NSAID-Induced Acute Phase Response is Due to Increased Intestinal Permeability and Characterized by Early and Consistent Alterations in Hepatic Gene Expression

Abstract

Toxicogenomics using a reference database can provide a better understanding and prediction of toxicity, largely by creating biomarkers that tie gene expression to actual pathology events. During the course of building a toxicogenomic database, an observation was made that a number of non-steroidal anti-inflammatory compounds (NSAIDs) at supra-pharmacologic doses induced an acute phase response (APR) and displayed hepatic gene expression patterns similar to that of intravenous lipopolysaccharide (LPS). Since NSAIDs are known to cause injury along the gastrointestinal tract, it has been suggested that NSAIDs increase intestinal permeability, allowing LPS and/or bacteria into the systemic circulation and stimulating an APR detectable in the liver. A short term study was subsequently conducted examining the effects of aspirin, indomethacin, ibuprofen, and rofecoxib to rats and a variety of endpoints were examined that included serum levels of inflammatory cytokines, histologic evaluation, and hepatic gene expression. Both indomethacin and ibuprofen injured the gastrointestinal tract, induced an APR, and increased serum levels of LPS, while rofecoxib and aspirin did not affect the GI tract or induce an APR. In treatments that eventually showed a systemic inflammatory response, hepatic expression of many inflammatory genes was noted as early as 6 hours after treatment well before alterations in traditional clinical pathology markers were detected. This finding led to the creation of a hepatic gene expression biomarker of APR that was effectively shown to be an early identifier of imminent inflammatory injury. In terms of the relative gastrointestinal safety and the NSAIDs studied, an important safety distinction can be made between the presumptive efficacious dose and the APR-inducing dose for indomethacin (1—2-fold), ibuprofen (5-fold), and rofecoxib (~250-fold). Our data support the notion that NSAID-induced intestinal injury results in leakage of commensural bacteria and/or LPS into the circulation, provoking a systemic inflammatory response and that hepatic gene expression-based biomarkers can be used as early and sensitive biomarkers of APR onset.

[The table referenced in this paper is not printed in this issue of Toxicologic Pathology. It is available as a downloadable text file in the online edition of Toxicologic Pathology, 34(2). In order to access the full article online, you must have either an individual subscription or a member subscription accessed through www.toxpath.org.]

Introduction

Over the past several years, the use of genome-wide gene expression profiling of drugs and drug development candidates has uncovered the mechanisms of cellular injury, organ toxicity, and pharmacological activity at the gene expression level (Fielden et al., 2004). This new mechanistic knowledge will have a positive impact on the accuracy and sensitivity of candidate selection. Importantly, toxicogenomics has been increasingly used to identify biomarkers of efficacy and/or toxicity that are often more sensitive and accurate than approaches traditionally used for preclinical optimization and clinical development.

A toxicogenomics reference database has been built containing information on 638 drugs and chemicals of which 433 are FDA approved drugs (Ganter et al., 2005). The value created from mining vast amounts of gene expression data has allowed the linking of gene expression changes with traditional measurements of pharmacological and toxicological activity. This connection of traditional and novel data provides context and meaning to the alterations in gene expression caused by a candidate drug or chemical (Amin et al., 2002; Bushel et al., 2002; Hamadeh et al., 2002a, 2002b; Steiner et al., 2004; Fielden et al., 2005; Ruepp et al., 2005). Further, this database has been mined using rigorous, statistical approaches based on sophisticated classification algorithms and logistic regression producing a library of linear, robust binary classifiers (Ganter et al., 2005; Natsoulis et al., 2005). These classifiers can be used to characterize, diagnose, and predict pharmacologic and toxicologic properties using gene expression data (Steiner et al., 2004; Ganter et al., 2005, Natsoulis et al., 2005).

Leveraging a toxicogenomic database, the pathological impact of nonsteroidal anti-inflammatory drugs (NSAIDs) on hepatic gene expression as a result of gastrointestinal toxicity was examined. NSAIDs are widely used for the clinical management of pain and inflammation and are known to induce a well-recognized yet poorly understood toxicity towards the gastrointestinal tract (Smecuol et al., 2001; Hawkey, 2002; James and Hawkey, 2003). Clearly, lack of selectivity towards the cycloxygenase (COX)-1 and COX-2 isoforms is a primary mechanism by which GI risk is exacerbated. Data supporting this concept compare older, relatively more GI toxic drugs such as ibuprofen, indomethacin, and aspirin, which are non-selective COX-1 inhibitors, with COX-2-selective NSAIDs such as rofecoxib (Vioxx^TM) and celecoxib (Celebrex^TM) that were successfully and specifically designed to avoid GI toxicity. Several clinical studies have shown that rofecoxib is significantly safer towards the GI tract than are ibuprofen and other older NSAIDs (Laine et al., 1999; Bombardier et al., 2000; Hawkey et al., 2000).

In the course of building our toxicogenomics database, many NSAIDs were observed to elicit clinical pathology and gene expression changes similar to those produced by intravenous administration of endotoxin (LPS) (Table 1). Since NSAIDs are known to compromise the integrity of the gastrointestinal barrier (Smecuol et al., 2001; Hawkey, 2002; James and Hawkey, 2003), it was hypothesized that NSAID-induced GI injury releases pro-inflammatory bacteria and/or LPS into systemic circulation and into the liver, where such foreign material stimulates an APR in Kupffer and other cells. A study was performed where rats were treated with NSAIDs (aspirin, rofecoxib, indomethacin, and ibuprofen), examined for GI toxicity, monitored for signs of APR using traditional means and gene expression of the liver. A goal of this study was to determine whether increased gastrointestinal permeability is responsible for the systemic acute phase response observed in rats. A second goal was to determine the ability of hepatic gene expression to be a sensitive indicator of NSAID-mediated APR. Our studies indicate rofecoxib requires much larger doses relative to the efficacious dose compared to ibuprofen and indomethacin in order to induce an APR. More importantly, we demonstrate that a gene expression-based biomarker can predict as well as diagnose the future occurrence of NSAID-mediated APR.

Methods

Reference Database (DrugMatrix^TM)

A reference database containing gene expression data, molecular pharmacology data, curated literature information, clinical chemistry, hematology, and histopathology data derived from short-term in vivo studies of drugs, toxicants, and pharmacological standards was built in a rigorous and consistent manner (Ganter et al., 2005). Briefly, male Sprague–Dawley rats aged 6–8 weeks were treated daily at 1 of 2 dose levels (maximum tolerated and fully effective) and necropsied after 0.25, 1, 3, or 5 (or in limited cases 7 to 28) days of dosing. The 0.25 day animals received 1 dose of compound and were sacrificed 6 hours later, the 1 day animals received a single dose of compound and were sacrificed 24 hours later, the 3 day animals received 3 daily compound doses and were sacrificed 24 hours after the final dose, and so on for the lengthier treatments. The high dose represents the upper limit of tolerability as determined by a 5-day range finding study while the low dose was intended to approximate the fully effective pharmacologically active dose. Twelve (12) tissues were collected, RNA was extracted, and microarray hybridization (RU1 CodeLink rat microarray, General Electric) was performed on each selected tissue from each animal according to previously described protocols (Ganter et al., 2005). The signal intensity for each probe from the three animals per time and dose group were averaged and are expressed as the mean log₁₀ ratios relative to an averaged group of vehicle-treated controls.

1- and 3-Day Gazstrointestinal and Hepatotoxicity Study of Aspirin, Indomethacin, Ibuprofen, Rofecoxib, and LPS in Male Rats

Male Charles River Sprague–Dawley rats, acclimated after receipt for 5–7 days, were 6–8 weeks old (approximately 150–250 grams), at the time of study initiation. Animals were maintained at 30–70% humidity in an approximate 12-hour light, 12-hour dark cycle, housed individually on corn cob bedding, and fed Certified Purina Rodent Chow 5002 and municipal water ad libitum. At study initiation, the rats were randomized by weight into groups of three for dosing. The test compounds were indomethacin (5 and 10 mg/kg/dose), ibuprofen (90 and 275 mg/kg/dose), rofecoxib (800 and 250 mg/kg/dose), aspirin (167 and 500 mg/kg/dose) and LPS (1.25 mg/kg/dose).

Indomethacin and aspirin were purchased from Sigma-Aldrich (St Louis, MO); LPS from EMD Biosciences (San Diego, CA), ibuprofen and rofecoxib were obtained from Sequoia (Oxford, United Kingdom). Indomethacin, ibuprofen, rofecoxib, and aspirin were administered by oral gavage at 10 mL/kg of body weight using a 16-gauge stainless steel feeding needle. The appropriate amount of oral test article was suspended in corn oil, and control animals received only the corn oil vehicle. LPS was given intravenously in 2.5 mL of saline per kg of body weight in the tail vein to achieve a dose of 1.25 mg/kg.

Animals were dosed daily for 1 day and 3 days, with dosing occurring each day within ±1 hour. The 1 day treatment animals were sacrificed 24 hours after a single dose, whereas the 3-day treated animals were sacrificed 24 hours after the third of 3 doses. Blood was drawn once at the time of sacrifice from the inferior vena cava after CO₂ anesthesia. Necropsy, tissue processing, and histology were performed by Seventh Wave Laboratories (St. Louis, MO) as per a recent publication (Ganter et al., 2005). Microarray analysis was performed on liver tissue that was flash frozen in liquid nitrogen.

Rat cytokine and LPS measurements were performed on flash frozen serum samples by Linco Research (St. Charles, MO) and Associates of Cape Cod Inc, (Falmouth, MA), respectively. Clinical chemistry (cholesterol, aspartate aminotransferase, alanine aminotransferase, total bilirubin, triglycerides, alkaline phosphatase, creatinine, blood urea nitrogen, glucose, phosphorus, calcium, total protein, albumin, lactate dehydrogenase, creatine phosphokinase, albumin to globulin (A/G) ratio, sodium, potassium, and chloride) and hematology (numbers of white blood cells, red blood cells, platelets, erythrocytes, segmented neutrophils, band neutrophils, lymphocytes, monocytes, eosinophils, basophils, and nucleated red blood cells, and measurements of hemoglobin, hematocrit, mean corpuscular volume, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, red cell distribution width, and mean platelet volume) assays were performed by the St. Louis University Department of Comparative Medicine. Occult blood was analyzed in cecal ingesta using the Hemoccult ^® Fecal Occult Blood Test from Beckman Coulter (Fullerton, CA).

Drug Signature Derivation

Mining large-scale gene expression data in order to derive Drug Signatures, or multigene biomarkers, has been previously described (Natsoulis et al. 2005). In brief, a Drug Signature is a mathematical model that can discriminate two classes of experiments using their associated gene expression values as variables. Natsoulis et al. evaluated several supervised classification algorithms, and chose an algorithm (sparse linear program) related to support vector machines because of superior performance and interpretability. In order to derive a Drug Signature, the two classes of experiments that are to be distinguished are labeled independently of knowledge of the gene expression results. The signatures are trained on a fraction of the data and then tested against the unseen portion. This training / testing paradigm is iterated through at least 20 random partitions of the data set for each signature, and the results are averaged to obtain the best estimate of a given signature’s “real-world” performance on novel data. A signature is validated by splitting the data such that 60% of the experiments are used for training the signature and 40% of the experiments are used to test the signature.

Drug Signatures are applied through the use of Scalar Product Scores. Scalar products are calculated using the following equation:

$$\text{S}=\mathrm{\Sigma }{\text{w}}_{\text{i}}{\text{x}}_{\text{i}}-\text{b},$$

where w_i are the weights for each gene assigned by the algorithm, x_i are the log₁₀ ratios values for each signature gene in that sample, and b is a bias term (Natsoulis et al., 2005). Signatures are binary classifiers with zero as the decision boundary and as such a scalar product greater than zero indicates that the sample is a member of the positive class, whereas a scalar product less than zero indicates that the sample is not a member of the positive class. Forward validation of a signature involves evaluating gene expression data from compound experiments not used in the signature derivation by determining the signature score or scalar product.

Drug Signatures for the Acute Phase Response

To generate a signature specifically designed to classify an acute phase response in liver based solely on changes in gene expression, microarray data were separated into two classes: one from compound-treated rats experiencing an APR and another from compound-treated rats not experiencing an APR. Serum albumin levels were used as a surrogate to separate each drug-dose-time combinations (treatments) into one of two classes. Serum albumin levels decreased by at least 1.2-fold from normal were considered positive for an active APR. A signature was also derived to detect early events associated with an acute phase response prior to known biomarkers. This “early response” signature was derived by using microarray data solely from animals necropsied 24 hours after a single dose of compound. These experiments were placed into the positive or negative class based on the serum albumin levels found in cohort treatments from animals that received 5 daily doses of compound.

The positive group was defined as those 1 day treatments whose 5 day cohort treatments showed 1.2- to 2-fold decreases in serum albumin. The 17 drug-dose combinations represent the following 16 unique compounds: 4,4′-methylenedianiline, 4-octylphenol, aminosalicylic acid, chloroxylenol, diclofenac, gentian violet, ibuprofen, indomethacin, isotretinoin, LPS, meloxicam, nystatin, oxiconazole, pramoxine, sulindac, and trichloroacetic acid. The negative class was defined as the 1-day treatments whose 5 day cohort treatments had essentially no change in serum albumin; the range was from a 1.03-fold decrease to about a 1.05 increase. The negative class encompassed 106 treatments representing 89 compounds. The remainder of the 1 day data was excluded from signature derivation (Natsoulis et al., 2005).

Results

While analyzing gene expression profiles of the NSAIDs, an unexpected connection between gene expression and classical toxicological parameters for anti-inflammatory NSAID-type drugs and the pro-inflammatory molecule lipopolysaccharide (LPS, or endotoxin) was observed. In particular, both NSAIDs and LPS elicited similar gene expression changes in the liver consistent with a systemic acute phase response (APR) (Figure 1). In this figure, gene expression data from treatments with 4 different classes of compounds, steroids, PPARalpha agonists, HMG CoA reductase inhibitors (statins) and LPS were clustered by principal components analysis. Each of the four classes forms a distinct arm emanating from an origin, where treatments that induce few changes cluster. Both high dose NSAIDs and LPS treatments cluster together along principal component axis 3 and the genes that drive this cluster pattern are primarily involved in the acute phase response.

Specifically, of the top 20 genes in principal component 3, at least ten are known to have a role in the inflammatory response. The 20 genes (ranked in order of importance for PCA; inflammatory genes noted with asterisks) are lipocalin 2^*, lipopolysaccharide binding protein^*, alpha-2-macroglobulin^*, S100 calcium binding protein A9 (calgranulin B)^*, chemokine (C-X-C motif) ligand 1^*, guanylate cyclase 2C^*, amiloride-sensitive cation channel 4 pituitary, putative SH3BGR protein, procollagen C-proteinase enhancer protein, phosphatidylserine-specific phospholipase A1, colony stimulating factor 2 receptor beta 1 low-affinity (granulocyte-macrophage)^*, follistatin^*, gp49B2, follistatin^*, regulator of G-protein signaling 4, heme oxygenase (decycling) 1^*, fatty acid binding protein 4, adipocyte, inositol polyphosphate 5-phosphatase, orosomucoid 1^*, and squalene epoxidase.

In contrast, aspirin, which has anti-inflammatory properties manifested through a different mechanism than the NSAIDs that cause an APR, did not induce an APR as indicated by gene expression (Figure 1).

In addition to sharing similarities in gene expression, high dose NSAIDs and LPS shared similarities in clinical chemistry and hematology pathology. The concentration of serum albumin, an indicator of APR (Kaysen, 1998; Margarson and Soni, 1998; Franch-Arcas, 2001), was decreased in high dose NSAID- as well as LPS-treated animals (Table 1). Both LPS and high doses of NSAIDs generally caused increases in neutrophil counts, also an indication of an APR (data not shown). In addition, high doses of NSAIDs caused erythrocyte counts to decrease, suggesting compromise of the GI tract, and leading to loss of blood (Table 1). Aspirin did not induce any similar changes in either clinical chemistry or hematology (Table 1). Together, the gene expression findings and the clinical chemistry and hematology findings confirm that a number of the high dose NSAIDs and LPS induce an acute phase response in rats. This paradoxical finding of NSAIDs inducing a pro-inflammatory state was the observation which initiated a subsequent study to investigate this potential adverse toxicological phenomenon. The hypothesis was that NSAIDs induce intestinal permeability changes that allow systemic exposure to LPS, which induces an acute phase response (Figure 2). The ability of gene expression based biomarkers to be able to detect as well as precede changes in traditional parameters was also examined.

In order to test this hypothesis, a study was conducted by treating rats with several NSAIDs that were previously studied and shown to induce various degrees of APR (Table 1). Of the 4 compounds used in this study, 2 were previously shown to elicit a profound acute phase response at high doses (ibuprofen and indomethacin), while aspirin did not induce an APR at a high dose, and rofecoxib produced a weak response at a high dose (for study design see Table 2). The low dose for each NSAID was the efficacious dose used in rat inflammation models, while the high dose was the upper limit of tolerability as established from our range finding studies and previous array studies (Ganter et al., 2005). LPS was used as a positive control and intravenously administered daily at the 1.25 mg/kg

Body weight gain was suppressed over the 3 days of the study at the high dose of LPS and with all other compounds except rofecoxib (Figures 3A and 3B). For rofecoxib and to a lesser degree, aspirin, the high dose level had minimal generalized toxicity despite being administered at ~250× and ~50× their pharmacologic dose, respectively. The low doses of aspirin, ibuprofen, and rofecoxib had no effect on weight gain over the three-day study. The low dose of indomethacin did suppress weight gain (Figure 3A), indicating at the efficacious dose in rodents, rats are sensitive to indomethacin-induced toxicity.

Gross and microscopic gastrointestinal pathology findings were evident at the high dose of both ibuprofen and indomethacin. A representative segment of the jejunum of the indomethacin-treated rats after 3 days of dosing at high doses highlights the presence of transmural ulcers (Figure 3C). The ibuprofen-treated small intestine was similar in appearance (data not shown). The ceca from indomethacin and ibuprofen-treated rats were dark and contained intraluminal blood (melena) and had a reduced quantity of ingesta (Figure 3D). These observations were most pronounced with indomethacin. The ceca from rofecoxib- and aspirin-treated animals were normal in appearance. Gross intestinal injury of indomethacin-treated rats has been well documented (Yamada et al., 1993; Bertrand et al., 1998), but not as thoroughly for ibuprofen. Histologically, either multifocal epithelial ulceration or inflammatory cell infiltrates were observed within the small intestine of ibuprofen- and indomethacin-treated rats. No treatment-related gastrointestinal pathology findings were noted in rofecoxib-, LPS-, or aspirin-treated animals.

Clinical pathology data confirmed the presence of an APR in the high dose of indomethacin- and ibuprofen-treated rats. The high doses of indomethacin and ibuprofen reduced levels of serum albumin, caused fecal occult hemoglobin, sporadically elevated serum levels of LPS, and increased the serum protein markers of inflammation, MCP-1 and GRO-1/KC (Table 3). Intravenous administration of LPS likewise sporadically increases the inflammatory markers MCP-1 and GRO-1/KC and increased the systemic levels of LPS, but did not induce fecal occult hemoglobin. By comparison, the low doses of the drugs did not alter any of the inflammatory markers consistent with their lack of damage to the gastrointestinal tract.

The nature and degree of the intestinal injury and acute phase response were further characterized through hepatic gene expression analysis. Examination of the genes induced in common across the treatments revealed that LPS, indomethacin, and ibuprofen increased the relative expression of a number of inflammatory response genes (Figure 4). Hierarchical clustering of known inflammatory genes such as MCP-1, GRO-1, ceruloplasmin, fibrinogen, LPS-binding protein, S100, alpha-2 macroglobulin, orosomucoid 1, and lipocalin 2 resulted in LPS, high-dose ibuprofen, and both doses of indomethacin residing within a single cluster (correlation coefficient of 0.6). The other treatments, rofecoxib and aspirin, perturbed very few inflammatory genes. Interestingly, although the serum levels of albumin decreased after several days of inflammation-inducing NSAID or LPS administration, the mRNA for albumin did not change in any treatment.

The next step was to leverage the reference database to create a gene expression-based biomarker, or signature, from treatments that induce an APR. The signature described herein utilizes gene expression data from 17 positive treatments (16 unique compounds) and 106 negative treatments (89 unique compounds) from a 24-hour time point. The early time point was selected with the rationale of being capable of identifying treatments that are about to or are actively experiencing APR. The signature was created using an algorithm similar to the support vector machine algorithm, which has been employed widely as binary classifier (Natsoulis et al., 2005). The algorithm identifies a relatively short list of genes (i.e., 22 genes in this case) that mathematically distinguish two sets of treatments based on their gene expression. A signature “score” for a given signature is determined by calculating the sum of the gene expression values for signature gene multiplied by the algorithm-determined weight (the “impact”) plus a bias term (Natsoulis et al., 2005). A positive number indicates the compound is in the class, in this case, an APR-causing molecule. If the signature score is negative, the treatment is not in the class. Importantly, the genes in the signature are selected by the algorithm based on performance to create a linear separation between the 2 classes of treatments and not based on biology. Thus, examination of the biological properties of the genes in the signature can lend insight into the biological “meaning” for a signature in an unbiased manner.

For the APR signature, 22 genes are sufficient to create an APR classifier (Table 4). Shown in this table are the hepatic gene expression values for 5 representative positive or APR-inducing compounds (indomethacin, diclofenac, nystatin, meloxicam, and pramoxine) and 5 negative or non-APR inducing compounds (valproic acid, ethylene glycol, griseofulvin, norethindrone acetate, and propylthiouracil). The positive compounds perturb the expression of several of the genes contained in the signature; in particular, lipocalin 2, complement component 9, and guanylate cyclase 2C are known to have roles in inflammation. By contrast, the negative compounds alter the expression of very few of the signature genes. The 22 genes are used to calculate the scalar product (score), which shows all positive compounds are greater than 1 and in the APR class while the negative compounds are less than 0 (i.e., not in the class). This result is an expected outcome of the signature, and the next step is to examine the performance of the signature on independent samples.

The concordance of the signature results relative to the measured serum albumin was examined on a set of independent samples (Table 5). A large number of treatments from the toxicogenomic reference database were stratified into 3 groups based on serum albumin levels; less than 1.05 fold decrease (i.e., no change), greater than 1.2-fold change (i.e., mild and greater APR), and greater than 1.25 fold change (i.e., moderate and greater APR). In 795 treatments (i.e., dose-time point combinations) from 311 compounds where no change in albumin was noted, 753 were correctly called as negative by the gene expression signature (94.7% true negative rate). In 68 treatments (40 compounds) of mild and greater albumin decrease, 52 were correctly identified as positive (76.5% true positive rate). If the threshold for albumin decrease is moved to 1.25-fold change, 40 of 44 are correctly identified as positive (90.9%).

The response of the signature outcome relative to the time course of the change in actual serum albumin was examined for 13 compounds that were not included in the original signature derivation (Table 6). A positive match to the APR signature was observed as early as 6 hours (the shortest time point examined) after a single high dose of ibuprofen, isotretinoin, or naproxen, while nimesulide was very close to the decision threshold (−0.02). At the 0.25-day time point, serum albumin levels are normal, and only begin to decrease after 1 or 3 days. At the later time points, a consistent relationship between decreased albumin and the signature score is seen for all compounds.

Each of the treatments in the study described herein along with the treatments not available at the time of signature derivation were scored against the APR signature. Those treatments that induced an APR (e.g., indomethacin, ibuprofen, and the LPS treatment) scored positive (Figure 5), as well as a number other treatments, including other NSAIDs. Generally, the magnitude of a positive signature score correlates with the magnitude of serum albumin reduction, in particular when serum albumin levels decrease by ~12%. As discussed above, the signature can identify treatments that cause an APR before serum albumin is reduced. In Figure 5, these treatments are LPS (1.25 mg/kg) at one day, IL-1β (0.01 mg/kg) at 1 day, and TNF-α (0.25 mg/kg) at one day, all known pro-inflammatory molecules. LPS at 3 days still matched the signature and also reduced serum albumin by 10% (IL-1 and TNF were not tested later than 1 day).

Discussion

Many NSAIDs, widely used for clinical management of pain and inflammation, induce a well-documented yet poorly understood toxicity towards the gastrointestinal tract (Smecuol et al., 2001; Hawkey, 2002; James and Hawkey, 2003). In this paper, the paradoxical observation of NSAIDs inducing a proinflammatory state was confirmed and investigated further. The hypothesis was confirmed that NSAIDs can damage the gastrointestinal epithelium and allow LPS access to the systemic environment, leading to an acute phase response (APR). Extending on this putative mechanism of injury, a set of 22 genes was identified whose hepatic expression can serve as an early sentinel marker of APR when used as part of a toxicogenomic evaluation.

In this study, high doses of indomethacin and ibuprofen damaged the gastrointestinal tract and caused an APR. The degree of intestinal injury caused by these compounds was severe enough that intestinal epithelial integrity was compromised as evidenced by detectable levels of LPS in the serum in some of the animals and occult hemoglobin appearing in the feces. Clinical pathology suggested APR, in that albumin was reduced, and the blood levels of neutrophils and the inflammatory cytokines MCP-1 and GRO/KC were each elevated. While it is widely known that chronic NSAID use can cause ulceration of the stomach, the consequences of NSAID use on the intestines are less well publicized. Despite the lesser attention to the gut, an estimated 16,500 or more people die annually in the U.S. from complications associated with long term NSAID use (e.g., intestinal ulceration, bleeding, perforations, and strictures (Hirschowitz, 1994; Laney et al., 1994)), especially among elderly, arthritic patients (Wolfe et al., 1999). Moreover, an appreciated aspect of persistent low-level LPS exposure due to NSAID use is elevated risk of drug-induced idiosyncratic hepatic injuries (Roth et al., 2003), atherosclerosis (Blake and Ridker 2001; Shishehbor and Bhatt, 2004) and alcohol-induced hepatic injury (Mathurin et al., 2000; Schafer et al., 2002). To this point, pretreatment of rats with a mild APR inducing dose of LPS followed by a dose of ranitidine that in naïve animals is only mildly hepatotoxic synergistically increases hepatic injury (Luyendyk et al., 2003).

Gastrointestinal toxicity of NSAIDs has been alleviated by developing molecules that are selective for the inducible isoform (COX-2) over the constitutive form (COX-1) (Laine, 2002; Pronai et al., 2003, Laporte et al., 2004; Micklewright et al., 2003). It has been suggested that GI toxicity arises when both COX-1 and COX-2 enzymes are inhibited, and that inhibiting either one alone does not cause injury to the gastrointestinal mucosa (Takeuchi et al., 2003; Tanaka et al., 2002a, 2002b). Our data presented herein may align with this hypothesis, as rofecoxib did not induce an APR at 250-fold its presumptive efficacious dose (Riendeau et al., 2001). By contrast, the COX-1 inhibitor indomethacin has an effective dose of 3 mg/kg (Romay et al., 1998), and produced an APR at 5 mg/kg in this study, resulting in an extremely narrow gastric safety window of 1.7-fold. Ibuprofen has an effective dose of 54 mg/kg (Perianin et al., 1988), an APR-inducing dose of 275 mg/kg, and therefore a calculated gastric safety margin of 5-fold. Although not tested, it may be that at the high doses, indomethacin or ibuprofen rapidly achieve dual inhibition of both COX-1 and COX-2 in the intestine, while rofecoxib maintains its selectivity. Irrespective of its favorable GI safety, rofecoxib was recently withdrawn voluntarily from the market by its manufacturer due to an increased risk of cardiovascular events (Burnier, 2005).

APR is a component of innate immunity normally arising in response to exposure to foreign matter such as bacteria, and tissue injury such as that induced by surgery, burns, and damage due to exposure to toxic compounds (Koj, 1985; Yoo and Desiderio, 2003). A major stimulus of APR is LPS, a component of the cell wall of gram-negative bacteria, which activates a signal through TLR-4 on antigen presenting cells such as macrophages and dendritic cells. These cells respond by producing cytokines such as tumor necrosis factor-α, IL-1β, and IL-6, which induce or repress expression of a number of proteins that work in concert to limit a bacterial infection. The proteins affected by an APR include complement components that serve an anti-infective role, procoagulation factors, and proteins that promote fatty acid synthesis and increased breakdown of lipids and glycogen. As discussed next, the gene expression modulation that occurs during APR can be utilized in part as an early detector of inflammation.

A gene expression-based biomarker (Drug Signature) was created that can identify drug treatments causing, or on the verge of causing, an APR due to GI toxicity. Interestingly, for some treatments, the biomarker described herein can detect inflammation from hepatic gene expression changes within six hours (the shortest time point examined) after dosing. Although the treatments used to create the signature demonstrated clinically measurable decreases in serum albumin, in some cases, the Drug Signature identified early changes in gene expression preceding the eventual decrease in albumin by several days. For example, the hepatotoxicant chloroform matched the acute phase signature, but did not decrease serum albumin at the tested doses, suggesting gene expression-based signatures may identify liver-specific inflammatory hepatotoxicity signals that do not elicit an APR that is detectable by a decrease in albumin. The signature can also indicate the intensity of an APR through the magnitude of the signature score.

The biomarker described herein is a 22-gene set of which the 3 most important in terms of impact are lipocalin 2, complement component 9 and guanylate cyclase 2C. Lipocalin 2 (also called NGAL, human neutrophil lipocalin, 24p3, uterocalin, or neu-related lipocalin) functions as a bacteriostatic defense system by binding enterochelin, an iron-binding siderophore of E. coli and Salmonella, starving bacteria of iron (Goetz et al., 2000, 2002; Yang et al., 2002, Yang et al., 2003). Lipocalin2 is also secreted by epithelial cells, where it acts as an intracellular iron delivery system and plays a part in development and carcinogenesis. Complement component 9 is the most abundant member of the membrane attack complex (MAC) that binds to and lyses invading bacterial cells (Esser, 1982; Müller-Eberhard, 1984; Shiver et al., 1986). Guanylate cyclase 2C is also called the heat stable enterotoxin receptor or GC-C. It controls chloride secretion in the intestine, and in the liver it is induced under conditions of liver regeneration and acute phase response (Laney et al., 1994). Thus, the three highest impact genes in the signature are not only inflammatory genes but specifically involved in the mammalian response to bacterial infection. It is worth noting that the gene for albumin does not appear in the signature, as the levels of mRNA in the liver for albumin did not vary. This suggests that albumin protein levels are regulated through synthesis, degradation, and through leakage into extravascular spaces rather than transcription (Schreiber et al., 1989).

The data presented herein demonstrate a model of toxicity in which a hepatic gene expression biomarker was identified. The 22-gene biomarker was independently tested and confirmed to be a robust, accurate, and early sentinel marker of APR. As toxicogenomics continues to be employed in mechanistic toxicology, the described approach to create and test a gene expression biomarker is likely to be a useful, representative strategy.