toxicological sciences 137(1), 234–248 2014 doi:10.1093/toxsci/kft232 Advance Access publication October 17, 2013

Utilization of Causal Reasoning of Hepatic Gene Expression in Rats to Identify Molecular Pathways of Idiosyncratic Drug-Induced Liver Injury Daphna Laifenfeld,*,1,2 Luping Qiu,†,2 Rachel Swiss,† Jennifer Park,* Michael Macoritto,* Yvonne Will,† Husam S. Younis,‡ and Michael Lawton§  *Selventa Inc., Cambridge, Massachusetts 02140; †Compound Safety Prediction Group, Pfizer Worldwide Research and Development, Groton, Connecticut 06340; ‡ISIS Pharmaceuticals, Carlsbad, California 92010; and §Drug Safety Research and Development, Pfizer Worldwide Research and Development, Groton, Connecticut 06340 To whom correspondence should be addressed at Selventa Inc., 1 Alewife Centre, Cambridge, MA 02140. Fax: +972779520860. E-mail: [email protected]. 2 These authors contributed equally to this study. Received June 16, 2013; accepted September 12, 2013

Drug-induced liver injury (DILI) represents a leading cause of acute liver failure. Although DILI can be discovered in preclinical animal toxicology studies and/or early clinical trials, some human DILI reactions, termed idiosyncratic DILI (IDILI), are less predictable, occur in a small number of individuals, and do not follow a clear dose-response relationship. The emergence of IDILI poses a critical health challenge for patients and a financial challenge for the pharmaceutical industry. Understanding the cellular and molecular mechanisms underlying IDILI is key to the development of models that can assess potential IDILI risk. This study used Reverse Causal Reasoning (RCR), a method to assess activation of molecular signaling pathways, on gene expression data from rats treated with IDILI or pharmacologic/chemical comparators (NON-DILI) at the maximum tolerated dose to identify mechanistic pathways underlying IDILI. Detailed molecular networks involved in mitochondrial injury, inflammation, and endoplasmic reticulum (ER) stress were found in response to IDILI drugs but not negative controls (NON-DILI). In vitro assays assessing mitochondrial or ER function confirmed the effect of IDILI compounds on these systems. Together our work suggests that using gene expression data can aid in understanding mechanisms underlying IDILI and can guide in vitro screening for IDILI. Specifically, RCR should be considered for compounds that do not show evidence of DILI in preclinical animal studies positive for mitochondrial dysfunction and ER stress assays, especially when the therapeutic index toward projected human maximum drug plasma concentration is low. Key Words:  toxicogenomics; systems biology; biological modeling; risk assessment; safety evaluation; liver; systems toxicology.

Drug-induced liver injury (DILI) remains a major challenge for the pharmaceutical industry. It accounts for 13% of acute liver failure cases in the United States and has led to market withdrawals, prescribing restrictions, and black box warnings

(Chalasani et al., 2008). As a whole, DILI in humans has the poorest correlation with animal toxicity tests (Olson et  al., 1998). Drugs that produce DILI are often identified postapproval because they occur at very low incidence, which explains the poor predictivity of animal models. Although acute druginduced liver failure is rare (approximately 1 per 100,000), 13%–17% of all acute liver failure cases are attributed to idiosyncratic drug reactions (Hussaini and Farrington, 2007). To date, there is insufficient molecular understanding of IDILI and a corresponding lack in methods to systematically study idiosyncratic drug toxicity early in the drug development process. The mechanism(s) of idiosyncratic DILI (IDILI) are largely unknown, but multiple mechanisms have been implicated. For example, drug metabolism and metabolic polymorphisms, unique physiochemical properties, drug immunogenicity, environmental factors, and genetics may be inter- or codependent contributing factors in IDILI responses (Li, 2002; Ulrich, 2007). Specifically, drug-induced low-grade inflammation, endoplasmic reticulum (ER) stress, and mitochondrial damage may have more direct contributions. Inflammation has been postulated to play a role by reducing the threshold for liver injury (Deng et al., 2008; Kaplowitz, 2005; Roth et al., 2003). The sporadic occurrence of inflammation may not cause hepatotoxicity directly but instead may increase susceptibility to low-grade effects of xenobiotics via inflammatory cell activation and/or proinflammatory cytokine production. ER stress has been shown to play a significant role in liver injury caused by the well-studied hepatotoxicants carbon tetrachloride (Moore et al., 1976; Plaa, 2000) and acetaminophen (Nagy et al., 2010). The bioactivation of xenobiotics by metabolizing enzymes within the ER renders this organelle uniquely susceptible to a localized stress response that may serve as an underlying mechanism for IDILI. Drug-induced mitochondrial injury is highly implicated in IDILI and DILI as a whole. For example, mitochondrial

© The Author 2013. Published by Oxford University Press on behalf of the Society of Toxicology. All rights reserved. For permissions, please email: [email protected].

Downloaded from http://toxsci.oxfordjournals.org/ at University of Cambridge on November 17, 2014

1



MOLECULAR PATHWAYS IMPLICATED IN IDILI

Materials and Methods Compounds Compounds were purchased from Sigma (St Louis, Missouri) and USP (Rockville, Maryland), with the exception of PF194223, which was synthesized at Pfizer. Nine drugs that have been reported to cause IDILI in humans and lacked any reported evidence of liver injury in animals were selected as follows: trovafloxacin, leflunomide, flutamide, a Pfizer internal compound PF194223, glafenine, nimesulide, sulindac, tolcapone, and nefazodone. Eight compounds, generally considered to be nonhepatotoxic (NON-DILI) in both animals and humans, of the same therapeutic class as the selected IDILI compounds, were selected to serve as negative controls for chemistry and/or pharmacology and were as follows: levofloxacin, methotrexate, finasteride, rosiglitazone, metformin, valdecoxib, entacapone, and buspirone. A  list of the drugs tested, their DILI classification, the dose administered to rats (mg/ kg), and the human maximum plasma drug concentration (Cmax) is provided in Table 1. Causal Reasoning of Gene Expression Responses in Rats In vivo rat studies.  All experiments were performed in compliance with the USA Animals Scientific Procedures Act 1986. Six- to eight-week-old male Sprague Dawley rats (200–250 g) were purchased from Charles River Laboratories, Wilmington, Massachusetts. Animals were paired and housed in standard open cages with a 12-h light/12-h dark cycle. All rats were fed ad libitum a normal control diet (2014, 14% Teklad protein rodent maintenance diet; Harlan Laboratories). Animals (n = 3 per group) received a single dose of

compound or vehicle (Table 1). All compounds and their corresponding vehicle treatments were administered orally once a day except for trovafloxacin and methotrexate (and corresponding vehicle treatments), which were administered twice a day PO and IV, respectively. At 2 and 6 h posttreatment and at termination (24 h), blood was drawn, and alanine transaminase (ALT), aspartate aminotransferase (AST), glutamate dehydrogenase (GLDH), bilirubin, albumin, blood urea nitrogen, and total protein levels were assayed. At termination, sections of the right medial lobe from the liver of each animal were collected and flash frozen in liquid nitrogen for RNA isolation with subsequent gene expression analysis. Selection of molecular gene expression changes for analysis.  RNA expression data were generated using Affymetrix Rat Genome 230 2.0 expression microarrays and analyzed using the “affy” and “limma” packages of the Bioconductor suite of microarray analysis tools (Gentleman, 2005). RNA background correction and quantile normalization were used to generate microarray expression values. An overall linear model was fit to the data for all sample groups, and specific contrasts of interest were evaluated to generate raw p values for each probe set on the expression array. The Benjamini-Hochberg FDR method was then used to correct for multiple testing effects. RCR is optimally powered by data sets that have between 200 and 1000 significant gene expression changes, and criteria for gene expression changes considered significant are adjusted accordingly. Overall, gene expression changes were considered significant if they had an adjusted p value of .05 or less; average expression intensity was above 250 in at least 1 treatment group and a fold change greater than or equal to 1.3. To accommodate for an optimal number of State Changes to power RCR (200–1000), gene expression cutoff criteria were adjusted for some of the compounds: For glafenine, the criteria were made more stringent, and genes were considered changed if their average expression intensity was above 250, a fold change of at least or equal to 1.5, and a .01 or lower p value. For chlorpheniramine, the criteria were relaxed and genes were considered changed if their average expression intensity was above 150, and there was a fold change of 1.3 or greater with a .05 or lower p value. Knowledgebase and rat knowledge assembly model.  The substrate for the analysis of statistically significant gene expression changes observed in response to various compounds in rat livers compared with vehicle is the rat knowledge assembly model, which is derived from a global Selventa knowledgebase. This knowledgebase is a collection of biological concepts and entities and their causal relationships. It is derived from peer-reviewed scientific literature as well as other public and proprietary databases. The rat knowledge assembly model is the set of rat-specific causal assertions that has been augmented with orthologous causal assertions derived from either mouse or human sources and is competent for Causal Reasoning. An example for causal assertion would be an increased transcriptional activity of nuclear factor kappa B (NFkB) causing an increase in the expression of interferon regulatory factor 1 (Irf1). Each such causal assertion has a specific scientific citation, and the assembled collection of these causal assertions is referred to as the rat knowledge assembly model in this article. RCR.  Causal Reasoning interrogates the rat knowledge assembly model to identify upstream controllers for statistically significant gene expression changes observed in the experiment. These upstream controllers are called hypotheses, which are defined as statistically significant potential explanations of the gene expression changes. Hypothesis generation is performed automatically by a computer program that utilizes the rat knowledge assembly model to identify hypotheses that explain the input gene expression changes, prioritized by multiple statistical criteria. Each hypothesis is scored according to 2 probabilistic scoring metrics, richness and concordance, which examine distinct aspects of the probability of a hypothetical cause explaining a given number of gene expression changes. Richness is the probability that the number of observed statistically significant gene expression changes connected to a given hypothesis could have occurred by chance alone. Concordance is the probability that the number of observed statistically significant gene expression changes that match the directionality of the hypothesis (eg, increased or decreased kinase activity for a kinase, increased or decreased transcriptional

Downloaded from http://toxsci.oxfordjournals.org/ at University of Cambridge on November 17, 2014

swelling, loss of cristae, and reduced matrix density have been observed microscopically in hepatocytes of patients treated with the IDILI compound, tolcapone (Spahr et  al., 2000). Mitochondrial impairment has also been demonstrated for other drugs reported to produce IDILI, such as troglitazone, nefazodone, and nimesulide (Dykens et al., 2008; Haasio et al., 2002; Mingatto et  al., 2000). Interestingly, heterozygous SOD2+/− (superoxide dismutase) mice that are at reduced capacity to scavange oxidants in the mitochondria are more susceptible to DILI and IDILI (Kashimshetty et al., 2009; Lee et al., 2008). Reverse Causal Reasoning (RCR) utilizes large-scale molecular data together with the Selventa Knowledgebase to identify networks and pathways that underlie pathogenic processes of disease or drug effects (Deehan et al., 2012; Laifenfeld et al., 2010). In this study, RCR methodology was utilized to identify plausible molecular pathways underlying IDILI based on hepatic gene expression profiling from rats treated with IDILI compounds. A  panel of IDILI drugs from multiple chemical and pharmacologic classes, along with nonhepatotoxic (NONDILI) comparator compounds of the same chemical and/or pharmacologic classes, was selected for this research. A single dose of IDILI or NON-DILI drugs was administered to rats at the maximum tolerated dose (MTD) for 24 h, at which point the liver was harvested for gene expression analysis and subsequent RCR analysis. In the absence of liver injury, RCR identified key molecular pathways involved in mitochondrial injury, inflammation, and ER stress in response to IDILI drugs relative to NON-DILI drugs. In vitro studies assessing mitochondrial or ER activity suggest that these systems were also perturbed at the functional level and validate the in vivo findings.

235

Levofloxacin Trovafloxacin Methotrexate Leflunomide Finasteride Flutamide Rosiglitazone PF194223 Metformin Valdecoxib Glafenine Nimesulide Sulindac Entacapone Tolcapone Buspirone Nefazodone

NON-DILI IDILI NON-DILI IDILI NON-DILI IDILI NON-DILI IDILI NON-DILI NON-DILI IDILI IDILI IDILI NON-DILI IDILI NON-DILI IDILI

Human DILI Classification 100 100 200 100 300 500 250 500 2500 500 1000 100 50 ? 600 100 100

In Vivo Dose to Rat, mg/kg 15.773 5.024 0.7731 24.815 0.099 0.362 1.0436 6.5 12.388 0.512 1.8813 21.083 31.986 3.9344 20.879 0.005 0.9253

Cmax, µM No No No Yes Yes Yes No Yes No Yes Yes Yes No No No No Yes

Mitochondrial Injury Demonstrated by RCR NE NE NE 74.9 ± 8.4 NE 99.1 ± 19.2 238.2 ± 9.8 87 ± 16 NE NE 262 ± 11.3 5.32 ± 1.7 NE 152.6 ± 20.3 < 3.12 NE NE

RST UC50, nmol/mg Protein NE NE NE NE NE 69.7 ± 25.8 226 ± 14.3 38 ± 6.6 NE 102 ± 7.37 143 ± 11.6 NE NE 426.3 ± 19.4 167.4 ± 16.7 NE 32.2 ± 7.2

RST IC50, nmol/mg Protein

3-Fold CHOP mRNA Increase < 100 × Cmax NE 0.3 ± 0.01 @ 62.5μM NE 4.1 ± 1.3 @ 500μM NE NE 3.0 ± 1.4 @ 125μM NE 7.8 ± 2.1 @ 250μM NE NE 5.0 ± 2.2 @ 125μM 8.7 ± 2.1 @ 300μM 3.0 ± 0.4 @ 3000μM NE NE NE 4.4 ± 0.3 @ 62.5μM

ER Stress Demonstrated by RCR Yes Yes No Yes No Yes No Yes No No Yes Yes Yes No No Yes Yes

Downloaded from http://toxsci.oxfordjournals.org/ at University of Cambridge on November 17, 2014

No No No Yes No Yes No Yes No Yes Yes Yes Yes No No No Yes

Inflammation Demonstrated by RCR

Notes. The in vivo administered dose to the rats was literature based. IC50 (inhibition) and UC50 (uncoupling) values as measured in the mitochondrial oxygen consumption assay (Respiratory Screening Technology, RST) and RCR_prediction of such event. NE, no effect. Data mean ± SD (n = 3). CHOP mRNA expression in HepG2 cells as a measurement of ER stress and Causal Reasoning prediction of such event. Data are mean ± SE (n = 3). NE means < 3-fold increase in CHOP mRNA expression at less than 100 × Cmax.

Antianxiolytic

COMT

COX

PPAR/Diabetes

Androgen pathway

DMARDS

Fluoroquinolone

Pharmacologic Class

Compound Name

Table 1 Drugs Utilized in Our In Vivo and In Vitro Studies, Their Pharmacological Class, Their DILI Classification, the In Vivo Administered Dose for the Rat Studies, and Their Human Maximum Plasma Concentration of the Drug (Cmax)

236 LAIFENFELD ET AL.



237

MOLECULAR PATHWAYS IMPLICATED IN IDILI

activity for a transcription factor, etc.) could have occurred by chance alone. A hypothesis is considered to be statistically (although not necessarily biologically) significant if it met the richness and concordance cutoffs of 0.1. RCR on this data set comparisons yielded many statistically significant hypotheses, which were further investigated and prioritized by evaluation of their biological relevance to the experimental context, whether they are causally linked to phenotypes and processes relevant to liver injury in the literature and if they are causally downstream of the compound’s molecular target. Using both the information stored in the rat knowledge assembly model and from the literature, causal links between selected hypotheses were identified to produce an overall biological network capable of explaining the effects of these compounds. Criteria for Positive/Negative Compound Determination Based on RCRIdentified IDILI Processes

In Vitro Assays Oxygen consumption measurement in isolated rat liver mitochondria.  Mitochondria were prepared from the liver of naïve male Sprague Dawley rats (150–180 g). Rats were subject to CO2 euthanasia, and the liver was rapidly excised and placed in ice-cold Buffer I  (210mM mannitol, 70mM sucrose, 5mM N-2-Hydroxyethylpiperazine-N’-2-Ethanesulfonic Acid [HEPES], 1mM EGTA, 0.5% bovine serum albumin, pH 7). Liver tissue (6 g) was finely minced and washed repeatedly in Buffer I to get rid of blood, and then homogenized using 6–8 passes of a smooth glass grinder with Teflon pestle driven by a power drill on low speed. The homogenate was then adjusted to 8 volumes with Buffer I and centrifuged at 700 × g at 4°C for 10 min, then filtered through 2 layers of cheesecloth, and recentrifuged for 10 min at 14 000 × g to precipitate the mitochondrial fraction. The mitochondrial pellet was resuspended in 20 ml of isolation Buffer I and recentrifuged at 10 000 × g for 10 min at 4°C. This wash step was repeated using buffer II (210mM mannitol, 70mM sucrose, 10mM MgCl2, 5mM K2HPO4, 10mM MOPS, 1mM EGTA, pH 7.4), and the mitochondrial pellet was resuspended in 0.7 ml of Buffer II and stored on ice until required. Oxygen consumption was monitored in 96-well plate format using a phosphorescent oxygen-sensitive probe as previously described (Hynes et  al., 2006): Briefly, MitoXpress oxygen probe (Luxcel, Cork, Ireland) was reconstituted in 10.5 ml of buffer II to a concentration of approximately 100nM. Hundred microliters of this solution were pipetted into each well of a 96-well plate yielding 10 pmol/well. For drug treatments, compound stock solutions were prepared in dimethyl sulfoxide (DMSO) and added to the wells to give the indicated final concentrations and maintaining final DMSO concentration below 0.5% (vol/vol). All drug concentrations are presented as nanomoles per milligram of mitochondrial protein. After drug or vehicle addition, 50  μl of substrate (12.5/12.5mM glutamate/malate final concentration) with or without adenosine diphosphate (1.65mM final concentration) in respiration buffer (250mM sucrose, 15mM KCl, 1mM EGTA, 5mM MgCl2, 30mM K2HPO4, pH 7.4) and 50  μl of mitochondria stock solution were added to each well giving the desired final concentration of mitochondria. Finally, 100  μl of mineral oil was added to each well to seal the samples from ambient oxygen, and the plate was placed in a fluorescence plate reader (Safire2 Tecan, Innsbruck,

Measurement of CHOP induction in HepG2 cells.  Human hepatocellular carcinoma (HepG2) cells (ATCC, Manassas, Virginia; no. HB8065) were cultured in Dulbecco’s Modified Eagle Medium (Gibco, Grand Island, New York; no. 11885) supplemented with 10% heat-inactivated fetal bovine serum (Sigma; F-4135), 1% penicillin-streptomycin (Gibco 15140), 2mM glutamine (Gibco; no. 25030), and 5mM HEPES (Gibco; no. 15630). Cells were maintained in a standard cell culture incubator at 37°C and 5% CO2 with a water reservoir for humidity control. For the quantitative RT-PCR CHOP (C/EBP homologous protein) analysis, HepG2 cells were seeded on collagen type I-coated 6-well plates (BD Bioscences, Woburn, Massachusetts; no. 356400) at 1 × 105 cells/cm2. HepG2 cells were treated with 5 concentrations of each compound. The concentrations of the compounds tested were in the range of 100 times therapeutically appropriate drug exposure levels defined by average plasma maximum concentration (Cmax) values observed in humans (see Table  1; Xu et  al., 2008). Total RNA was extracted from HepG2 cells treated with compounds at 24 h using the Qiagen RNeasy plus mini kit (Invitrogen, Carlsbad, California; no. 74134). One micrograms of total RNA were reverse transcribed to cDNA using the First-strand cDNA Synthesis Kit (Qiagen, Valencia, California; no.  11752-050). Real-time qPCR was performed in a 384-well plate using the TaqMan Gene Expression Master Mix (Applied Biosystems, Foster City, California, no.  4370074)  and a CHOP primer pair (Applied Biosystems; no. 4331182) in an ABI PRISM 7900HT Sequence Detection System (Applied Biosystems). qPCR was performed with an initial step at 95°C for 10 min, followed by 40 cycles at 95°C for 3 s and 60°C for 30 s. The CHOP expression levels were calculated after conversion to numerical values by the ABI PRISM 7700 SDS software and are shown as ratios relative to the expression level of RPL19.

Results

Identification of Molecular Pathways via RCR on Gene Expression Responses in Rats A single maximal tolerated dose based on a 24-h exposure period was determined to provide the greatest opportunity to identify molecular pathways in the liver that may be indicative of a hepatotoxic response for IDILI drugs. As expected, none of the IDILI or NON-DILI compounds produced any evidence of liver injury in rats based on microscopic evaluation of the liver or assessment of liver function tests (ALT, AST, Alb, total protein, and bilirubin; data not shown). RCR of hepatic gene expression data identified 3 main processes and their associated molecular pathways as underlying IDILI: mitochondrial injury, inflammation, and/or ER stress. The majority (67%) of IDILI compounds demonstrated activation of all 3 of these mechanisms, whereas not a single NON-DILI drug showed support for all 3 (Tables 2A, 2B, and C).

Downloaded from http://toxsci.oxfordjournals.org/ at University of Cambridge on November 17, 2014

Statistically significant RCR-generated hypotheses were identified for all compounds and assembled into a mechanistic network, resulting in identification of 3 processes in IDILI—mitochondrial injury, inflammation, and ER stress. Although each process comprised multiple underlying molecular mechanisms, a compound was designated with positive or negative responses based on the following criteria: for inflammation and ER stress processes, a compound was considered positive if at least one of the molecular mechanisms was a statistically significant hypothesis in the expected direction for injury, whereas at least 2 of the molecular mechanisms had to be a statistically significant hypothesis for mitochondrial injury due to the large number of mechanisms involved in this process; in some cases, there was a mixed response, at which point a compound was only considered positive if multiple mechanisms supported positive determination of injury with only a single mechanism in an opposing direction.

Austria) equilibrated at 30°C and monitored over a period of 20 min measuring probe fluorescence signal in each well every 1.5 min in kinetic mode using 380/650 nm excitation/emission, a delay time of 30 μs and a measurement window of 100 μs. To ensure gas and temperature equilibration, all the dispensing steps were carried out at 30°C using prewarmed solutions and a Multi-Blok plate heater (Barnstead/LabLine, Melrose Park, Illinois). After completion of fluorescence measurements, time profiles of fluorescence intensity in each well were analyzed using Magellan (Tecan) and MS Excel (Microsoft, Redmond, Washington) to determine the rates of oxygen consumption based on the known relationship between probe fluorescence and oxygen concentration. Rates of change of dissolved oxygen were subsequently determined from the slopes of these concentration profiles over the initial 8 min.

▲ [17]

▲ [11] ▲ [7]

▲ [10]

▲ [34]

▲ [10]

▲ [13]

▲ [17]

▲ [42] ▲ [185]

▼ [30]

▼ [27]

▼ [7]

▼ [35]

▲ [8] ▲ [57]

▲ [22]

▲ [100]

▼ [15]

▲ [5]

▲ [86] ▲ [20] ▲ [20] ▲ [25]

Leflunomide

▲ [18] ▲ [8] ▲ [8]

▼ [24] ▲ [70]

▲ [17] ▲ [24]

▲ [83]

Finasteride

NON-DILI

IDILI

▲ [41] ▲ [41]

▲ [195]

▼ [27] ▲ [185] ▲ [79] ▼ [43] ▲ [46] ▲ [9] ▲ [7]

▲ [201] ▲ [21] ▲ [44] ▲ [74] ▲ [22] ▲ [7]

Flutamide

Cancer

Androgen Pathway/

Rosiglitazone

NON-DILI

▼ [57] ▼ [31]

▲ [22] ▼ [42] ▼ [55]

▼ [180]

▲ [8] ▲ [7]

▼ [23] ▲ [97]

▲ [8] ▲ [4]

▲ [94]

PF194223

IDILI

PPAR/Diabetes

Metformin

NON-DILI

▲ [23] ▲ [31]

▲ [44] ▲ [117]

▲ [92]

▼ [26] ▲ [55]

▲ [94] ▲ [16] ▲ [20] ▲ [39]

Valdecoxib

NON-DILI

▲ [54] ▲ [66] ▼ [13]

▲ [143] ▲ [52] ▲ [58] ▲ [203] ▲ [19] ▲ [33]

▲ [59] ▲ [43] ▼ [27]

▲ [37]

▲ [67]

Glafenine

IDILI

▲ [39]

▼ [45] ▼ [50]

▲ [43]

▲ [30] ▲ [8] ▲ [7]

▼ [21] ▲ [87] ▲ [49]

▲ [109] ▲ [23] ▲ [31] ▲ [42] ▲ [9]

Nimesulide

IDILI

COMT Inhibitor/

Entacapone

Sulindac

▲ [21] ▲ [31]

▲ [14]

▲ [63]

▲ [37] ▲ [18]

▲ [16]

NON-DILI

▼ [21]

Tolcapone

IDILI

Parkinson’s Disease

IDILI

COX-2 Inhibitor/NSAID/Inflammation

Downloaded from http://toxsci.oxfordjournals.org/ at University of Cambridge on November 17, 2014

▼ [9]

▼ [35]

▼ [41]

▼ [82]

▼ [53]

▲ [42] ▲ [23]

▼ [60]

▲ [131]

▲ [31] ▲ [8] ▲ [7]

▼ [27] ▲ [132] ▲ [64]

▲ [161] ▲ [23] ▲ [35] ▲ [51] ▲ [15] ▲ [7]

▼ [60] ▼ [18] ▼ [27]

Nefazodone

IDILI

Buspirone

NON-DILI

SSRI/Antianxiolytic

Notes. Arrows indicate the predicted directionality of each molecular mechanism, and the number in brackets indicates the number of supporting genes for the mechanism. (A) The mitochondrial injury network is activated in IDILI compounds. The mitochondrial injury network identified by RCR includes 2 main subnetworks: ROS-associated mechanisms and an increased antioxidant response. taof(Nfe2l2), transcriptional activity of Nfe2l2l; catof(SOD1), catalytic activity of SOD1; gtpof(Hras), gtp-bound activity of Hras. (B) An increased inflammatory response is observed in IDILI compounds. An inflammatory imbalance that can contribute to injury includes an increase in the hepatotoxic proinflammatory network and a suppression of the hepatoprotective network. An increased hepatotoxic response is supported by elements of a proinflammatory network including increased Tnf, IL1b, Il6, NFKB activity, and LPS-like and lipoteichoic-like signaling. A decreased hepatoprotective effect is supported by a decrease in elements of Ifng/Stat3 pathway. taof(stat3), transcriptional activity of stat3; taof(NFkB), transcriptional activity of NFkB. (C) An imbalance in the ER stress response contributes to IDILI. ER stress imbalance was demonstrated via an increased ER stress response, accompanied in some compounds by an activation of ER stress proteins (eg, Xbp1) or proapoptotic proteins.

(A)   taof (Nfe2l2)  TFEC  ROS   Oxidative stress   taof (Nrf1)   catof (Nos1)   catof (Nos3)   catof (Proteasome)  Insulin   gtpof (Hras)   catof (Sod1)   Response to heat  Hsf1  Hsf2 (B)  Tnf  Il1b  taof(NFKB)  LPS   Lipoteichoic acid  Il6  Ifng  taof(Stat3) (C)   ER stress  taof(Xbp1)  taof(Foxo3a)  taof(Tp53)  taof(ATF6) in vitro

Trovafloxacin

Levofloxacin

Methotrexate

NON-DILI

IDILI

NON-DILI

IDILI

DMARDs/Antirheumatic

Fluoroquinolone/Antibiotics

Table 2 Processes Supporting Elements of Three Injury Networks in Response to Idili Compounds

238 LAIFENFELD ET AL.



MOLECULAR PATHWAYS IMPLICATED IN IDILI

(nuclear factor (erythroid-derived-2)-like), Nrf1 (nuclear respiratory factor), Nos (nitric oxide synthase), and decreased catalytic activity of SOD1 (superoxide dismutase 1), all of which can be induced by ROS production. Collectively, these responses support a mitochondrial injury response (Fig. 1) at varying degrees for 6 of the 9 IDILI compounds (leflunomide, flutamide, PF194223, glafenine, nimesulide, and nefazadone). In contrast, only 2 of the 8 NON-DILI compounds (finasteride and valdecoxib) produced a mitochondrial injury type response (Table 2A). Inflammatory response.  The inflammatory response network consisted of increased tumor necrosis factor (TNF), interleukin-1 beta (IL-1β), IL-6, NFkB activity, and lipopolysaccharide (LPS) like and lipoteichoic like (Fig. 2; Table 2B). This was present in 78% (7/9) of IDILI compounds (all but tolcapone). Also suggestive of an inflammatory response

Fig. 1.  The mitochondrial injury network is activated in IDILI compounds. The mitochondrial injury network identified by RCR includes 2 main subnetworks: ROS-associated mechanisms and an increased antioxidant response. Arrows indicate direction of mechanism when the network is activated. Abbreviations: catof(SOD1), catalytic activity of SOD1; gtpof (Hras), gtp bound activity of Hras; IDILI, idiosyncratic drug-induced liver injury; RCR, Reverse Causal Reasoning; ROS, reactive oxygen species; Taof(Nfe212), transcriptional activity of Nfe212.

Fig. 2.  An increased inflammatory response is observed in IDILI compounds. An inflammatory imbalance that can contribute to injury includes an increase in the hepatotoxic proinflammatory network and a suppression of the hepatoprotective network. An increased hepatotoxic response is supported by elements of a proinflammatory network including increased Tnf, Il1b, NFKB activity, and LPS- and lipoteichoic-like signaling. A decreased hepatoprotective effect is supported by a decrease in elements of the Il6/Ifng/Stat3 pathway. Arrows indicate direction of mechanism when the network is activated. Abbreviations: taof(stats3), transcriptional activity of stats3; taof (NFkB), transcriptional activity of NFkB.

Downloaded from http://toxsci.oxfordjournals.org/ at University of Cambridge on November 17, 2014

Mitochondrial injury.  The mitochondrial injury network was evident for 66% (6/9) of IDILI compounds by a subnetwork of molecular processes indicative of reactive oxygen species (ROS) production and an antioxidant response (Fig. 1, Table  2A). In addition to a general signature consistent with ROS production, molecular processes of oxidative stress, Hras (GTPase HRas) activation, proteasome inhibition, and increased insulin signaling were present in IDILI drugs. Hras is a redoxsensitive protein that can be activated in response to changes in redox balance, such as those induced by ROS production, as well as in response to insulin signaling. Insulin signaling is prolonged in the face of proteasome inhibition, and proteasome inhibition stabilizes Hras. Finally, there is a reciprocal relationship between ROS and the proteasome in that ROS interferes with proteasome activity, and reduced activity of the proteasome can lead to ROS production. Components of the antioxidant response include increased transcription activity of Nfe2l2

239

240

LAIFENFELD ET AL.

was a decreased hepatoprotective effect, which consisted of a decrease in elements of the Interferon-gamma (Ifng)/signal transducer and activator of transcription 3 (Stat3) pathway (Table 2B). This reduced hepatoprotective network was present in 3 out of 9 IDILI compounds (PF194223, nimesulide, and nefazodone), all of which also showed an element of a proinflammatory response, with trovafloxacin and PF194223 also showing some anti-inflammatory activity. In contrast, of the 8 NON-DILI compounds, only valdecoxib demonstrated a proinflammatory response (Table 2B).

Fig.  3.  An imbalance in the ER stress-response contributes to IDILI. ER stress imbalance was demonstrated via an increased ER stress response, accompanied in some compounds by an activation of ER stress proteins (eg, Xbp1) or proapoptotic proteins. Arrows indicate direction of mechanism when the network is activated. Abbreviations: taof (Tp53), transcriptional activity of Tp53; taof(foxo3a), transcriptional activity of foxo3a; taof (xbp1), transcription activity of xbp1.

In Vitro Assessment of RCR Mechanisms The IDILI and NON-DILI compounds were tested in in vitro assays of mitochondrial injury and ER stress to determine whether the molecular patterns identified by RCR had a functional consequence. Specifically, the same set of compounds tested in rats was also tested in isolated rat liver mitochondria to detect possible mitochondrial impairment as marked by oxygen consumption (Will et al., 2006) and CHOP assessment in HepG2 cells as a marker of ER stress (Given a lack of a compelling inflammatory in vitro model, this process is not represented in the in vitro assessment.). Isolated rat liver mitochondrial assay.  The mitochondrial assay employed measures both uncoupling (UC) and inhibitory (IC) mechanisms of mitochondrial impairment. The in vitro mitochondrial respiration assay identified 89% (7/9) of the IDILI compounds as affecting mitochondrial function (Table 1), validating the RCR findings. Overall, mitochondrial impairment was found in 78% (7/9) IDILI compounds and in 38% (3/8) NON-DILI compounds. Nimesulide (Fig. 4G) and tolcapone (Fig.  4I) uncoupled mitochondrial respiration with UC50 values < 10 nmol/mg protein and leflunomide (Fig. 4A) and PF194223 (Fig. 4D) uncoupled mitochondrial respiration with UC50 values < 100 nmol/mg protein. Nefazodone (Fig. 4J) showed an inhibitory effect on mitochondrial respiration, with an IC50 value < 50 nmol/mg mitochondrial protein. Flutamide (Fig.  4B) caused both an uncoupling and an inhibitory effect (UC50 > 99.1; IC50 = 69.7) and so did glafenine (Fig. 4F), but much less potent (UC50 > 262; IC50 = 143). Three NON-DILI compounds—rosiglitazone (Fig.4C), valdecoxib (Fig.4E), and entacapone (Fig. 4H)—also showed a response on mitochondrial oxygen consumption. Rosiglitazone (Fig. 4C) had an UC50 value of 238 nmol/mg protein and an IC50 value of 226 nmol/mg protein. Valdecoxib (Fig.  4E) did not exhibit uncoupling and displayed an IC50 value of 102  nmol/mg protein. Entacapone (Fig. 4H) showed an IC50 value of 426 nmol/mg protein and a UC50 values of 152 nmol/mg protein. ER stress hypothesis.  To examine ER stress in vitro, CHOP mRNA expression was measured in HepG2 cells treated with the IDILI and NON-DILI compounds for 24 h. CHOP is a protein involved in making the cell death decision associated with ER stress. All compounds were tested below or at approximately 100 times the maximum human plasma drug concentrations (Cmax) (Table  1). Six out of the 9 (66%) IDILI compounds (PF194223 [Fig. 5A], nimesulide

Downloaded from http://toxsci.oxfordjournals.org/ at University of Cambridge on November 17, 2014

ER stress.  An ER stress response (Fig. 3), including activation of ER stress proteins such as Xbp1 (X-box binding protein) and ATF6 (activating transcription factor 6) is an adaptive xenobiotic detoxification response (Yamamoto et  al., 2004). RCR identified an ER stress response in 88% (8/9) of IDILI compounds (all but tolcapone; Table  2C). In contrast, an ER stress response was only supported in 2 NON-DILI compounds, levofloxacin and buspirone. In the majority of the IDILI compounds, a general signature for ER stress was accompanied by specific support for the activation of the ER stress protein Xbp1, which, depending on the extent of the stress and degree of protein activation, can be adaptive or injurious (Table 2C). In response to leflunomide, there is an increase in the cytotoxic

ER stress pathway, involving activation of TP53 (protein 53 gene) and/or FOXO3A (Forkhead box A3) (Table 2C). Finally, in response to PF194223, there is a decrease in the cytoprotective ER stress response as demonstrated by a decrease in the general signature for ER stress, as well as a decrease in the activity of Xbp1, which can result in decreased xenobiotic detoxification (Table 2C).



MOLECULAR PATHWAYS IMPLICATED IN IDILI

241

Downloaded from http://toxsci.oxfordjournals.org/ at University of Cambridge on November 17, 2014 Fig. 4.  Oxygen consumption measurements in rat liver mitochondria. The assay was run under basal (left panels) and ADP-stimulated (right panels) conditions. Data are mean ± SD (n = 3). Testing concentrations were as follows: 300 nmol/mg protein (closed circles), 150 nmols/mg protein (closed squares), 75 nmol/ mg protein (closed triangle), 37.5 nmol/mg protein (closed inverted triangles), 18.75 nmol/mg protein (open diamonds), 9.375 nmol/mg protein (open circles), and dimethyl sulfoxide control (open squares). (A) leflunomide, (B) flutamide, (C) rosiglitazone, (D) PF194223, (E) valdecoxib, (F) glafenine, (G) nimesulide, (H) entacapone, (I) tolcapone, and (J) nefazodone.

LAIFENFELD ET AL.

242

Downloaded from http://toxsci.oxfordjournals.org/ at University of Cambridge on November 17, 2014

Fig. 4.  Continued

MOLECULAR PATHWAYS IMPLICATED IN IDILI



243

[Fig. 5B], nefazodone [Fig. 5C], leflunomide [Fig. 5D], sulindac [Fig.  5E], and glafenine [Fig.  5F]) produced at least a 3-fold increase in CHOP mRNA expression at a concentration less than their 100 × Cmax compared with vehicle-treated cells (Table 1). One IDILI drug, trovafloxacin (Fig. 5G), produced a decrease in CHOP mRNA expression by 3-fold. None of the NON-DILI compounds induced CHOP expression by > 3-fold at a concentration less than their 100 × Cmax compared with the vehicle-treated cells.

Discussion

IDILI is a rare but serious side effect of many drug treatments and represents one of the largest undetected liabilities for new drugs in development. General safety assessments in laboratory animals do not show toxicity for compounds found to cause IDILI in humans, and, therefore, hold no predictive value for this type of toxicity based on evaluation of standard parameters. Understanding the underlying injury mechanisms responsible for IDILI is the first step toward identifying risk in preclinical models. To this end, this research utilized RCR on gene expression data from livers of rats treated with multiple pharmacologically distinct IDILI compounds to identify common pathways that these compounds may be modulating. This analysis led to the identification of 3 key injury pathways in response to IDILI compounds—mitochondrial injury,

inflammation, and ER stress. Mitochondrial injury and ER stress were subsequently assessed and validated in relevant in vitro assays. The mitochondrial injury network, which consists of a subnetwork of molecular processes leading to ROS production and an antioxidant response (Fig. 1), was identified in the majority (6 of 9) of IDILI compounds. This ROS production subnetwork comprised increased ROS and oxidative stress, as well as Hras activation, proteasome inhibition, and increased insulin signaling. Hras is a redox-sensitive protein that can be activated in response to changes in redox balance (Abe et al., 2000), such as those induced by ROS production, as well as in response to insulin signaling (Sasaoka et  al., 1996). Proteasome inhibition prolongs insulin signaling (Haruta et  al., 2000) while stabilizing Hras (Svegliati et  al., 2005). Finally, ROS interferes with proteasome activity (Torres and Perez, 2008), and reduced activity of the proteasome can lead to ROS production (Wu et al., 2002). Molecular changes that cause mitochondrial injury can also drive antioxidant processes as an adaptive response. Components of the antioxidant response include activation of Nfe2l2, Nrf1, and Nos, all of which can be induced by ROS production (Piantadosi and Suliman, 2006; Venugopal and Jaiswal, 1998; Zhen et al., 2008). Consistent with RCR results, mitochondrial effects have been previously associated with several of the IDILI compounds (Berson et al., 2001; Dykens et al., 2008; Haasio et al., 2002; Hail et al., 2010; Kashimshetty et al., 2009; Ong et al., 2007).

Downloaded from http://toxsci.oxfordjournals.org/ at University of Cambridge on November 17, 2014

Fig. 4.  Continued

244

LAIFENFELD ET AL.

Downloaded from http://toxsci.oxfordjournals.org/ at University of Cambridge on November 17, 2014

Fig. 5.  Real-time PCR analysis of HepG2 cells after induction of ER stress. HepG2 cells were treated with concentrations of compounds, and changes in CHOP mRNA expression were detected 24 h after treatment. Data were normalized to DMSO-treated control samples and results are expressed as a percentage of DMSO vehicle controls. Error bars indicate mean ± SD (n = 3). (A) PF194223, (B) nimesulide, (C) nefazodone, (D) leflunomide, (E) sulindac, (F) glafenine, and (G) trovafloxacin. Abbreviation: DMSO, dimethyl sulfoxide.



MOLECULAR PATHWAYS IMPLICATED IN IDILI

Mild inflammation due to infection or other environmental stressors has been implicated in IDILI. For example, endotoxin cotreatment with IDILI compounds has been shown to produce liver injury in rats (Roth et al., 2003), consistent with the notion that an incidental inflammatory response in conjunction with a low-grade inflammatory response caused by IDILI compounds may culminate in an inflammatory effect that results in injury. Indeed, our findings are consistent with the idea that mild inflammation can be elicited by IDILI compounds. A number of compounds tested in our study, such as trovafloxacin, levofloxacin, and sulindac, have been tested in rats that were exposed to a nontoxic dose of LPS(Shaw et al., 2009). In this model, liver injury in rats was caused by a mild inflammation from exposure to LPS. Specifically, idiosyncratic hepatotoxicants such as trovafloxacin and sulindac induced significant liver injury in rats cotreated with LPS, whereas their considered nontoxic counterparts such as levofloxacin failed to induce liver injury in LPS-treated rats (Shaw et al., 2009). The inflammation mechanism of trovafloxacin-induced toxicity was previously studied by Liguori et al. (2005), who revealed that trovafloxacin uniquely regulates a larger number of genes that are important in major biological processes, such as the inflammatory response. In addition, liver gene-expression profiling from mice identified distinct gene-expression patterns induced by trovafloxacin/LPS coexposure, including changes in expression of genes involved in interferon signaling, which led to the mechanistic finding that IFN-γ and IL-8 play critical roles in trovafloxacin-/LPS-induced liver injury (Shaw et al., 2009). It may be that trovafloxacin is producing a low-grade inflammatory response that may prime the liver for an additional inflammatory insult. The ER stress response involves a complex interplay among proteins, and the ability of the cell to appropriately respond to ER stress determines the injurious versus adaptive nature of the response. RCR identified ER stress effects in 8 of the 9 IDILI compounds. Given the importance of drug-metabolizing cytochrome P450 enzymes in the ER and their role in xenobiotic bioactivation, it is not surprising that ER stress may play a role in IDILI. RCR demonstrated an increased signature for a general ER stress response in 6 of the IDILI compounds, of which 4 showed the expected accompanying increase in the activity of the ER stress protein Xbp1. In those compounds that did not show activation of ER stress proteins despite induction of ER stress, the insult to the ER may have been too great and activation of ER-associated cell death pathways may have been triggered, contributing to injury. Indeed, in 2 of these compounds, leflunomide and trovafloxacin, there is activation of ER stress-responsive proapoptotic proteins, Foxo3a and Tp53. None of the NON-DILI compounds showed induction of ER stress, with the exception of levofloxacin. We suggest, therefore, that in compounds associated with idiosyncratic DILI, one of the mechanisms contributing to injury could be an overwhelming degree of ER insult and stress.

Downloaded from http://toxsci.oxfordjournals.org/ at University of Cambridge on November 17, 2014

Of particular interest were the findings with glafenine, an analgesic that has also been associated with hepatotoxicity (Stricker et  al., 1986). To the best of our knowledge, this is the first report of glafenine showing an effect on respiration of isolated liver mitochondria. Although glafenine has not been directly implicated in mitochondrial injury, it has been suggested that the bioactivation of glafenine and subsequent binding of reactive metabolite(s) to critical cellular proteins play a causative role in its toxicity (Wen and Moore, 2011). In addition, the presence of fever and eosinophilia suggested immunologic idiosyncrasy as a possible mechanism for glafenine’s hepatotoxicity (Zimmerman, 1999). The specific molecular mechanisms involved in glafenine toxicity as demonstrated by RCR are consistent with this and represent novel insight for understanding the molecular mechanisms underlying toxicity of glafenine. One of the molecular mechanisms associated with ROS production and supported by causal analysis in response to IDILI compounds is proteasome inhibition. Proteasome inhibition is one of 2 molecular processes most prevalent with IDILI compounds (the other being Nfe2l2 activation). Proteasome inhibition has been implicated in response to oxidative stress (Torres and Perez, 2008), mitochondrial injury, and in alcoholic liver injury (Osna and Donohue, 2007). Therefore, it is not surprising to identify proteasome inhibition as a plausible contributor in IDILI. Because proteasome function gradually deteriorates with age (Deocaris et al., 2008), it is possible that further inhibition of the proteasome with drugs may render the elderly population susceptible to IDILI. We observed elements of an increased inflammatory network in response to all but one of the IDILI drugs tested in vivo. The inflammatory response was absent in all but one of the NONDILI drugs. Our data suggest that Tnf activation through NFKB may lead to an exaggerated inflammatory response that leads to overt injury. In drugs that lead to liver injury, susceptibility to injury correlates with the activity of Tnf (Bourdi et al., 2002). Of the 7 compounds whose gene expression signatures support increased inflammation, 4 had increased Tnf. Tnf alone, however, may be insufficient to cause overt hepatocellular damage (Deng et  al., 2008). In all but one IDILI compound in which Tnf was induced, there is a lack of downstream Tnf signaling based on the absence of support for NFKB activation or the induction of additional proinflammatory cytokines. The increase in an inflammatory response constitutes one part of an immune imbalance theory for IDILI. A complementary component is the decrease in molecular elements implicated in hepatoprotection, resulting in increased susceptibility to inflammation. Indeed, causal reasoning revealed decreased activity of the Il6/Stat3 pathway, which is considered hepatoprotective (Taub, 2003), resulting in decreased expression of the anti-inflammatory acute phase reactants in response to multiple idiosyncratic compounds. Thus, in addition to increased inflammation, decreased hepatoprotection may contribute to increased susceptibility to inflammation with exposure to idiosyncratic compounds.

245

246

LAIFENFELD ET AL.

the quantitative criteria for use in screening for compound safety. As noted earlier, measurement of routine parameters (clinical chemistry, histopathology) from standard animal models cannot predict the potential for some compounds to cause DILI in humans. Some groups have had success in using alternative animal models to detect such compounds. Examples already cited include LPS-treated rodents (Roth et  al., 2003) and SOD+/− mice (Ong et al., 2007) but also a genetically diverse panel of inbred mice (Harrill et  al., 2009). These models all try to mimic some aspect of diversity in the human population whether it is environmental, disease, or genetic. In the approach described herein, a transcriptional response in the standard rat model produced known mechanisms of hepatic injury in the absence of toxicological response (ie, increased serum ALT or liver necrosis). It is important to note that it is the signaling of the molecules that is associated with IDILI, rather than the abundance level. The confirmation for aberrant signaling was accomplished through the functional assays of these mechanisms. It is possible that although there are some subtle cellular or biochemical changes associated with the molecular response, the liver as a whole is able to respond in a way that protects it from cellular injury. Interestingly, compounds that do cause DILI in animals, such as acetaminophen and carbon tetrachloride, activate the same molecular pathways but with a much stronger response (data not shown), suggesting that although the molecular response to IDILI compounds in rats is detectable, it might not be sufficiently strong to elicit cellular injury. Two important considerations in implementing predictive in vitro assays are the physiological relevance of the assays and the parameters by which testing concentrations are established. For example, how does one differentiate true positive from false-positive compounds in the absence of an accurate Cmax projection to understand the safety margin? It is also clear that assays have to be developed that adequately recapitulate the in vivo response. This is challenging giving the limitations of most cell-based models. Whether coculture and 3D models will improve the predictivity remain to be seen. Khetani et al. (2013) did demonstrate that long-term dosing in a hepatocytes coculture model increases predictivity toward DILI without increasing the false-positive rate for NON-DILI compounds. Recently, Thompson et al. (2012) proposed an in vitro risk matrix, where the sum of several assay results was correlated with in vivo hepatotoxicity outcomes. Thompson’s assay suite contained the mitochondrial assay but could benefit from including an ER stress assessment and an improved inflammation assay. The importance of Cmax when extrapolating in vitro assay data has been discussed by Xu et  al. (2008) and more recently by Porceddu et al. (2012). Among the 9 IDILI compounds, leflunomide, tolcalpone, nimesulide, and PF194223 were found to have UC50 or IC50 values that were lower than their 100  × Cmax concentrations. Nefazodone, flutamide, and glafenine had IC50s or UC50 values that were close to their 100 × Cmax values.

Downloaded from http://toxsci.oxfordjournals.org/ at University of Cambridge on November 17, 2014

There is evidence that the ER stress response is decreased in response to PF194223. The decrease in ER stress suggests that vehicle control treatment leads to more of an ER stress response than the drug. Given the role of the ER in detoxification together with reports of induction of the detoxifying Cyp enzymes in response to ER stress, a reduced level of ER stress induced by these compounds can contribute to susceptibility of these compounds to cellular injury. Indeed, this is consistent with the notion of a cytoprotective role for ER stress in the xenobiotic response. Together, our results support ER stress as a fine balance between protection and injury: The ER stress response is required for normal function, such that levels too high or too low can contribute to cellular injury. In total, based on gene expression data, RCR identified 3 major mechanisms predominantly found in IDILI compounds that are well supported in the literature. Of note, Zhang et al. (2012) applied statistical measures alone on transcriptomic data from rats treated with a similar set of compounds and achieved only minimal accuracy for predicting human hepatotoxicity, demonstrating the importance of a mechanistic approach to understanding underlying mechanisms in predicting toxicity. Following the results from RCR, the same compounds were assessed by selected in vitro assays to confirm the in vivo findings and to possibly establish high-throughput assays for preclinical screening for IDILI. Mitochondrial injury was confirmed in vitro for all but 2 of these compounds, trovafloxacin and sulindac. The in vitro assay likely did not identify a mitochondrial liability for sulindac because that it is a metabolite—sulindac sulfide—that is known to cause mitochondrial impairment (Leite et  al., 2006). Most of the compounds that were predicted to have ER stress in vivo were also confirmed in HepG2 cells by measuring CHOP induction. We attempted to test the inflammation hypothesis in vitro using the cytokine synergy model (Cosgrove et al., 2009). However, only two of the IDILI compounds—nefazodone and trovafloxacin—were active in our in vitro assay (data not shown). This could be due to the fact that this assay assesses a compound’s ability to synergize with inflammatory elements to cause LDH release, rather than a compound’s ability to induce inflammation as predicted by RCR. A key outcome of the current work is the insight gained into molecular pathways underlying IDILI and the difference in these pathways between IDILI and NON-DILI compounds of the same pharmacologic class. Although some individual components of the pathways were also supported in NON-DILI compounds, it appears that the combination of multiple mechanisms is what underlies the IDILI phenomena because although 78% of the IDILI compounds tested showed activation of 2 or more of the pathways, only a single NON-DILI compound (constituting 12.5%) showed activation in more than one of the pathways, and the vast majority of the non-DILI compounds that showed activation in any of the pathways demonstrated support for far fewer mechanisms within them. Follow-up work with an extended list of compounds will enable refinement of

MOLECULAR PATHWAYS IMPLICATED IN IDILI



References Abe, J., Okuda, M., Huang, Q., Yoshizumi, M., and Berk, B. C. (2000). Reactive oxygen species activate p90 ribosomal S6 kinase via Fyn and Ras. J. Biol. Chem. 275, 1739–1748. Berson, A., Descatoire, V., Sutton, A., Fau, D., Maulny, B., Vadrot, N., Feldmann, G., Berthon, B., Tordjmann, T., and Pessayre, D. (2001). Toxicity

of alpidem, a peripheral benzodiazepine receptor ligand, but not zolpidem, in rat hepatocytes: Role of mitochondrial permeability transition and metabolic activation. J. Pharmacol. Exp. Ther. 299, 793–800. Bourdi, M., Masubuchi, Y., Reilly, T. P., Amouzadeh, H. R., Martin, J. L., George, J. W., Shah, A. G., and Pohl, L. R. (2002). Protection against acetaminophen-induced liver injury and lethality by interleukin 10: Role of inducible nitric oxide synthase. Hepatology 35, 289–298. Chalasani, N., Fontana, R. J., Bonkovsky, H. L., Watkins, P. B., Davern, T., Serrano, J., Yang, H., Rochon, J.; Drug Induced Liver Injury Network (DILIN). (2008). Causes, clinical features, and outcomes from a prospective study of drug-induced liver injury in the United States. Gastroenterology 135, 1924–34, 1934.e1. Cosgrove, B. D., King, B. M., Hasan, M. A., Alexopoulos, L. G., Farazi, P. A., Hendriks, B. S., Griffith, L. G., Sorger, P. K., Tidor, B., Xu, J. J., et al. (2009). Synergistic drug-cytokine induction of hepatocellular death as an in vitro approach for the study of inflammation-associated idiosyncratic drug hepatotoxicity. Toxicol. Appl. Pharmacol. 237, 317–330. Deehan, R., Maerz-Weiss, P., Catlett, NL., Steiner, G., Wong, B., Wright, M. B., Blander, G., Elliston, K. O., Ladd, W., Bobadilla, M., Mizrahi, J., Haefliger, C., and Edgar, A. (2012). Comparative transcriptional network modeling of three PPAR-α/γ co-agonists reveals distinct metabolic gene signatures in primary human hepatocytes. PLoS One. 7, e35012. Deng, X., Lu, J., Lehman-McKeeman, L. D., Malle, E., Crandall, D. L., Ganey, P. E., and Roth, R. A. (2008). p38 Mitogen-activated protein kinase-dependent tumor necrosis factor-alpha-converting enzyme is important for liver injury in hepatotoxic interaction between lipopolysaccharide and ranitidine. J. Pharmacol. Exp. Ther. 326, 144–152. Deocaris, C. C., Takano, S., Priyandoko, D., Kaul, Z., Yaguchi, T., Kraft, D. C., Yamasaki, K., Kaul, S. C., and Wadhwa, R. (2008). Glycerol stimulates innate chaperoning, proteasomal and stress-resistance functions: Implications for geronto-manipulation. Biogerontology 9, 269–282. Dykens, J. A., Jamieson, J. D., Marroquin, L. D., Nadanaciva, S., Xu, J. J., Dunn, M. C., Smith, A. R., and Will, Y. (2008). In vitro assessment of mitochondrial dysfunction and cytotoxicity of nefazodone, trazodone, and buspirone. Toxicol. Sci. 103, 335–345. Gentleman, R. (2005). Bioinformatics and computational biology solutions using R and bioconductor. Stat. Biol. Health xix, 397–420. Haasio, K., Koponen, A., Penttilä, K. E., and Nissinen, E. (2002). Effects of entacapone and tolcapone on mitochondrial membrane potential. Eur. J. Pharmacol. 453, 21–26. Hail, N., Jr, Chen, P., and Bushman, L. R. (2010). Teriflunomide (leflunomide) promotes cytostatic, antioxidant, and apoptotic effects in transformed prostate epithelial cells: Evidence supporting a role for teriflunomide in prostate cancer chemoprevention. Neoplasia 12, 464–475. Harrill, A. H., Watkins, P. B., Su, S., Ross, P. K., Harbourt, D. E., Stylianou, I. M., Boorman, G. A., Russo, M. W., Sackler, R. S., Harris, S. C., et al. (2009). Mouse population-guided resequencing reveals that variants in CD44 contribute to acetaminophen-induced liver injury in humans. Genome Res. 19, 1507–1515. Haruta, T., Uno, T., Kawahara, J., Takano, A., Egawa, K., Sharma, P. M., Olefsky, J. M., and Kobayashi, M. (2000). A rapamycin-sensitive pathway down-regulates insulin signaling via phosphorylation and proteasomal degradation of insulin receptor substrate-1. Mol. Endocrinol. 14, 783–794. Hsiao, C. J., Younis, H., and Boelsterli, U. A. (2010). Trovafloxacin, a fluoroquinolone antibiotic with hepatotoxic potential, causes mitochondrial peroxynitrite stress in a mouse model of underlying mitochondrial dysfunction. Chem. Biol. Interact. 188, 204–213. Hussaini, S. H., and Farrington, E. A. (2007). Idiosyncratic drug-induced liver injury: An overview. Expert Opin. Drug Saf. 6, 673–684. Hynes, J., Marroquin, L. D., Ogurtsov, V. I., Christiansen, K. N., Stevens, G. J., Papkovsky, D. B., and Will, Y. (2006). Investigation of drug-induced

Downloaded from http://toxsci.oxfordjournals.org/ at University of Cambridge on November 17, 2014

In contrast, the remaining 2 IDILI compounds, sulindac and trovafloxacin, were negative. No in vitro value for trovafloxacin has been reported to date to know whether it truly inhibits mitochondrial function. A recent study by Hsiao et al. (2010) demonstrated that trovafloxacin caused mitochondrial peroxynitrite stress in a mouse model of underlying mitochondrial dysfunction. All the NON-DILI compounds had either no effect on mitochondria or the IC50s/UC50 values were higher than their 100  × Cmax values. A  similar trend was also observed for the CHOP mRNA expression data, where 7/9 IDILI compounds were found to increase CHOP mRNA expression at more than 3-fold at a testing concentration below 100 × Cmax, while none of the NON-DILI compounds did. Our data demonstrate that using 100 × Cmax value as a reference more accurately predicts what in vivo or clinical outcomes. Overall, the correlations between the in vivo RCR predictions and the in vitro assay results were high: 15/17 for mitochondrial injury and 14/17 for ER stress. The exceptions for mitochondrial toxicity were the IDILI compound tolcapone, where RCR failed to predict the mitochondrial toxicity (but was observed in the in vitro assay). Also, RCR predicted the NON-DILI compound finasteride to have mitochondrial toxicity, something that was not confirmed in vitro. For ER stress, RCR predicted the NON-DILI compounds levofloxacin and buspirone to be positive, but this was not confirmed in vitro. Similarly, the IDILI compound flutamide was positive by RCR but negative in the in vitro assay. Possible explanations for the discrepancies include the known limitations of in vitro toxicity assays and their high false-negative predictivity (Xu et al., 2008). In addition, the use of a MTD paradigm for the in vivo study could lead to false positives for the RCR predictions. To better predict the potential for a compound to cause IDILI, it is clear that a variety of assays will need to be utilized, that IDILI compounds need to be benchmarked against compounds causing other organ toxicities, and that a safety margin can only be established if a large retrospective study has been conducted such as in the case of Xu et al. (2008). Based on the results presented here, we suggest that a short-term, high-dose rodent study using gene expression profiling and causal reasoning analysis when applied to compounds that test positive in at least 2 of the in vitro assays (ER stress and mitochondrial toxicity) but are negative in exploratory rat toxicology studies could be used to increase the likelihood of identifying IDILI compounds. Follow-up studies can aid in developing a more focused gene/protein signature that captures the molecular mechanisms involved in the activation of the 3 networks identified here as implicated in IDILI.

247

248

LAIFENFELD ET AL. Roth, R. A., Luyendyk, J. P., Maddox, J. F., and Ganey, P. E. (2003). Inflammation and drug idiosyncrasy–is there a connection? J. Pharmacol. Exp. Ther. 307, 1–8. Sasaoka, T., Ishiki, M., Sawa, T., Ishihara, H., Takata, Y., Imamura, T., Usui, I., Olefsky, J. M., and Kobayashi, M. (1996). Comparison of the insulin and insulin-like growth factor 1 mitogenic intracellular signaling pathways. Endocrinology 137, 4427–4434. Shaw, P. J., Ditewig, A. C., Waring, J. F., Liguori, M. J., Blomme, E. A., Ganey, P. E., and Roth, R. A. (2009). Coexposure of mice to trovafloxacin and lipopolysaccharide, a model of idiosyncratic hepatotoxicity, results in a unique gene expression profile and interferon gamma-dependent liver injury. Toxicol. Sci. 107, 270–280. Spahr, L., Rubbia-Brandt, L., Burkhard, P. R., Assal, F., and Hadengue, A. (2000). Tolcapone-related fulminant hepatitis: Electron microscopy shows mitochondrial alterations. Dig. Dis. Sci. 45, 1881–1884. Stricker, B. H., Blok, A. P., and Bronkhorst, F. B. (1986). Glafenine-associated hepatic injury. Analysis of 38 cases and review of the literature. Liver 6, 63–72. Svegliati, S., Cancello, R., Sambo, P., Luchetti, M., Paroncini, P., Orlandini, G., Discepoli, G., Paterno, R., Santillo, M., Cuozzo, C., et  al. (2005). Platelet-derived growth factor and reactive oxygen species (ROS) regulate Ras protein levels in primary human fibroblasts via ERK1/2. Amplification of ROS and Ras in systemic sclerosis fibroblasts. J. Biol. Chem. 280, 36474–36482. Taub, R. (2003). Hepatoprotection via the IL-6/Stat3 pathway. J. Clin. Invest. 112, 978–980. Thompson, R. A., Isin, E. M., Li, Y., Weidolf, L., Page, K., Wilson, I., Swallow, S., Middleton, B., Stahl, S., Foster, A. J., et  al. (2012). In vitro approach to assess the potential for risk of idiosyncratic adverse reactions caused by candidate drugs. Chem. Res. Toxicol. 25, 1616–1632. Torres, C. A., and Perez, V. I. (2008). Proteasome modulates mitochondrial function during cellular senescence. Free Radic. Biol. Med. 44, 403–414. Ulrich, R. G. (2007). Idiosyncratic toxicity: A  convergence of risk factors. Annu. Rev. Med. 58, 17–34. Venugopal, R., and Jaiswal, A. K. (1998). Nrf2 and Nrf1 in association with Jun proteins regulate antioxidant response element-mediated expression and coordinated induction of genes encoding detoxifying enzymes. Oncogene 17, 3145–3156. Wen, B., and Moore, D. J. (2011). Bioactivation of glafenine by human liver microsomes and peroxidases: Identification of electrophilic iminoquinone species and GSH conjugates. Drug Metab. Dispos. 39, 1511–1521. Will, Y., Hynes, J., Ogurtsov, V. I., and Papkovsky, D. B. (2006). Analysis of mitochondrial function using phosphorescent oxygen-sensitive probes. Nat. Protoc. 1, 2563–2572. Wu, H. M., Chi, K. H., and Lin, W. W. (2002). Proteasome inhibitors stimulate activator protein-1 pathway via reactive oxygen species production. FEBS Lett. 526, 101–105. Xu, J. J., Henstock, P. V., Dunn, M. C., Smith, A. R., Chabot, J. R., and de Graaf, D. (2008). Cellular imaging predictions of clinical drug-induced liver injury. Toxicol. Sci. 105, 97–105. Yamamoto, K., Yoshida, H., Kokame, K., Kaufman, R. J., and Mori, K. (2004). Differential contributions of ATF6 and XBP1 to the activation of endoplasmic reticulum stress-responsive cis-acting elements ERSE, UPRE and ERSE-II. J. Biochem. 136, 343–350. Zhang, M., Chen, M., and Tong, W. (2012). Is toxicogenomics a more reliable and sensitive biomarker than conventional indicators from rats to predict drug-induced liver injury in humans? Chem. Res. Toxicol. 25, 122–129. Zhen, J., Lu, H., Wang, X. Q., Vaziri, N. D., and Zhou, X. J. (2008). Upregulation of endothelial and inducible nitric oxide synthase expression by reactive oxygen species. Am. J. Hypertens. 21, 28–34. Zimmerman, H. J. (1999). Hepatotoxicity. The adverse effect of drugs and other chemicals on the liver, 2nd ed. Lippincott Williams & Wilkins, Philadelphia, PA.

Downloaded from http://toxsci.oxfordjournals.org/ at University of Cambridge on November 17, 2014

mitochondrial toxicity using fluorescence-based oxygen-sensitive probes. Toxicol. Sci. 92, 186–200. Kaplowitz, N. (2005). Idiosyncratic drug hepatotoxicity. Nat. Rev. Drug Discov. 4, 489–499. Kashimshetty, R., Desai, V. G., Kale, V. M., Lee, T., Moland, C. L., Branham, W. S., New, L. S., Chan, E. C., Younis, H., and Boelsterli, U. A. (2009). Underlying mitochondrial dysfunction triggers flutamide-induced oxidative liver injury in a mouse model of idiosyncratic drug toxicity. Toxicol. Appl. Pharmacol. 238, 150–159. Khetani, S. R., Kanchagar, C., Ukairo, O., Krzyzewski, S., Moore, A., Shi, J., Aoyama, S., Aleo, M., and Will, Y. (2013). Use of micropatterned cocultures to detect compounds that cause drug-induced liver injury in humans. Toxicol. Sci. 132, 107–117. Laifenfeld, D., Gilchrist, A., Drubin, D., Jorge, M., Eddy, S. F., Frushour, B. P., Ladd, B., Obert, L. A., Gosink, M. M., Cook, J. C., Criswell, K., Somps, C. J., Koza-Taylor, P., Elliston, K. O., and Lawton, M. P. (2010). The role of hypoxia in 2-butoxyethanol-induced hemangiosarcoma. Toxicol Sci. 113, 254–66. Lee, Y. H., Chung, M. C., Lin, Q., and Boelsterli, U. A. (2008). Troglitazoneinduced hepatic mitochondrial proteome expression dynamics in heterozygous Sod2(+/-) mice: Two-stage oxidative injury. Toxicol. Appl. Pharmacol. 231, 43–51. Leite, S., Martins, N. M., Dorta, D. J., Curti, C., Uyemura, S. A., and dos Santos, A. C. (2006). Mitochondrial uncoupling by the sulindac metabolite, sulindac sulfide. Basic Clin. Pharmacol. Toxicol. 99, 294–299. Li, A. P. (2002). A review of the common properties of drugs with idiosyncratic hepatotoxicity and the “multiple determinant hypothesis” for the manifestation of idiosyncratic drug toxicity. Chem. Biol. Interact. 142, 7–23. Liguori, M. J., Anderson, M. G., Bukofzer, S., McKim, J., Pregenzer, J. F., Retief, J., Spear, B. B., and Waring, J. F. (2005). Microarray analysis in human hepatocytes suggests a mechanism for hepatotoxicity induced by trovafloxacin. Hepatology 41, 177–186. Mingatto, F. E., dos Santos, A. C., Rodrigues, T., Pigoso, A. A., Uyemura, S. A., and Curti, C. (2000). Effects of nimesulide and its reduced metabolite on mitochondria. Br. J. Pharmacol. 131, 1154–1160. Moore, L., Rodman Davenport, G., and Landon, E. J. (1976). Calcium uptake of a rat liver microsomal subcellular fraction in response to in vivo administration of carbon tetrachloride. J. Biol. Chem. 251, 1197–1201. Nagy, G., Szarka, A., Lotz, G., Dóczi, J., Wunderlich, L., Kiss, A., Jemnitz, K., Veres, Z., Bánhegyi, G., Schaff, Z., et al. (2010). BGP-15 inhibits caspaseindependent programmed cell death in acetaminophen-induced liver injury. Toxicol. Appl. Pharmacol. 243, 96–103. Olson, H., Betton, G., Stritar, J., and Robinson, D. (1998). The predictivity of the toxicity of pharmaceuticals in humans from animal data–an interim assessment. Toxicol. Lett. 102-103, 535–538. Ong, M. M., Latchoumycandane, C., and Boelsterli, U. A. (2007). Troglitazoneinduced hepatic necrosis in an animal model of silent genetic mitochondrial abnormalities. Toxicol. Sci. 97, 205–213. Osna, N. A., and Donohue, T. M., Jr. (2007). Implication of altered proteasome function in alcoholic liver injury. World J. Gastroenterol. 13, 4931–4937. Piantadosi, C. A., and Suliman, H. B. (2006). Mitochondrial transcription factor A induction by redox activation of nuclear respiratory factor 1. J. Biol. Chem. 281, 324–333. Plaa, G. L. (2000). Chlorinated methanes and liver injury: Highlights of the past 50 years. Annu. Rev. Pharmacol. Toxicol. 40, 42–65. Porceddu, M., Buron, N., Roussel, C., Labbe, G., Fromenty, B., and BorgneSanchez, A. (2012). Prediction of liver injury induced by chemicals in human with a multiparametric assay on isolated mouse liver mitochondria. Toxicol. Sci. 129, 332–345. Rachek, L. I., Yuzefovych, L. V., Ledoux, S. P., Julie, N. L., and Wilson, G. L. (2009). Troglitazone, but not rosiglitazone, damages mitochondrial DNA and induces mitochondrial dysfunction and cell death in human hepatocytes. Toxicol. Appl. Pharmacol. 240, 348–354.