Article pubs.acs.org/jpr
Pharmacometabonomic Investigation of Dynamic Metabolic Phenotypes Associated with Variability in Response to Galactosamine Hepatotoxicity Muireann Coen,*,† Françoise Goldfain-Blanc,‡ Gael̈ le Rolland-Valognes,§ Bernard Walther,∥ Donald G. Robertson,⊥ Elaine Holmes,† John C. Lindon,† and Jeremy K. Nicholson*,† †
Biomolecular Medicine, Department of Surgery and Cancer, Faculty of Medicine, Imperial College London, United Kingdom Toxicology, Biologie Servier, 905 Route de Saran, 45520 Gidy, France § Molecular Pharmacology and Pathophysiology, Institut de Recherches Servier, 125 Chemin de Ronde, 78290 Croissy Sur Seine, France ∥ DMPK, Technologie Servier, 2527 Rue Eugène Vignat, Orleans 45000, France ⊥ Research and Development, Bristol-Myers Squibb, Princeton, New Jersey, United States ‡
S Supporting Information *
ABSTRACT: Galactosamine (galN) is widely used as an in vivo model of acute liver injury. We have applied an integrative approach, combining histopathology, clinical chemistry, cytokine analysis, and nuclear magnetic resonance (NMR) spectroscopic metabolic profiling of biofluids and tissues, to study variability in response to galactosamine following successive dosing. On re-challenge with galN, primary nonresponders displayed galN-induced hepatotoxicity (induced response), whereas primary responders exhibited a less marked response (adaptive response). A systems-level metabonomic approach enabled simultaneous characterization of the xenobiotic and endogenous metabolic perturbations associated with the different response phenotypes. Elevated serum cytokines were identified and correlated with hepatic metabolic profiles to further investigate the inflammatory response to galN. The presence of urinary N-acetylglucosamine (glcNAc) correlated with toxicological outcome and reflected the dynamic shift from a resistant to a sensitive phenotype (induced response). In addition, the urinary level of glcNAc and hepatic level of UDP-N-acetylhexosamines reflected an adaptive response to galN. The unique observation of galN-pyrazines and altered gut microbial metabolites in fecal profiles of non-responders suggested that gut microfloral metabolism was associated with toxic outcome. Pharmacometabonomic modeling of predose urinary and fecal NMR spectroscopic profiles revealed a diverse panel of metabolites that classified the dynamic shift between a resistant and sensitive phenotype. This integrative pharmacometabonomic approach has been demonstrated for a model toxin; however, it is equally applicable to xenobiotic interventions that are associated with wide variation in efficacy or toxicity and, in particular, for prediction of susceptibility to toxicity. KEYWORDS: galactosamine, metabonomics, NMR spectroscopy, pharmacometabonomics, inter-individual variability
■
INTRODUCTION Drug-induced toxicity represents a serious clinical challenge that is associated with high morbidity and mortality rates.1 The combination of preclinical toxicity (animals) and clinical adverse drug reactions (humans) accounts for up to a third of the cases of drug attrition.2 A meta-analysis of a cohort of U.S. hospital data reported the incidence of serious and fatal adverse drug reactions in hospitalized patients as 6.7% and 0.32%, respectively.3 Fatal adverse drug reactions were found to be the sixth leading cause of death in this cohort of hospitalized patients, highlighting the importance of this clinical challenge.3 It is clear that novel methods for understanding the mechanistic bases for preclinical and clinical toxicological outcomes are still urgently required. An enhanced understanding of the © 2012 American Chemical Society
mechanisms underlying idiosyncratic toxic response, together with a means of prediction of susceptibility to drug-induced toxicity, would significantly drive the development of personalized healthcare. The aminosugar galactosamine (galN) is not found in isolation in mammalian systems and has widely been used as a model of acute liver injury. The key biochemical understanding of galN-induced hepatotoxicity relates to perturbation of the hepatic uridylate pool as a consequence of UDP-glucose depletion.4−6 In addition to the primary metabolic insult, an important role for inflammatory processes in the pathogenesis Received: November 23, 2011 Published: March 2, 2012 2427
dx.doi.org/10.1021/pr201161f | J. Proteome Res. 2012, 11, 2427−2440
Journal of Proteome Research
Article
We also present analysis of circulating serum cytokines and correlation of multicompartmental metabolic profile data with cytokine profiles to further understand the relationship between the primary galN-induced metabolic insult and secondary inflammatory response. This metabonomic and cytokine study was anchored with traditional histopathological and clinical chemistry assessment, which together facilitated progress towards a complete understanding of differential response phenotypes. This integrative approach has provided much mechanistic insight into variability in response for a model toxin, however, it is equally applicable to xenobiotic interventions that are associated with wide variation in efficacy or toxicity and, in particular, for prediction of susceptibility to toxicity.
of galN-induced injury has become clear. Inflammation is believed to result from increased gut permeability and bacterial translocation to portal blood, arterial blood, and intraabdominal organs, which results in endotoxemia.7−9 Indeed, the extent of liver injury is greatly attenuated following colectomy7,9 or exacerbated on co-treatment with both galN and endotoxin.10 Mice are relatively resistant to galN, which is believed to be due to a higher capacity for de novo uridylate synthesis. However, co-administration of galN and endotoxin (lipopolysaccharide) to mice represents a model of synergistic toxicity as sensitivity to galN is increased by several thousand fold.11 Metabonomic approaches, utilizing advanced spectroscopic platforms coupled with multivariate statistical modeling, have been successfully applied to study the systems-level metabolic response to a diverse range of disease and toxin-induced alterations in both preclinical and clinical studies.12−16 The development of pharmacometabonomics has enabled prediction of variability in drug metabolism and toxic outcome from baseline urinary metabolic profiles.17,18 This predictive approach has been applied to the study of the metabolism and toxicity of acetaminophen in humans and rats19−21 and streptozotocin-induced toxicity in a rat model of diabetes.22 NMR-based pharmacometabonomics has also been successfully applied in the clinic to predict toxicological outcomes to capecitabine in patients with inoperable colorectal cancer23 and to predict response to breast cancer treatment.24 The value of NMR-based metabolic profiling in preclinical toxicological assessment has been assessed in the COMET consortium project, which carried out a large number of toxicological metabonomic studies (approximately 150 wideranging toxins and treatments25−27). The present work on variable response to galN-induced hepatotoxicity has been carried out as part of the second phase of the COMET consortium project (COMET-2). The COMET-2 project applied both NMR spectroscopic and mass spectrometric metabonomic platforms to elucidate the mechanistic bases for toxicity induced by selected model hepatic and renal toxins.28−33 We have previously identified extreme interindividual variability in the rat in response to galN from numerous studies that incorporated metabonomic, clinical chemistry, and histopathological assessment.34−37 More recently we have determined distinct metabolic phenotypes based on NMR spectroscopic profiles of liver, feces, urine, and serum, which provided mechanistic insight into susceptible (responders) or resistant (non-responders) phenotypes to galN-induced toxicity.28,30 In the present study we have applied an integrative,38−41 multicompartmental, metabonomic, and pharmacometabonomic approach to further our understanding of galN-induced variable response phenotypes through a study of successive dosing of galN. Importantly, as will be shown herein, we have investigated whether a second dose of galN, after an appropriate time period, has an effect on the primary responder/non-responder phenotype. This systems-level study of dynamic metabolic phenotypes (metabotypes) has involved the acquisition of NMR spectroscopic profiles of urine, feces (pre- and post-treatment), liver, and sera (post-treatment) from rats treated with successive doses of galN. Multivariate statistical modeling of pre-treatment metabolic profiles enabled characterization of predose urinary and fecal metabolites that discriminated with respect to post-treatment response to galN.
■
EXPERIMENTAL PROCEDURES
Animals
Male Sprague−Dawley (SPF Crl:CD(SD)IGS BR) rats (8 wks old, 275−325 g) were obtained from Charles River Laboratories, France (Domaine des Oncins, 69210 SaintGermain sur l’Arbresle, France) and acclimatized to the facility for 7 days. Animals were housed in temperature (20−24 °C) and humidity (40 − 70% RH) controlled rooms with a 12-h light/dark cycle throughout the study and handled and maintained in accordance with the requirements of the EEC guideline (1986). During sample collection periods, rats were housed in metabolism cages. When samples were not collected, animals were housed in individual stainless steel wire hanging cages. Animals were provided with food (sterilized A04C-10 feed pellets purchased from SAFE, Villemoisson-sur-Orge, France) and filtered drinking water ad libitum throughout the study. GalN Administration
On Day −1, animals were selected and assigned to 2 groups (control and galN-treatment), using a stratified randomization procedure based on bodyweight. Galactosamine hydrochloride was purchased from Sigma-Aldrich (St. Louis, MO), dissolved in 0.9% saline to give a free base concentration of 41.5 mg/mL, and filter sterilized using a 0.2 μm filter. On Day 1, all rats were divided into two groups and given a single intraperitoneal (ip) injection of 0 (n = 8) or 415 (n = 30) mg/kg galactosamine in a dose volume of 10 mL/kg (freshly prepared dosing solution). This dose level was selected to complement our previous in vivo metabonomic studies of galN hepatotoxicity28−30 and was expected to induce serum clinical chemistry and histopathological responses indicative of acute but reversible liver injury. After the first administration (dose 1), the animals were separated according to their clinical pathology results into subgroups (responders and non-responders) intended for a second galactosamine administration (dose 2), which was given after an 11-day wash-out period. All rats were euthanized, by exsanguinations under isoflurane anesthesia, 24 h after dose 2. Body weights of all rats were obtained once during the acclimation phase and on each day of galN administration. Sample Collection
All animals were placed in metabolism cages 24 h before each administration, and urine was collected over the following time periods: −24 to 0; 0 to 24 h after each dosing, and 24 to 48 and 48 to 72 h after the first dosing. Animals were not fasted during these collection periods. Urine samples were collected into chilled tubes containing 1 mL of 1% sodium azide. All samples 2428
dx.doi.org/10.1021/pr201161f | J. Proteome Res. 2012, 11, 2427−2440
Journal of Proteome Research
Article
(1H, δ 0). 1H NMR spectra were acquired on a Bruker Avance600 spectrometer, operating at 600.13 MHz 1H frequency and a temperature of 300 K, using a Bruker flow injection probe (Bruker Biospin, Rheinstetten, Germany) with an active volume of 120 μL and an automated sample-handling unit (BEST, Bruker). Samples were transferred from the 96-well plate (with a cooling rack at 277 K) to the NMR flow probe using a Gilson 215 automatic sample handling system (Gilson, Middleton, Wisconsin, USA). For each sample, 500 μL of urine was injected at a rate of 3 mL/min from the well into the transfer line which was maintained at a temperature of 303 K. Urine samples were separated from subsequent samples by approximately 500 μL of push solvent (1% sodium azide in H2O), and each solution was separated by an air bubble. Blank samples were run after every 10th sample to confirm that there was no cross contamination between samples. NMR spectra were acquired using the standard one-dimensional solvent suppression pulse sequence (relaxation delay - 90° pulse - 4 μs delay-90° pulse-mixing time-90° pulse-acquire FID42). For each sample, 128 transients were collected into 64 K data points using a spectral width of 12,000 Hz with a relaxation delay of 4 s, an acquisition time of 2.72 s, and a mixing time of 100 ms. The water resonance was selectively irradiated during the relaxation delay and the mixing time. A line-broadening function of 0.3 Hz was applied to all spectra prior to Fourier transformation (FT).
were centrifuged (3000 rpm, 10 min) to remove particulate matter, divided into aliquots, and stored at −70 °C. Blood was collected from all animals 24 h postdose (dose 1 and 2) into tubes containing a serum separator, for clinical chemistry analysis and metabonomic analysis. Serum was collected by rapid centrifugation after clotting and kept at 4 °C for clinical chemistry measurements, while that used for NMR spectroscopic analysis was stored at −70 °C. Fecal pellets were collected from each rat for 24 h predose and from 0 to 24 h postdose (dose 1 and 2) and stored at −70 °C until analyzed by NMR. Necropsy was performed immediately after euthanasia (24 h postdose 2) with a section of the left lateral lobe of the liver being immediately frozen in liquid nitrogen and stored at −70 °C until analyzed by NMR spectroscopy. Clinical Chemistry
Serum was analyzed for albumin, total protein, glucose, creatinine, triglycerides, alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), glutamate dehydrogenase (GLDH), bile acids, and total bilirubin levels using a Hitachi 917 analyzer (Roche Diagnostics). Serum Cytokine Profiling
Cytokines were profiled using the Mesoscale Discovery (MSD, Gaithersburg, MD20877) multiplex sandwich immunoassay kit for rat serum that simultaneously measures 7 cytokines (Interferon gamma, IFN-γ; Interleukin-1 beta, IL-1β; Interleukin-4, IL-4; Interleukin-5, IL-5; Interleukin-13, IL-13; Chemokine (C-X-C motif) ligand 1/Rat cytokine-induced neutrophil chemoattractant (rodent IL-8 like), KC/GRO; Tumor necrosis factor-alpha, TNF-α; 7-plex kit; MSD K15014C). This panel of cytokines was measured in duplicate aliquots of serum (20 μL) and was calibrated using a 12-point calibration curve from authentic standards of the cytokines. The data were analyzed using the MSD Workbench software that enabled cytokine concentration (pg/mL, ± SEM) for each duplicate pair to be determined. The cytokines IL-5 and KC/ GRO were reliably detected in the sera samples; the remainder of the cytokine panel was found to be below the limit of detection or could not be fitted to the calibration curve. Statistically significant differences were determined between groups by calculation of a p-value from a two-tailed, un-paired Mann−Whitney t test (Prism 5, Graphpad Software).
1
H NMR Spectroscopy of Serum
Serum samples were thawed, vortexed, and allowed to stand for 10 min prior to mixing aliquots (200 μL) with saline containing 20% D2O (400 μL). Samples were spun at 10,000 rpm for 10 min. Samples (500 μL) were placed in NMR tubes (5 mm) and NMR spectra were acquired at a 1H observation frequency of 600.13 MHz and temperature of 300 K. Chemical shifts were referenced to that of α-glucose (H-1, δ 5.23) and D2O provided a field-frequency lock. The Carr−Purcell−Meiboom−Gill (CPMG) spin-echo pulse sequence with a fixed spin-spin relaxation delay, 2nτ of 200 ms (n = 250, τ = 400 μs), was applied to acquire 1H NMR spectra of all sera samples. For each sample, 128 transients were collected into 64 K data points using a spectral width of 12,000 Hz with a relaxation delay of 4 s and an acquisition time of 2.72 s. A line-broadening function of 0.3 Hz was applied to all spectra prior to FT. 1 H NMR Spectroscopy of Water-Soluble Liver Tissue Extracts
Histopathology
All tissues selected for microscopic examination were embedded in paraffin wax, cut at approximately 4 μm in thickness and stained with hemalun eosin saffron (HES). Microscopic examination was performed on liver sections that were assigned a histological score (HS) relative to control livers according to the following severity scale: HS0 = no lesions, HS1 = minimal lesions, HS2 = mild lesions, HS3 = moderate lesions, and HS4 = marked lesions.30
Liver tissue samples (median sample weight of 80 mg) were added to 1.5 mL of cold acetonitrile/water (50:50) and homogenized for 8 min using a ball-bearing tissue lyser (QiagenTissueLyser, Retsch GmBH, Haan Germany). The homogenized samples were spun at 13,000 rpm for 10 min, and the supernatant was removed and lyophilized prior to reconstitution in 600 μL of D2O/H2O (90:10) containing TSP (1.3 mM) and sodium azide (1.4 mM) in 5 mm NMR tubes. NMR spectra were acquired using the standard onedimensional solvent suppression pulse sequence. For each sample, 128 transients were collected into 64 K data points using a spectral width of 12,000 Hz with a relaxation delay of 4 s, an acquisition time of 2.72s, and a mixing time of 100 ms. The water resonance was selectively irradiated during the relaxation delay and the mixing time. An exponential function corresponding to a line-broadening of 0.3 Hz was applied to all spectra prior to FT.
1
H NMR Spectroscopy of Urine
Urine samples were thawed, vortexed, and allowed to stand for 10 min prior to mixing aliquots (400 μL) with phosphate buffer (200 μL, 0.2 M containing 10% deuterium oxide (D2O), 3 mM 3-(trimethylsilyl)-[2,2,3,3-2H4]-propionic acid sodium salt (TSP), and 3 mM sodium azide) and centrifuged at 13,000 rpm for 10 min. Supernatants (550 μL) were transferred into 96-well plates (1 mL, Lablinks, U.K.). The D2O provided a field frequency lock, and TSP provided a chemical shift reference 2429
dx.doi.org/10.1021/pr201161f | J. Proteome Res. 2012, 11, 2427−2440
Journal of Proteome Research
Article
1
H NMR Spectroscopy of Aqueous Fecal Extracts
Fecal samples (median sample weight of 100 mg) were suspended in 3 mL of D2O and homogenized, and the resultant samples were spun at 13,000 rpm for 10 min. The supernatant was removed and placed in 5 mm NMR tubes (500 μL). TSP in D2O (50 μL, 1 mg/mL) was added as a chemical shift reference. NMR spectra were acquired using the same protocol as for the aqueous liver extract analyses. Statistical Analysis of NMR Spectral Data
Full-resolution NMR data were imported into MATLAB (R2006a, Mathworks Inc., 2006). The regions corresponding to water/HDO (δ 4.7−4.9) and TSP (δ −0.2−0.2) were removed from all spectra. In addition, the urea -NH resonance was removed from all spectra (δ 5.6−6). The spectral data (urine, feces, liver) were then normalized using the probabilistic quotient normalization method43 and scaled to unit variance. The serum metabolic profiles were not normalized. Data were modeled using principal component analysis (PCA), partial least-squares (PLS) regression, and pairwise orthogonalprojection on latent structures-discriminant analysis (O-PLSDA). O-PLS-DA is a supervised pattern recognition algorithm that prefilters classification-irrelevant variation from data and improves interpretability of spectral variation between classes.44−46 To prevent overfitting of spectral data, the 7-fold cross validation method was used, from which the crossvalidation parameter Q2 was calculated.45,46 In addition, permutation tests were carried out to test the validity of models, where the Y vector was permuted randomly 1000 times and the p-value for Q2Y calculated.
■
Figure 1. Mean ALT following a first galN administration (dose 1) for controls, responders (R1), and non-responders (NR1) and a second galN administration (dose 2) for controls, induced responders (iR2), and responders (R2). Error bars relate to the SEM for each cohort. *p
