Environ. Sci. Technol. 2008, 42, 4188–4194
Investigating Compensation and Recovery of Fathead Minnow (Pimephales promelas) Exposed to 17r-Ethynylestradiol with Metabolite Profiling D . R . E K M A N , * ,† Q . T E N G , † D. L. VILLENEUVE,‡ M. D. KAHL,‡ K. M. JENSEN,‡ E. J. DURHAN,‡ G. T. ANKLEY,‡ AND T. W. COLLETTE† Ecosystems Research Division, U.S. EPA, 960 College Station Road, Athens, Georgia 30605, and Mid-Continent Ecology Division, U.S. EPA, 6201 Congdon Boulevard, Duluth, Minnesota, 55804
Received January 8, 2008. Revised manuscript received March 6, 2008. Accepted March 10, 2008.
1H
NMR spectroscopy was used to profile metabolite changes in the livers of fathead minnows (Pimephales promelas) exposed to the synthetic estrogen 17R-ethynylestradiol (EE2) via a continuous flow water exposure. Fish were exposed to either 10 or 100 ng EE2/L for 8 days, followed by an 8 day depuration phase. Livers were collected after days 1, 4, and 8 of the exposure, and at the end of the depuration phase. Analysis of polar extracts of the liver revealed a greater impact of EE2 on males than females, with metabolite profiles of the former assuming similarities with those of the females (i.e., feminization) early in the exposure. Biochemical effects observed in the males included changes in metabolites relating to energetics (e.g., glycogen, glucose, and lactate) and liver toxicity (creatine and bile acids). In addition, amino acids associated with vitellogenin (VTG) synthesis increased in livers of EE2-exposed males, a finding consistent with increased plasma concentrations ofthelipoproteininthefish.Usingpartialleast-squaresdiscriminant analysis (PLS-DA), the response trajectories of the males at both exposure concentrations were compared. This revealed an apparent ability of the fish to compensate for the presence of EE2 later in the exposure, and to partially recover from its effects after the chemical was removed.
Introduction The release of endocrine disrupting chemicals (EDCs) by sewage treatment plants and various nonpoint sources into aquatic ecosystems can adversely impact fish and other organisms (1, 2). Although the toxicities of these chemicals are varied, adverse reproductive and developmental effects are often observed (1). To more effectively evaluate exposures of ecologically relevant species to EDCs for regulatory purposes, methods for determining the specific mechanism of action (MOA) for biologically active chemicals in complex mixtures are being sought. It is believed * Corresponding author e-mail:
[email protected]; phone: 706-355-8250; fax: 706-355-8202. † Ecosystems Research Division. ‡ Mid-Continent Ecology Division. 4188 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 11, 2008
that the “omics” technologies are potentially well poised to address this need (3). We are interested specifically in assessing effects of EDCs with well-defined MOAs within the hypothalamic-pituitarygonadal (HPG) axis using small fish models. In the long term, we seek to develop “omic”-based biomarkers and other related tools and information that can be used for decisionmaking in ecological risk assessments. This goal requires studying responses of fish as a function of many different exposure parameters. For example, in addition to understanding responses related to chemical concentration, it is critically important with EDCs to determine differential effects related to gender. Since a large number of biological processes are under some degree of hormonal control, the effects of EDC exposures are often quite different for males and females. Furthermore, an assessment of the evolution of responses over the time course of an exposure, and also after removal of the chemical, is critical from a regulatory perspective. In addition to determining adverse effects associated with toxicity, this approach allows one to examine the ability of the organism to compensate for, or even recover from, the exposure (4). This type of temporal information is essential to accurately assess ecological risk (5). Metabolomics involves measuring changes in levels of endogenous metabolites in tissues and biofluids to determine the biochemical changes that occur as a result of exposure(s). Although the body of work in this area with regard to environmental applications is still relatively small, there has been steady growth in recent years, with a number of reports published examining the impact of pollutants or environmental stressors on ecologically relevant organisms (6–11). This growth is likely due to some notable advantages offered by this technique. For example, unlike transcriptomic and proteomic techniques, a sequenced genome for the organism under study is not a requirement for metabolomics. This is a particular advantage for many ecologically important species because their genomes have yet to be fully sequenced. In addition, the two most widely used analytical techniques in metabolomicssnuclear magnetic resonance (NMR) spectroscopy and mass spectrometry (MS)scan be rapid and highly automated, making these techniques amenable to high-throughput analyses. Moreover, when considered on a per-sample basis, a metabolomics approach is often significantly less expensive than other “omic” techniques. A final advantage of metabolomics is that metabolite profiles in an organism are more proximal to apical changes (and, hence, possible adverse effects) than are alterations in gene or protein expression (12). Given these advantages, metabolomics is an attractive option for studying the responses of ecologically relevant species upon exposure to EDCs, particularly when multiple variables (gender, concentration, duration, etc.) must be considered. With these considerations in mind, we have applied NMRbased metabolomics with the commonly used small fish toxicity model, Pimephales promelas (fathead minnow), exposed to the contraceptive estrogen 17R-ethynylestradiol (EE2). A previous study by Samuelsson et al. demonstrated the utility of NMR-based metabolomics for determining responses to EE2 using plasma from juvenile rainbow trout collected after two weeks of exposure (13). In this paper we expand on that work by demonstrating the utility of metabolomics for investigating endocrine disruption relative to temporal as well as gender-specific responses, in adult fish exposed to two test concentrations of EE2. In addition, a 10.1021/es8000618
Not subject to U.S. Copyright. Publ. 2008 Am. Chem. Soc.
Published on Web 04/30/2008
depuration phase was included in the design to determine the extent to which the fish could recover once the exposure had ended.
Materials and Methods Experimental Design and Sample Collection. Prior to and during the exposures, fish were fed adult brine shrimp (Artemia sp.; San Francisco Bay Brand, Newark, CA), twice daily to satiation except on sampling days, in which case the fish were only fed in the afternoon after sampling was completed. Exposures were conducted using eight replicate tanks (20 L each) per treatment group (0, 10, 100 ng EE2/L). A stock solution of EE2 (Sigma Chemical Co., St Louis, MO) was dissolved in UV-filtered Lake Superior water without the use of a carrier solvent and delivered after appropriate dilution to the test tanks at a flow-rate of about 45 mL/min. Twenty-four hours after starting chemical delivery, exposures were initiated by randomly assigning four male and four female fathead minnows, obtained from an on-site culture facility at the US EPA, Duluth, MN, to each tank. The time of fish addition and sampling was staggered by replicate so that all samples were collected within 1 h of the intended exposure duration. At each time point (i.e., 1, 4, 8 d, and 8 d postexposure), one male and one female fish were sampled from each replicate tank. Fish were anesthetized in a buffered solution of tricaine methanesulfonate (MS-222; 100 mg/L buffered with 200 mg/L NaHCO3; Finquel, Argent, Redmond, WA). Blood was collected from the caudal artery/vein with a microhematocrit tubule, and plasma was isolated by centrifugation for 3 min at 15 000 rcf. Fish were weighed, and livers, gonads, and brain with pituitary were removed, transferred to preweighed microcentrifuge tubes, and flashfrozen in liquid nitrogen. All samples were stored at -80 °C until extracted and/or analyzed. Average (+/- SD) fish masses were as follows: males 3.02 +/- 0.62 g; females 1.30 +/- 0.27 g. Dorsal fatpads were excised from males and weighed, but not stored for further analysis. Vitellogenin (VTG) concentrations were determined in plasma using an enzyme linked immunosorbent assay (ELISA) with a polyclonal antibody to fathead minnow VTG (14, 15). Purified fathead minnow VTG protein was used as a standard. Statistical tests used to assess the fatpad and VTG data are described in the Supporting Information. EE2 levels were verified in the tanks during the exposure and were found to be 70-80% of the nominal values (see Supporting Information). Extraction of Liver Metabolites and NMR Analysis. Liver samples were extracted using a dual-phase extraction procedure adapted from Viant (16). This procedure generates both a polar metabolite phase and a lipophilic metabolite phase. Further information regarding the preparation of the extracts prior to NMR analysis is reported in the Supporting Information. For each extract a one-dimensional (1D) 1H NMR spectrum of the polar phase was obtained. All of these spectra were acquired at 25 °C on a Varian Inova 800 MHz spectrometer (799.7 MHz, 1H) using a 5 mm cryogenic triple resonance probe. Additional information on the acquisition parameters used for these spectra as well as the twodimensional (2D) NMR experiments used to confirm metabolite identities can be found in the Supporting Information. An example of a typical 1D 1H NMR spectrum used for the metabolomic analyses with several metabolites labeled is also shown in the Supporting Information (Figure S1). NMR Data Analysis. After phase and baseline correction, spectra were referenced to TSP. The residual water resonance was then removed and all spectra were normalized to the same total intensity. Next, spectra (0.50-10.00 ppm) were segmented into 0.01 ppm wide bins using ACD/1D NMR Manager (Advanced Chemistry Development, Toronto,
Canada) and imported into Microsoft Excel (Microsoft Corporation, Redmond, WA). The Excel spreadsheet of binned spectra was imported into SIMCA-P+ (Umetrics Inc., Umea, Sweden) for multivariate data analysis. (All analyses in SIMCA-P+ were conducted on bins that were mean-centered and Paretoscaled.) Initially, principal components analysis (PCA) was used with the entire data set to identify outliers. Eight fish (about 4%) were observed to fall outside of the Hotelling’s T2 ellipse at the 95% confidence interval in a scores plot of the first two components and were removed from the data set prior to subsequent analysis. Following PCA, exposure-response trajectories were generated for male fish using partial least-squares discriminant analysis (PLS-DA) within SIMCA-P+. One PLS-DA model was built using all male samples, which were assigned among nine separate classes. All male controls (i.e., those sampled at day 1, 4, 8, and 16) were assigned to a single class (N ) 30) because changes in controls over time were small relative to changes due to the exposures. The eight additional classes (N ) 7 or 8) were defined by both treatment level (10 or 100 ng/L) and day sampled (1, 4, 8, or 16). Note that the region of 3.26-3.28 ppm was excluded from this analysis. This region, which contains coincident and abundant peaks from betaine and TMAO, was highly variable, exhibited very strong leverage on the model, and was uninformative. A twocomponent model was found optimal using cross validation. Subsequently, this model was further validated using the permutation testing routine within SIMCA-P+ (17) (For example, 200 randomly permuted models all exhibited considerably less significance than the “correct” model.) Also, several smaller PLS-DA models (constructed in the same fashion, but using a subset of the male samples) were validated externally by predicting class membership for samples that were excluded during model construction. Univariate analysis was conducted within Excel on the spreadsheet of spectra that was generated by ACD/1D NMR Manager. (Recall that these spectra were normalized and binned, but not scaled). Initially, an “average class spectrum” was calculated by averaging the binned spectra across all class members (N ) 6, 7, or 8). Note that class was defined by sex, exposure level (including controls), and day sampled. Next, a difference spectrum was generated by subtracting the averaged bins of the relevant control class from those of each exposed class. Then, a t test (one-tailed, assuming equal variances) was conducted on each bin to see if the average for the exposed class differed significantly from that of the relevant control class, using a p value
