Metabolomic Profiling of In Vivo Plasma Responses to Dioxin

Apr 15, 2013 - Institute for Global Food Security, School of Biological Sciences, Queen's .... Comparison of common components analysis with principal...
1 downloads 0 Views 1MB Size
Article pubs.acs.org/est

Metabolomic Profiling of In Vivo Plasma Responses to DioxinAssociated Dietary Contaminant Exposure in Rats: Implications for Identification of Sources of Animal and Human Exposure Anthony A. O’Kane,* Olivier P. Chevallier, Stewart F. Graham, Christopher T. Elliott, and Mark H. Mooney Institute for Global Food Security, School of Biological Sciences, Queen’s University Belfast, David Keir Building, Stranmillis Road, Belfast, Northern Ireland, U.K., BT9 5AG S Supporting Information *

ABSTRACT: Dioxin contamination of the food chain typically occurs when cocktails of combustion residues or polychlorinated biphenyl (PCB) containing oils become incorporated into animal feed. These highly toxic compounds are bioaccumulative with small amounts posing a major health risk. The ability to identify animal exposure to these compounds prior to their entry into the food chain may be an invaluable tool to safeguard public health. Dioxin-like compounds act by a common mode of action and this suggests that markers or patterns of response may facilitate identification of exposed animals. However, secondary cocontaminating compounds present in typical dioxin sources may affect responses to compounds. This study has investigated for the first time the potential of a metabolomics platform to distinguish between animals exposed to different sources of dioxin contamination through their diet. Sprague−Dawley rats were given feed containing dioxin-like toxins from hospital incinerator soot, a common PCB oil standard and pure 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) (normalized at 0.1 μg/kg TEQ) and acquired plasma was subsequently biochemically profiled using ultra high performance liquid chromatography (UPLC) quadropole time-of-flight-mass spectrometry (QTof-MS). An OPLS-DA model was generated from acquired metabolite fingerprints and validated which allowed classification of plasma from individual animals into the four dietary exposure study groups with a level of accuracy of 97−100%. A set of 24 ions of importance to the prediction model, and which had levels significantly altered between feeding groups, were positively identified as deriving from eight identifiable metabolites including lysophosphatidylcholine (16:0) and tyrosine. This study demonstrates the enormous potential of metabolomic-based profiling to provide a powerful and reliable tool for the detection of dioxin exposure in food-producing animals.



INTRODUCTION Dioxin contamination of food and the environment remains a significant concern for public health. While huge advances have been made in reducing the production and release of dioxins from industrial processes, incinerators, and other sources,1concontamination still occurs with unfortunate regularity with considerable costs to the economies of affected countries through associated product recalls and reputational loss.2−4 Detection of dioxin-related contamination at critical points in the food chain is an important safeguard but presents many technical challenges. The range of compounds with characterized dioxin-like toxicity includes congeners of polychlorinated dibenzo-p-dioxins (PCDDs), polychlorinated dibenzofurans (PCDFs), and polychlorinated biphenyls (PCBs). This diverse variety of potential contaminating agents poses an analytical problem in assessing the risk within any given sample. Current reference dioxin detection methods require extraction of all relevant congeners and their individual quantification by © 2013 American Chemical Society

capillary gas chromatography-high resolution mass spectrometry (GC-HRMS)) instrumentation.5,6 The concentration of each compound is then adjusted in terms of equivalence to 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) by a toxic equivalence factor (TEF) giving an overall toxic equivalence quotient (TEQ).7 Cell-based methods have been devised to screen larger numbers of samples for dioxin-like activity and are based on the ability of dioxins to induce, in a dose-dependent manner, the cytochrome P450 1A1 (CYP1A1) enzyme system in cultured rat liver cells, which is determined using ethoxyresorufin-Odeethylase (EROD) activity or by a dioxin-responsive chemically activated luciferase gene expression (DR-CALUX) assay.8−13 However, such biological-based methods may be Received: Revised: Accepted: Published: 5409

January 2, 2013 April 11, 2013 April 15, 2013 April 15, 2013 dx.doi.org/10.1021/es305345u | Environ. Sci. Technol. 2013, 47, 5409−5418

Environmental Science & Technology

Article

metabolomic approaches may be the most useful approach. Metabolomic-profiling using highly sensitive ultrahigh performance liquid chromatography quadropole (UPLC) Time-offlight (QTof) mass spectrometry (MS) offers the capability to monitor the relative abundance of the full range of small molecular weight (100−1200 Da) compounds within any given biological sample. In the present study a metabolomic-based plasma-profiling technique has been developed and optimized which is capable of identifying dioxin exposure of animals through contaminated feed sources. To replicate typical dioxin exposure scenarios and to account for interactions with co-contaminating toxins, male Sprague−Dawley rats were fed materials containing specific amounts of dioxin from a number of sourcesnamely incinerator soot, a PCB oil standard (Aroclor 1254), and pure dioxin (TCDD). Extracts of plasma from exposed animals were analyzed using UPLC-QTof-MS techniques and the acquired metabolomic profiles used to build multivariate models of exposure which were tested for their ability to correctly predict source of exposure based upon their metabolomic profiles. Where possible, metabolite compounds of importance to the discrimination of the feeding groups were identified on the basis of the spectral information obtained.

subject to interference from environmental pollutants such as heavy metals, pesticides and other organic chemicals as well as endogenous aryl hydrocarbon receptor (AhR) agonists and antagonists.14 Human exposure to dioxins occurs primarily through the ingestion of animal-derived food products17 with dioxins typically entering the food chain through contaminated animal feed sources. Dioxin compounds are typically stable, persistent, lipophilic chemicals, which accumulate readily in the adipose tissues of animals, and contaminated feed materials can easily lead to excess levels in a wide range of primary and processed foodstuffs.3,15,16 Animal feed contamination may occur as a consequence of incorporation of residues of combustion during drying of feed ingredients,18 or from the contamination of feed oil components by PCB-containing sources.19 Therefore, the main routes of dioxin exposure originating from contaminated animal feed also involve two distinct sets of co-contaminating toxic agents: polynuclear aromatic hydrocarbons (PAHs) and PCBs. Combustion will generate dioxin-like PCDDs and PCDFs but will also produce a relatively large yield of PAHs.20,21 PAHs are a group of important lipophilic toxins which also tend to bioaccumulate22 and are known to be potent carcinogens.23 The TEQ contribution in PCB oils contaminated with dioxins is principally from non-ortho coplanar PCB congeners rather than dioxins and furans, but a large amount of mono- and diortho-substituted noncoplanar PCBs are also present.3,24 Ortho-substituted PCBs also have distinctive neurotoxic and immunotoxic effects.25−29 The potential for PAHs and PCBs to influence downstream responses to dioxin exposure in vivo and in vitro is high, as is the likelihood of one or more of these compounds being present alongside dioxin in the majority of exposure scenarios related to contaminated feed materials. Detecting dioxin contamination within live farm animals would pre-empt the presence of dioxin in meat and other products but can only be performed currently using existing methods on fat tissue taken post slaughter, with results of such analyses often not available until after contaminated materials have entered the human food chain. More appropriate methods to detect dioxin exposure are required which can identify contamination at an early stage of food production. Such tests need to be able to measure identifiable components present in blood, urine or other accessible matrix types which can be obtained from live animals using minimal invasive techniques.30 Attempts have been made to adapt the DR-CALUX assay for the analysis of serum31 and other techniques based on EROD activity in circulating blood lymphocytes have been explored32 but with limited success. Ingested dioxins are quickly partitioned into fatty tissue and levels in blood are thereby incredibly low making analysis based on direct detection difficult.33,34 Dioxins produce a myriad of downstream biological responses in exposed organisms.35 The response and toxic outcome is highly dependent upon the developmental stage of the animal as well as the sex, diet, stress levels, and other chemical and environmental challenges.36−39 Markerbased detection of disease and exposure to chemicals is a growing area of interest to both animal diagnostics and food safety surveillance using techniques such as 2-dimensional gel electrophoresis40 (2D-GE), nuclear magnetic resonance41 (NMR) or liquid chromatography−mass spectrometry42 (LCMS). Proteomic and metabolomic techniques have been applied previously to help elucidate the toxic effects of TCDD35,40,43 and the findings of this work suggests that



EXPERIMENTAL SECTION TCDD and Aroclor 1254 (Lot 124−191) were purchased from Accustandard Inc. (New Haven, CT). Details of analysis of full PCB, PCDD, and PCDF content of this lot have been published previously24 with the TEQ content calculated to be 39.42 μg/g WHO-TEQ based on reported congener profiles. Aroclor 1254 was dispersed directly into corn oil at 1 μg TEQ/ mL and subsequently diluted to 0.02 μg TEQ/mL. TCDD was made up to 1 mg/mL in toluene, 100 μL was added to 1 mL of corn oil and the toluene removed by evaporation and verified by mass for subsequent incorporation at a final concentration of 0.1 μg TEQ/kg to animal feed. A sample of characterized incinerator soot was provided as a gift from Marchela Pandelova of the Helmholtz Centrum, Munich, Germany. Analysis of dioxin content of this material “Soot 2” has been published previously20 and the I-TEQ calculated as 19.71 ng ITEQ/g based on data presented. Additional analysis of incinerator soot for PAH and PCB content was carried out by Scientific Analysis Laboratories, Manchester, UK. Total PAH content was found to be 2.2 mg/kg by gas chromatography−mass spectrometry (GC-MS) with phenanthrene the most abundant congener at 0.5 mg/kg. None of the panel of PCBs measured (7 ortho-substituted and 12 nonorthosubstituted) were detectable by GC-MS with a limit of detection set to 500 pg/g. Incinerator soot was added to corn oil at a level of 0.02 μg TEQ/mL prior to incorporation into feed materials. 1-O-hexyl-2-C-methyl-3-phosphatidylcholine (LysoPC (16:0)) and L-tyrosine were obtained from Sigma Aldrich (Poole, UK). Experimental Feed Preparation. Powdered and gelatinized rat feed was obtained from Special Diet Services (Essex, UK) and supplied in two forms: a control diet containing corn oil at 0.5% (v/w) and a dry form to which corn oil had not been added, with the identical batch of corn oil also supplied for use as a vehicle. To each experimental diet, the exposure compound dispersed in 10 mL of corn oil was added gradually to a 2 kg batch of dry feed using a motorized mixer to a final concentration of 0.1 μg TEQ/kg diet. Diet was prepared for animals by the addition of ∼40 mL hot water (50−70 °C) to 20 5410

dx.doi.org/10.1021/es305345u | Environ. Sci. Technol. 2013, 47, 5409−5418

Environmental Science & Technology

Article

g powdered feed in individual dishes left overnight at 4 °C to form a firm gel. Weights of feed and water were recorded to allow estimation of daily individual animal exposure to dioxin. Animal Feeding Study. Twenty-four male Sprague− Dawley rats (5 weeks old, 120−140g) were received in two batches of 12 from Charles River (Kent, UK). Animals were acclimatized for 5 days in the animal facility before being transferred to metabolic cages. The rats were housed at 27 °C with 50% humidity for comfort and a 12 h light cycle. Feeding studies were carried out in two distinct replicate phases with each following an identical regime and using the same batch of prepared experimental feed. In each phase 12 rats were divided into four groups of three rats and each group was fed a diet containing 0.1 μg/kg TEQ dioxin from TCDD, soot, or Aroclor 1254 along with a matching group of animals which were fed a control diet as described above. Animals were provided with prepared diet daily as their only food source for 14 days. Water was provided ad libitum. Urine output was collected and weighed for each animal per 24 h period and bodyweight was recorded daily. At the end of the feeding period animals were euthanized under CO2 and exsanguinated by cardiac puncture and blood collected in tubes containing K2 EDTA (BD Vacutainer, Plymouth, UK). Blood was processed as plateletdepleted plasma, immediately frozen in aliquots under liquid N2 and stored at −80 °C prior to analysis. UPLC-QTof-MS Metabolomic Profiling of Plasma. One mL of ice-cold acetone was added to 300 μL plasma (Day 14) from each animal, vortexed, and centrifuged at 10 000g for 15 min. The supernatant was collected and mixed with 500 μL chloroform:methanol 1:1 (v/v) and the upper fraction was transferred to a glass tube. This process was repeated, the fractions combined and evaporated to dryness under N2 at 40 °C. The resulting residue was reconstituted in 150 μL water for analysis, forming a clear solution. Pooled samples for each feeding group were also prepared from 20 μl of extract from each animal; these were used for condition of the system and for quality control of data analysis. Replicate injections (n = 6) of a protein depleted plasma sample from each individual animal were randomly analyzed using an Acquity UPLC and Xevo G2 QTof system (Waters, Manchester, UK). LC separation was performed on an Acquity HSS T3 column (1.8 μm, 2.1 × 100 mm) with an injection volume of 7 μL and flow rate of 0.6 mL/min over a 20 min run time. The mobile phase consisted of water with 0.1% formic acid and acetonitrile with 0.1 formic acid under the following gradient conditions: 99% water until 2 min then gradually increasing acetonitrile to 100% at 14.5 min until 17 min, finally re-equilibrating the column with 99% water until 20.0 min. QTof-MS data was acquired in positive electrospray ionization (ESI+) mode using resolution mode, capillary voltage was 2.5 kV, sampling cone was 30 V, and extraction cone was 4.0 V. Source temperature was 120 °C and desolvation gas temperature was 600 °C. Cone gas flow was 50 L/h and desolvation gas flow was 900 L/h. Mass spectra data were acquired in centroid mode using MSE function (low energy: 6 eV; high energy: ramp from 15 to 30 eV) over the range m/z 100−1200 range with a scan time of 0.1 s. Pooled samples were additionally injected after every 10 sample injections throughout the course of metabolomic profiling runs44 to evaluate chromatographic reproducibility of retention times and peak intensities and also in six replicates before sample analysis to condition the system. A leucineenkephalin lock mass calibrant was used throughout at 10 μL/ min infusion.

Data Processing. Initial processing of acquired MS data involved feature identification and integration using Waters MarkerLynx (version 4.1, Waters Corporation, Milford, MA). Chromatographic features between 0 and 17.5 min were processed using a marker intensity threshold of 750 and a peakto-peak baseline noise limit of 2000 (arbitrary units). Features were further refined by manual checking to remove any items which were not present in more than 10% of the spectra. Preprocessed data were then exported to SIMCA 13.0 (Umetrics, Umea, Sweden) and used to construct multivariate statistical models. Prior to in-depth data analysis, data quality was assessed in terms of reproducibility by an approach adopted by other leading metabolomics researchers.45 Clustering of the pooled samples was assessed using principal component analysis (PCA) to reveal if platform stability had been achieved. The data were also exported to Metaboanalyst 2.046 for parallel multivariate and ANOVA analysis. Data were used to generate models for PCA, partial least squares− discriminant analysis (PLS-DA), and orthogonal partial least squares−discriminant analysis (OPLS-DA). These techniques consolidate the most important differences between the spectra produced into the smallest meaningful number of mathematical components revealing relationships and groupings within the data.47 The ability of each model to distinguish and classify animals into their respective exposure groups was assessed rigorously using a range of methods. Tentative identification of the most important exposure markers was performed by crossreferencing multiple metabolite databases. Spectra were examined for patterns denoting common adducts such as the addition of sodium ions or dimerization. Once potential parent ions had been putatively identified in this manner, further patterns of fragmentation were considered with reference to online databases including Massbank.jp and MetLin Confirmation of Marker Ion Identity. Standards were run to confirm the identity of the parent compounds responsible for a number of confirmed ions of interest including LysoPC (16:0) and L-tyrosine. Compounds were dissolved in water (50 μg/mL) and injected onto the same chromatographic system used for the original plasma metabolomic analysis. The chromatograms and spectra generated were subsequently compared to those recorded during the analysis of plasma.



RESULTS AND DISCUSSION Animal Feeding Study. Animals showed no sign of any recognizable symptoms of dioxin toxicity throughout the course of the study and were found to consume feed materials provided on a daily basis almost entirely with very little spillage noted. Exact weights of feed prepared for and consumed by each animal were recorded daily and used to calculate daily intake. Calculated TEQ intake ranged between 6.9 and 10.1 ng TEQ/kg body weight/day over the course of the study. Urine output and water intake was compared daily by ANOVA to assess for any impact of feeding group - no significant differences between the control and exposed groups were observed. Animals (n = 24) gained bodyweight in a linear manner (R2 = 0.971−0.995) from a mean weight of 184 g (Day −2) to 275 g (Day 14) and no differences were observed between control and respective exposed group animals. Weight gain was found to be within the 95% confidence limits for optimal weight gain as outlined by animal supplier guidelines.48 Daily water intake was between 8 and 17 mL, which is lower than expected for rats at this stage of growth. However, as the provided feed was prepared using water this may have 5411

dx.doi.org/10.1021/es305345u | Environ. Sci. Technol. 2013, 47, 5409−5418

Environmental Science & Technology

Article

Figure 1. Multivariate analysis of UPLC-QTof-MS profiling of plasma from rats fed dioxin-like toxins in SIMCA 13.0. A: PCA scores plot, B: PLSDA scores plot, C: OPLS-DA scores plot. Individual item labels represent experimental feeding study phase (P1/2) and animal number (R1−12). Variances for the first two components of each model are shown below each plot. Ellipses represent Hotelling’s T2 95% range. Feeding study groups are represented as follows: CTRL (green circle) = Control group; TCDD (blue square) = Pure TCDD group; SOOT (red upward triangle) = Soot group; AROC (yellow downward triangle) = Aroclor 1254 group.

clear separation between the control group and the spectra obtained from treated animals. The classification rate of this PLS-DA model was 100%, based upon removal and prediction of a third of available data, with a Fisher’s probability score of 1.9 × 10−28. An OPLS-DA model was then constructed, classed according to each of the four feeding groups and the resulting scores plot is shown in Figure 1C (R2X: 0.764, R2Y:0.965, Q2X: 0.942). Four distinct groupings are clearly observed, each within a separate quadrant of the plot demonstrating an exceptional level of separation. The rate of correct classification by this model was 100% based upon removal and prediction of one-third of inputted observations, with a Fisher’s probability of 1.9 × 10−28. These analyses demonstrate that metabolomic profile data from UPLC-QTof-MS analysis will produce clearly defined groupings and a highly valid model in PLS-DA and OPLS-DA analysis. As an important indicator of data quality, the spectra obtained from pooled samples for each group were found to be grouped near to the center of the clusters associated with each feeding groupthis was observed when the pooled data was inputted into all three model types.45 The same data was also imported into Metaboanalyst 2.0,46 and similar findings in terms of PCA and PLS-DA analysis were obtained (data not shown). In addition, an ANOVA using Tukey’s post hoc multiple comparison test was performed on this data set which identified a total of 761 variables out of the 1555 inputted, which showed significant variation (p =