparations. However, in Table 2, the CYP450 activities are shown for the different donors; small differences can be seen between the different donors, but these data alone do not seem to explain the variability observed.

Overall, the transcriptomic approach was satisfactory to distinguish RM-forming from non-RM-forming compounds in PHH when the donor-specific threshold approach was used to rule out the variability among the different donors. However, this approach was not applicable to HepG2 cells as specificity was very low. One of the reasons for the low predictivity of the HepG2 cells could be due to the low metabolic capacity compared to PHH and/or an altered Nrf2 response in these cells. Low specificity values are of particular concern at early stage of drug development because potential promising compounds could be filtered out due to false positive signals. Overall, based on the limited set of compounds used, it seems that the toxicogenomic approach is more favorable when using PHH.

In the second approach, the ARE Reporter Cell Line, AREc32, was used to screen compounds inducing the formation of RM as well as negative compounds. In this cellular model, Nrf2-mediated activation of ARE was measured by a luciferase reporter transgene. In total, 27 compounds were tested. The sensitivity and specificity obtained were 78% (14/18) and 88% (7/8), respectively. Overall, we concluded that the AREc32 assay seemed to be suitable for the identification of compounds that induce the formation of RM. Overall, the assay combined a good sensitivity (i.e., 78%) with an excellent specificity (i.e., 88%) based on the criteria defined by the ECVAM Management Team.³⁷ The somewhat lower sensitivity can be explained by a number of reasons. Acetaminophen was not detected in the assay. This was likely due to the low concentration tested (300 μM) in relation to the clinical hepatotoxic plasma exposure. It is known that acetaminophen, which is a highly prescribed drug, is a safe drug at a normal therapeutic dose (Cmax of 139 μM, maximum 4 g/day). However, when used at a high dose (7.5–10 g/day), it can cause hepatotoxicity due to the formation of RMs.³⁸ Aflatoxin B1 could not be detected in the AREC assay. This could be due to the fact that the CYP3A4 activity in AREC32 cells may be too low to metabolize the compound to generate RM. The same applies to flutamide for which the CYP1A2 enzyme is necessary.³⁹

The addition of S9 to the cells did not allow detecting any of these compounds (data not shown). The AREC32 cell model has proven its robustness in other studies as well, for example, Wu et al.¹⁹ successfully built a high-throughput assay using these cells. They tested more than 47,000 compounds and concluded that this assay is suitable to identify Nrf2 activators and provides novel insights into chemical scaffold that might prevent oxidative/electrophilic stress-induced toxicity. In another study of Wu et al.,⁴⁰ 54 natural dietary compounds were screened to check which natural dietary compounds activate the Keap1-Nrf2 pathway. The assay also proved to be useful for the detection of skin sensitizers. Indeed, Natsch and Emter⁴¹ tested 102 chemicals using the reporter cell line. When applying a threshold of 1.5, they obtained an overall accuracy of 83%, with a sensitivity and specificity of 81% and 87%, respectively.

Conclusion

Two different approaches have been tested to distinguish DILI from non-DILI compounds based on their capacity to form RM or not. This study showed that the transcriptomic approach suffered from donor variability that could be circumvented by the implementation of a donor-specific threshold. Overall, gene expression measured in PHH could be used to predict RM formation, whereas HepG2 cells displayed similar sensitivity, but very low specificity, and consequently were less suitable to predict RM formation. Finally, the AREC32 assay had shown good predictive values³⁷ and was reproducible and easy to use. Consequently, the AREC32 cell assay seems to be the most suited assay to investigate Nrf2 activation to detect compounds inducing electrophilic and oxidative stress at early stages of drug development.

Footnotes

Acknowledgments

The authors would like to thank Annelies Eerdekens and Gaëlle Toussaint for technical assistance and Etienne Hanon for the calculation of the donor-specific thresholds.

Author Disclosure Statement

No competing financial interests exist.

References

Waring

, Arrowsmith

, Leach

, et al. An analysis of the attrition of drug candidates from four major pharmaceutical companies. Nat Rev Drug Discov, 2015:14; 475–486.

Atienzar

, Delaunois

, Brouta

, et al. Investigative safety strategies to improve success in drug development. J Med Dev Sci, 2016:2; 1–28.

Atienzar

, Blomme

, Chen

, et al. Key challenges and opportunities associated with the use of in vitro models to detect human DILI: Integrated risk assessment and mitigation plans. Biomed Res Int, 2016:2016; 9737920.

Attia

. Deleterious effects of reactive metabolites. Oxid Med Cell Longev, 2010:3; 238–253.

Acikgöz

, Giri

, Bader

. Detection of nanolevel drug metabolites in an organotypic culture of primary human hepatocytes and porcine hepatocytes with special reference to a two-compartment model. Int J Nanomedicine, 2012:7; 5859–5872.

Fura

. Role of pharmacologically active metabolites in drug discovery and development. Drug Discov Today, 2006:11; 133–142.

Ramírez

, Vázquez-Sánchez

, Carrión-Robalino

, et al. Ion channels and oxidative stress as a potential link for the diagnosis or treatment of liver diseases. Oxid Med Cell Longev, 2016:2016; 3928714.

Russmann

, Kullak-Ublick

, Grattagliano

. Current concepts of mechanisms in drug-induced hepatotoxicity. Curr Med Chem, 2009:16; 3041–3053.

Park

, Boobis

, Clarke

, et al. Managing the challenge of chemically reactive metabolites in drug development. Nat Rev Drug Discov, 2011:10; 292–306.

10.

Stephens

, Andrade

, Lucena

. Mechanisms of drug-induced liver injury. Curr Opin Allergy Clin Immunol, 2014:14; 286–292.

11.

Fulda

, Gorman

, Hori

, et al. Cellular stress responses: Cell survival and cell death. Int J Cell Biol, 2010:2010; 214074.

12.

Singh

, Vrishni

, Singh

, et al. Nrf2-ARE stress response mechanism: A control point in oxidative stress-mediated dysfunctions and chronic inflammatory diseases. Free Radic Res, 2010:44; 1267–1288.

13.

Suzuki

, Otsuki

, Keleku-Lukwete

, et al. Overview of redox regulation by Keap1–Nrf2 system in toxicology and cancer. Curr Opin Toxicol, 2016:1; 29–36.

14.

Keum

, Choi

. Molecular and chemical regulation of the keap1-nrf2 signaling pathway. Molecules, 2014:19; 10074–10089.

15.

Gerets

, Hanon

, Cornet

, et al. Selection of cytotoxicity markers for the screening of new chemical entities in a pharmaceutical context: A preliminary study using a multiplexing approach. Toxicol In Vitro, 2009:23; 319–332.

16.

Takakusa

, Masumoto

, Mitsuru

, et al. Markers of electrophilic stress caused by chemically reactive metabolites in human hepatocytes. Drug Metab Dispos, 2008:36; 816–823.

17.

Gerets

HHJ

, Tilmant

, Gerin

, 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, 2012:28; 69–87.

18.

Wang

, Hayes

, Wolf

. Generation of a stable antioxidant response element-driver reporter gene cell line and its use to show redox-dependent activation of Nrf2 by cancer chemotherapeutic agents. Cancer Res, 2006:66; 10983–10994.

19.

, McDonald

, Liu

, et al. Implementation of a high-throughput screen for identifying small molecules to activate the Keap1-Nrf2-ARE pathway. PLoS One, 2012:7; 1–10.

20.

Atienzar

, Tilmant

, Gerets

, et al. The use of real-time cell analyzer technology in drug discovery: Defining optimal cell culture conditions and assay reproducibility with different adherent cellular models. J Biomol Screen, 2011:16; 575–587.

21.

Atienzar

, Gerets

, Tilmant

, et al. Evaluation of impedance-based label-free technology as a tool for pharmacology and toxicology investigations. Biosensors, 2013:3; 132–156.

22.

Wolf

, Vora

, Webster

, et al. Use of cassette dosing in sandwich-cultured rat and human hepatocytes to identify drugs that inhibit bile acid transport. Toxicol In Vitro, 2010:24; 297–309.

23.

Stephens

, Lucena

, Andrade

. Genetic variations in drug-induced liver injury (DILI): Resolving the puzzle. Front Genet, 2012:3; 253.

24.

Jennings

, Limonciel

, Felice

, et al. An overview of transcriptional regulation in response to toxicological insult. Arch Toxicol, 2013:87; 49–72.

25.

Wang

, Xiu

, Li

, et al. Vitamin A supplementation alleviates extrahepatic cholestasis liver injury through Nrf2 activation. Oxid Med Cell Longev, 2014:2014; 273692.

26.

, Paonessa

, Zhang

. Mechanism of chemical activation of Nrf2. PLoS One, 2012:7; e35122.

27.

Bryan

, Olayanju

, Goldring

, et al. The Nrf2 cell defence pathway: Keap1-dependent and -independent mechanisms of regulation. Biochem Pharmacol, 2013:85; 705–717.

28.

Atienzar

, Gerets

, Dufrane

, et al. Determination of phospholipidosis potential based on gene expression analysis in HepG2 cells. Toxicol Sci, 2007:96; 101–114.

29.

de Longueville

, Atienzar

, Marcq

, et al. Use of a low-density microarray for studying gene expression patterns induced by hepatotoxicants on primary cultures of rat hepatocytes. Toxicol Sci, 2003:75; 378–392.

30.

Cantoni

, Valaperta

, Ponsoda

, et al. Induction of hepatic heme oxygenase-1 by diclofenac in rodents: Role of oxidative stress and cytochrome P-450 activity. J Hepatol, 2003:38; 776–783.

31.

Shinkai

, Sumi

, Fukami

, et al. Sulforaphane, an activator of Nrf2, suppresses cellular accumulation of arsenic and its cytotoxicity in primary mouse hepatocytes. FEBS Lett, 2006:580; 1771–1774.

32.

Kim

, Jang

. Nrf2-mediated HO-1 induction coupled with the ERK signaling pathway contributes to indirect antioxidant capacity of caffeic acid phenethyl ester in HepG2. Cells Int J Mol Sci, 2014:15; 12149–12165.

33.

Rogue

, Lambert

, Spire

, et al. Interindividual variability in gene expression profiles in human hepatocytes and comparison with HepaRG cells. Drug Metab Dispos, 2012:40; 151–158.

34.

Zanger

, Schwab

. Cytochrome P450 enzymes in drug metabolism: Regulation of gene expression, enzyme activities, and impact of genetic variation. Pharmacol Ther, 2013:138; 103–141.

35.

Jetten

MJA

, Ruiz-Aracama

, Coonen

MLJ

, et al. Interindividual variation in gene expression responses and metabolite formation in acetaminophen-exposed primary human hepatocytes. Arch Toxicol, 2016:90; 1103–1115.

36.

Park

, Choi

, Kim

, et al. Differences in the epigenetic regulation of cytochrome P450 genes between human embryonic stem cell-derived hepatocytes and primary hepatocytes. PLoS One, 2015:10; e0132992.

37.

Genschow

, Spielmann

, Scholz

, et al. The ECVAM international validation study on in vitro embryotoxicity tests: Results of the definitive phase and evaluation of prediction models. European Centre for the Validation of Alternative Methods. Altern Lab Anim, 2002:30; 151–176.

38.

Shin

, Yang

, Ki

. Role of the Nrf2-ARE pathway in liver diseases. Oxid Med Cell Longev, 2013:2013; 763257.

39.

Porceddu

, Buron

, Roussel

, et al. Prediction of liver injury induced by chemicals in human with a multiparametric assay on isolated mouse liver mitochondria. Toxicol Sci, 2012:129; 332–345.

40.

, McDonald

, Liu

, et al. Screening of natural compounds as activators of the keap1-nrf2 pathway. Planta Med, 2013:80; 97–104.

41.

Natsch

, Emter

. Skin sensitizers induce antioxidant response element dependent genes: Application to the in vitro testing of the sensitization potential of chemicals. Toxicol Sci, 2008:102; 110–119.

42.

Pessayre

, Fromentry

, Berson

, et al. Central role of mitochondria in drug-induced liver injury. Drug Metab Rev, 2012:44; 34–87.

43.

Amacher

. Serum transaminase elevations as indicators of hepatic injury following the administration of drugs. Regul Toxicol Pharmacol, 1998:27; 119–130.

44.

Figueira-Coelho

, Pereira

, Picado

, et al. Acute hepatitis associated with the use of levofloxacin. Clin Ther, 2010:32; 1733–1737.

45.

Bale

, Vernetti

, Senutovitch

, et al. In vitro platforms for evaluating liver toxicity. Exp Biol Med, 2014:239; 1180–1191.

46.

Vroomen

, Groten

, van Muiswinkel

, et al. Identification of a reactive intermediate of furazolidone formed by swine liver microsomes. Chem Biol Interact, 1987:64; 167–179.

47.

Zimmerman

. Drug-induced liver disease. Drugs, 1978:16; 25–45.

48.

, Henstock

, Dunn

, et al. Cellular imaging predictors of clinical drug-induced liver injury. Toxicol Sci, 2008:105; 97–105.

49.

Nguyen

NSD

, Cottet-Maire

, Buetler

, et al. Metabolism-dependent stimulation of reactive oxygen species and DNA synthesis by cyclosporine A in rat smooth muscle cells. Free Radic Biol Med, 1999:27; 1267–1275.

50.

, Li

, Wong

, et al. Mechanism of clofibrate hepatotoxicity: Mitochondrial damage and oxidative stress in hepatocytes. Free Radic Biol Med, 2001:31; 659–669.

51.

Dohnal

, Wu

, Kuca

. Metabolism of aflatoxins: Key enzymes and interindividual as well as interspecies differences. Arch Toxicol, 2014:88; 1635–1644.

52.

Chen

, Vijay

, Shi

, et al. FDA-approved drug labeling for the study of drug-induced liver injury. Drug Discov Today, 2011:16; 697–703.

53.

Luyendyk

, Copple

, Barton

, et al. Augmentation of aflatoxin B1 hepatotoxicity by endotoxin: Involvement of endothelium and the coagulation system. Toxicol Sci, 2003:72; 171–181.