One proposed mechanism to explain some IDRs is that a mild underlying inflammatory stimulus lowers the therapeutic index for an otherwise safe drug [2]. Although this mechanism is unlikely to explain all IDRs, background information on a compound’s potential to exacerbate inflammation may be useful toward risk assessment of compounds in preclinical development. Currently, the most characterized inflammatory agent used to address the proposed mechanism is LPS [2,4,16]. LPS is a potent activator of the immune system and leads to a massive stimulation of gene transcription, especially in the bone marrow, spleen, and liver [18].

LPS is a useful inflammagen for mechanistic evaluation of drugs with known liabilities. However, the heterogeneous nature of LPS renders difficult the establishment of a standardized test protocol. Gene expression profiling confirmed that different preparations of LPS caused significantly different hepatic transcriptomic responses that were not attributable to individual animal variation. In contrast, TNFα resulted in remarkably consistent transcriptomic profiles regardless of preparation. Therefore, replacing LPS with TNFα could provide a model with enhanced reproducibility, reducing the need for extensive dose range finding studies.

In previous model characterizations, acute exposure to LPS at the doses used resulted in no overt liver injury [4]. Likewise, acute dosing with TNFα yielded no significant changes in serum chemistry or histology under these experimental conditions despite a clear signal at the level of transcription. The lack of a phenotypic effect is likely due to the acute dosing period (<6 h) and to the small dose administered. Both proinflammatory agents induced similar changes in several cell signaling networks, especially those related to innate immune response pathways. However, the LPS effects were broader, as evidenced by the greater number of pathways impacted and a more pronounced level of gene regulation.

LPS and other bacterial products mediate their effects partly via stimulation of toll-like receptors (TLRs), resulting in signal transduction and synthesis of various immune activators, such TRAF and NF-κB [19]. This results in rapid mobilization of immune mediators to the primary infection sites. LPS itself, via activation of receptors such as TLR4, mediates signaling to downstream MAP3Ks, which in turn are essential for secretion of native TNFα and for the subsequent innate immune response [20,21]. TNFα biosynthesis results from LPS stimulation of cells, and this cytokine mediates some of the effects produced by LPS. However, LPS stimulates activation of additional immune mediators, including IL-6 [22], IL-12 [23], MCP-1 [24], and IL-8 [25]. Therefore, the enhanced perturbation of gene expression by LPS relative to TNFα is not surprising.

Globally, TNFα/TVX treatment resulted in greater changes in the liver transcriptome compared to TNFα/LVX, despite the fact that LVX was administered at a 5-fold greater dose (comparable to the fold-difference in doses used clinically). From the TNFα/TVX group, genes involved with NF-κB activation such as RELA and TRAF2 were significantly induced. NF-κB activation occurs as a response to inflammatory stimuli and is integrally linked to TNFα signaling [26] TRAF2 regulates these signals via ubiquitin ligation to its targets (IκB), which releases NF-κB to exert its effects [27]. All of these perturbations were absent in the TNFα/LVX unique gene set, suggesting that the ability of TVX to potentiate an innate immune response is not shared by LVX. This supports the feasibility of evaluating novel drug development candidates for their ability to enhance an underlying inflammatory stimulus. Evaluation of additional IDR-inducing molecules would aid in the validation of such a model.

TNFα, acting through its receptors (TNFR1 and TNFR2), is a critical mediator for manifestation of TVX hepatotoxicity in LPS-treated mice [8,28]. Indeed, cotreatment with TNFα and TVX produced liver injury in mice [17]. Gene array data might facilitate elucidation of key mechanism(s) in this interaction. Furthermore, data from mice further support the choice of TNFα itself as an appropriate external inflammatory stimulus for these rodent models.

Also critical to the evaluation of the use of TNFα in lieu of LPS were the comparisons between TNFα/TVX and LPS/TVX responses. Unlike the mouse, there was no evidence of overt liver toxicity following TNFα/TVX administration at the early timepoint evaluated. The lack of a phenotypic effect at this early timepoint may reflect on the inability of TNFα to activate inflammatory signal transduction to the same extent as LPS in rats. The lack of an overt signal in rat may also suggest that mice are a more sensitive species to a TNFα/TVX interaction. For interpretation of gene expression data, early timepoints offer the advantage of enhanced insight into potential primary mechanisms of action. Indeed, genome-wide mRNA profiling of livers from TNFα/TVX-treated rats suggested a remarkable similarity in mode of action when compared to LPS/TVX in rats. More specifically, TNFα/TVX and LPS/TVX treatment perturbed similar immunological and stress response pathways, thus suggesting, despite differences in signaling robustness with no visible early injury, that treatment with LPS or TNFα ultimately led to similar toxicogenomic outcomes. Robust gene expression signals, such as those observed with TNFα/TVX treatment, are considered predictive of subsequent injury [10]. In addition, although a number of specific pathways did overlap between TNFα/TVX and LPS/TVX, a more important factor may be that TNFα/TVX cotreatment activates the necessary/key pathways that may contribute to liver pathogenesis.

Coadministration of TNFα/TVX resulted in marked autoinduction of TNFα mRNA, which was absent upon exposure to TNFα alone or TNFα/LVX. Transactivation of factors downstream of the TNFα receptor substantiated this observation for the TNFα/TVX-cotreated rats. Autoinduction could potentially perpetuate inflammatory cascades and immune cell activation. Indeed, in the mouse model, there is evidence for dysregulated amplification cycles in which TNFα enhances expression of other inflammatory cytokines, leading to further TNFα production [17]. This could eventually augment the presence of deleterious factors resulting in toxicity. In mice, TVX has been shown to be a critical mediator of hepatotoxicity at least partially through sustained TNFα signaling [17]. TNFα/TVX cotreatment primarily reduced the clearance of endogenous serum TNFα with a relatively smaller effect of increased production. This was a rare example of a xenobiotic reducing systemic clearance of a cytokine. In that mouse study, mRNA expression of hepatic TNFα was not evaluated. Here, using gene expression analysis, we showed that enhanced expression in liver represents one source of increased TNFα in serum. Interestingly, hepatic gene expression of TNFα was not elevated in livers of LPS/TVX–treated animals compared to those treated with LPS alone, suggesting that other factors also contribute to the toxicity. However, we cannot exclude the possibility that the failure to observe an influence of TVX on TNFα expression in LPS-cotreated rats was due to pharmacokinetic differences, i.e., differences in the time of initial TNFα exposure relative to sacrifice time in the LPS- and TNFα-cotreated groups.

The exact mechanism responsible for the TVX-induced autoinduction of TNFα is unknown and requires further investigation. Understanding this mechanism could lead to further insight into understanding TVX-induced hepatotoxicity since inflammatory mediators have been clearly linked to certain hepatotoxicities [29]. Relatively few reports exist describing this autoinduction effect. TNFα increases its own expression in some in vitro systems like human ZR-75-1 breast cancer cells [30] and rat tracheal epithelial cells [31]. In HL60 leukemia cells, autoinduction of TNFα required phospholipase A₂ and lipoxygenase activity with subsequent release of arachidonic acid metabolites [32]. In the present study, phospholipase A₂ activating protein (Plaa) mRNA was overexpressed in the TNFα and TNFα/TVX treated rats, but was not consistently increased in the TNFα/LVX group. No change was observed in lipoxygenase expression, but phospholipase A₁ (Pla1a) was increased in all rats treated with TNFα either alone or in combination with TVX or LVX.

All animal experiments for this study were conducted in accordance with the Guiding Principles in the Use of Animals in Toxicology (Anonymous 2002) and were approved by Abbott’s Institutional Animal Care and Use Committee (IACUC). Male Sprague-Dawley rats [Crl:CD^®(SD)IGS BR], weighing ∼250 g at study initiation were obtained from Charles River Laboratories, Inc. (Wilmington, MA). Rats were housed singly in ventilated, stainless steel, wire-bottom hanging cages and fed non-certified Rodent Chow (Harlan Labs, Madison, WI) and water ad libitum.

After 24 h fasting, the rats were injected with recombinant rat TNFα (Sigma Chemical, St Louis, MO; lot numbers 126K1053, 087K1290, or 098K1865) via tail vein (i.v.) at a dose of 0.01 mg/kg body weight with a delivery volume of 1 mL/kg. This dose of TNFα was based on rat studies contained in the DrugMatrix™ database [33]. For the lot comparison analysis, TNFα was given in separate studies which were conducted several months apart. Two hours after TNFα administration, rats were treated (i.v.) with TVX (30 mg/kg), LVX (150 mg/kg) or vehicle (dextrose 5% in water with 0.1 N equivalent HCl) in a volume of 5 mL/kg. The previous LPS rat model served as a basis for dose selection with a lower TVX dose due to formulation restriction [4]. Treatment nomenclature was designated as follows: Veh/Veh (n = 3), TNFα/Veh (n = 9), Veh/TVX (n = 3), Veh/LVX (n = 3), TNFα/TVX (n = 3), and TNFα/LVX (n = 3). TNFα (lot #126k1053) was used for coadministration with TVX and LVX. Four hours after the second treatment, the rats were sacrificed using CO₂ anesthesia and blood was collected. A portion of the liver was flash frozen in liquid nitrogen, and the remaining liver was preserved in 10% neutral-buffered formalin. One TNFα/LVX rat was misdosed and was therefore excluded from further analysis.

Two different lots of LPS were used to compare hepatic gene expression changes. These studies were conducted at Michigan State University (MSU, East Lansing, MI) using a protocol similar to that previously described [4]. Rats received humane care according to the criteria outlined in the Guide for the Care and Use of Laboratory Animals (1996) prepared by the National Academy of Sciences, and procedures were approved by the Michigan State University Committee on Animal Use and Care. Male Sprague-Dawley rats [Crl:CD (SD)IGS BR; Charles River, Portage, MI] (n = 5) weighing 250–350 g were used for these studies. Animals were fed standard chow (rodent chow/Tek 8640; Harlan Teklad, Madison, WI) and allowed access to water ad libitum. In one study, LPS derived from Escherichia coli serotype O55:B5 with an activity of 9.2 × 10⁶ EU/mg was used (catalog number L-2880, Lot 024K4067; Sigma-Aldrich, Inc., St. Louis, MO), designated in Figure 1 as Lot #2. This activity was determined using a QCL Chromogenic LAL Endpoint Assay from Cambrex (East Rutherford, NJ). Rats fasted for 24 h were given 44.4 × 10⁶ EU/kg LPS or its saline vehicle (Veh) i.v., and food was then returned. Two hours later, vehicle (50/50 sterile saline/sterile water) was administered i.v. Two hours later, rats were anesthetized with sodium pentobarbital (75 mg/kg i.p.) and euthanized by exsanguination. The right medial lobe of the liver was flash-frozen in liquid nitrogen for subsequent gene expression analysis. For Lot #1 in Figure 1 (Lot 51K4115), data files were used from a previously published study [34] in which rats were treated with LPS (from E. Coli, serotype 055:B5) and vehicle using the same dose (44.4 × 10⁶ EU/kg LPS) and treatment protocol as described for Lot #2.

Serum clinical chemistry parameters for the TNFα study were quantified using an Aeroset Clinical Chemistry Analyzer (Abbott Laboratories, Abbott Park, IL) and included: alanine amino transferase (ALT), aspartate amino transferase (AST), gamma glutamyltransferase (GGT), and alkaline phosphatase (ALP) activities. Formalin-fixed liver samples (left and right lobes) were embedded in paraffin and sections (6 μm) were stained with hematoxylin and eosin.

Frozen liver samples (approximately 100 mg of tissue per sample) were immediately added to 2 mL of QIAzol reagent (Qiagen, Valencia, CA) and homogenized using a Polytron 300D homogenizer (Brinkman Instruments, Westbury, NY). One mL of the tissue homogenate was transferred to a microfuge tube, and total RNA was extracted via chloroform extraction followed by nucleic acid precipitation with isopropanol. The pellet was washed with 80% ethanol and resuspended in molecular biology grade water. Nucleic acid concentration was determined by O.D. 260 nm (Smart-Spec; Bio-Rad Laboratories, Hercules, CA), and RNA integrity was evaluated using a bioanalyzer (model 2100; Agilent Technologies, Foster City, CA).

Microarray analysis of rat liver samples was performed using the standard protocol provided by Affymetrix, Inc. (Santa Clara, CA) as previously described [4] using Affymetrix Rat Genome RAE 230_2.0 or RAE230A arrays containing ∼31,000 or ∼15,000 probe sets (genes) respectively. Resolver software (Version 7.2; Rosetta Informatics, Seattle WA) was used to analyze the microarray data (available in supplemental Table 1). Ingenuity Pathways Analysis software (Version 8.0 Ingenuity Systems, Redwood City, CA) was applied for pathway evaluations. For TNFα/TVX comparisons to LPS/TVX, the original data files from the previous LPS/TVX study were used [4].

Clinical pathology parameters and mRNA fold changes were analyzed by analysis of variance (ANOVA) with Tukey’s post hoc test using GraphPad Prism software (Version 5; La Jolla, CA). The criterion for significance was p ≤ 0.05. For hepatic microarray analysis, the scanned image and intensity files (.cel files) were imported into Resolver software. Gene expression ratios were built for each drug-treated animal versus the respective vehicle-treated animals combined in silico using the Resolver error model. All hierarchical cluster and principal component analyses were completed with Resolver software. Gene expression data for each rat were entered into Ingenuity software (combined in silico), which then ranked and calculated percent pathway impact (number of genes regulated/total number of genes in pathway) with a corresponding p-value for each treatment. Pathways impacted similarly and differentially upon exposure to the respective treatment were identified. In a given analysis, the number of genes/probe sets represented depends on the statistical cutoffs and stringency parameters for each separate analysis, resulting in a slightly different magnitude of probe sets presented. All analysis cutoffs are listed in each figure description.