Environ. Sci. Technol. 2007, 41, 3375-3381
Direct Sampling of Organisms from the Field and Knowledge of their Phenotype: Key Recommendations for Environmental Metabolomics ADAM HINES,† G B O L A H A N S A M U E L O L A D I R A N , †,‡ JOHN P. BIGNELL,§ GRANT D. STENTIFORD,§ AND M A R K R . V I A N T * ,† School of Biosciences, The University of Birmingham, Birmingham, B15 2TT, United Kingdom, and Centre for Environment, Fisheries, and Aquaculture Science (Cefas), Weymouth Laboratory, Barrack Road, Weymouth, Dorset DT4 8UB, United Kingdom
Critical questions must be addressed to evaluate the potential of metabolomics for studying free-living wildlife. First, can metabolomics identify stress-induced phenotypes in animals experiencing a highly variable environment or must animals be stabilized in a controlled laboratory prior to sampling? Second, is knowledge of species and phenotype (gender and age) required to interpret metabolomics data? To address these questions, we characterized the metabolic variability of the mussel and determined if inherent variability masked the metabolic response to an environmental stressor, hypoxia. Specifically, we compared metabolic fingerprints of adductor muscle and mantle from four groups of Mytilus galloprovincialis: animals sampled directly from the field with and without hypoxia and those stabilized in a laboratory for 60 h, also with and without hypoxia. Contrary to expectation, laboratory stabilization increased metabolic variability in adductor muscle, thereby completely masking the response to hypoxia. The principal source of metabolic variability in mantle was shown to be gender-based, highlighting the importance of phenotypic anchoring of samples to known life history traits. We conclude that direct field sampling is recommended for environmental metabolomics since it minimizes metabolic variability and enables stress-induced phenotypic changes to be observed. Furthermore, we recommend that species and phenotype of the study organism must be known for meaningful interpretation of metabolomics data.
Introduction Marine mussels are widely studied and frequently used in environmental monitoring programs. They are ubiquitous, sedentary filter-feeders and inhabit coastal and estuarine habitats (1, 2). Advances in genomic technologies have expanded the study of this organism beyond the confines of * Corresponding author phone: +44-(0)121-414-2219; fax: +44(0)121-414-5925; e-mail:
[email protected]. † The University of Birmingham. ‡ Current address: School of Life and Health Sciences, Aston University, Birmingham, B4 7ET, U.K. § Weymouth Laboratory. 10.1021/es062745w CCC: $37.00 Published on Web 04/03/2007
2007 American Chemical Society
single biomarker approaches (3). Recent studies in Mytilus spp. have correlated changes in gene transcription with responses to mixtures of crude oil, mercury (4), copper (5), and benzo(a)pyrene (6). The response of mussels to oil has also been evaluated using proteomics (7). Nuclear magnetic resonance (NMR) spectroscopy-based metabolomics is another post-genomic approach that combines the highthroughput metabolic fingerprinting capabilities of 1H NMR with statistical bioinformatics (8), enabling assessment of an animal’s phenotypic response to environmental stressors (9, 10). Comparatively few environmental metabolomic studies have been reported (reviewed in ref 11). Previous studies of the metabolomic stress responses in aquatic invertebrates have been limited to characterizing withering syndrome in red abalone (10, 12). Despite the considerable potential of this technique to assess the metabolic health of free-living animals, it is still unproven in this role. For free-living animals that experience large variations in their metabolomes due to multiple environmental factors, two critical questions must be addressed before undertaking metabolomics studies. First: what is the optimal sampling strategy? Here, we need to consider whether animals should be sampled directly from the environment or whether they require a laboratory stabilization period under standardized conditions prior to measurement. Direct field sampling would preclude stress from husbandry and would capture the true metabolome of the wild animal, but environmentally induced metabolic variability may be considerable. Conversely, laboratory stabilization may reduce metabolic variation due to differences in food intake, temperature, and location between individuals but potentially introduces variation due to transportation and husbandry. Furthermore, laboratory stabilization may allow recovery from anthropogenic stressors encountered in the field, thus degrading the very information being sought. To address this question, we used 1H NMR metabolomics to (i) compare the overall metabolic variability between mussels sampled directly from the field and those returned to and then stabilized in the laboratory (with the hypothesis that laboratory stabilized animals exhibit less variability) and (ii) determine the visibility of a natural metabolic stress, hypoxia, against background metabolic variation in mussels sampled directly from the field and those stabilized in the laboratory (with the hypothesis that laboratory stabilized animals will exhibit a more pronounced metabolic change due to hypoxia since their underlying metabolic variability will be less). Hypoxia was selected as a stressor since it is easy to induce and has a wellcharacterized biochemistry (13, 14). The second critical question is how important is it to know the species and phenotype (gender, age, etc.) of experimental animals for interpretation of metabolomics data? Previous studies on fish have recommended that due to inherent variability present in field collected samples, phenotypic anchoring to life history data related to age, sex, species, and disease status is critical for interpretation of omics-based biomarker data (15, 16). This principle is also pertinent to marine mussels that are known to vary considerably in terms of sexual development, storage of tissue reserves, and disease status by season, sex, species, and location (17). Further examples of phenotypic effects in mussels include the influence of age on lipid peroxidation and oxidative stress (18, 19) and variation in stress resistance and metabolism with size and age (20, 21). To address this question, we characterized (or controlled for) several genetic and phenotypic traits in mussels from the test site at Port Quin, U.K. Since the metabolic variation that was observed in mantle VOL. 41, NO. 9, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
3375
tissue was suspected to arise from gender, additional mussels of known gender (based upon histology) were collected from Southampton Water, U.K. NMR metabolomics was then used to provide the first unbiased assessment of the effect of gender on mantle metabolome. The novel information gained was used to interpret the metabolic fingerprints of all mussels collected from the Port Quin site. Finally, we report recommended sampling strategies for metabolomics studies of freeliving wildlife.
Materials and Methods Mussel Sampling and Induction of Hypoxia. Mussels (Mytilus galloprovincialis, 4.3-4.8 cm long) were sampled from the mid-tide level at Port Quin, Cornwall, U.K. (latitude 50° 35′16′′N, longitude 04° 52′19′′W) in July 2004, shortly after emersion. Fourteen animals were resubmerged in the sea within a mesh bag for 2 h, and then adductor muscle, mantle, and gill were rapidly dissected and flash-frozen in liquid nitrogen. A second group of 14 mussels remained exposed to air for 2 h (hypoxic stress), and then tissues were harvested. A further group of 28 mussels was transported to Birmingham in moist packing at 4 °C using standard protocols (22). On arrival, ca. 10 h after collection from the field, they were placed in recirculating artificial seawater (Tropic Marin, Wartenburg, Germany; 15 °C, 34 psi) for a 60 h stabilization period. Next, 14 animals were rapidly dissected. The remaining group of 14 mussels was removed from the seawater for 2 h (hypoxic stress), and the tissues were rapidly harvested. In a separate study, mantle was harvested from 24 M. galloprovincialis samples from Southampton Water, U.K. between April and July 2005. Each mantle was divided into two for metabolomics and for a histological determination of mussel gender. Gender was determined by the presence of ovary or sperm follicles within mantle matrix (23). Metabolite Extraction. Polar metabolites were extracted from adductor muscle and mantle from Port Quin mussels using a methanol/chloroform method (24). Briefly, tissue was ground under liquid nitrogen using a mortar and pestle and then extracted in 4 mL/g of methanol, 0.85 mL/g of water, and 2 mL/g of chloroform. The mixture was shaken and centrifuged (5 min, 1500g, 4 °C), and the supernatant was removed. A total of 2 mL/g of chloroform and 2 mL/g of water were added, and the mixture was vortexed and then centrifuged (10 min, 1500g, 4 °C). The methanol layer was removed, dried in a centrifugal concentrator (Thermo Savant, Holbrook, NY), and stored at -80 °C. There was insufficient mantle material for two of the animals, so these were not extracted. Additionally, during extraction, two mantle samples showed poor phase separation and were discarded. Immediately prior to NMR analysis, the dried polar extracts were resuspended in sodium phosphate buffer (0.1 M in D2O, pH 7.4, containing 0.5 mM sodium 3-trimethylsilyl-2,2,3,3d4-propionate (TMSP) chemical shift standard). Mantles from Southampton mussels were homogenized in 4 mL/g of methanol and 0.85 mL/g of water using a Precellys-24 beadbased homogenizer (Stretton Scientific Ltd., U.K.). Next, 4 mL/g of chloroform and 4.4 mL/g of water were added, the mixture was vortexed and centrifuged, and the upper layer was removed and dried to isolate the polar extracts. Immediately prior to NMR analysis, the extracts were resuspended in sodium phosphate buffer (0.1 M in 10% D2O and 90% H2O, pH 7.0, containing 0.5 mM TMSP). 1H NMR Spectroscopy. Extracts of adductor muscle and mantle from Port Quin mussels were analyzed on a DRX-500 NMR spectrometer (Bruker Biospin, Coventry, U.K.) equipped with a cryoprobe and operated at 500.18 MHz (at 297 K). One-dimensional (1-D) 1H NMR spectra were obtained using a 8.4 µs (60°) pulse, 6 kHz spectral width, and 2.5 s relaxation delay with water presaturation, with 64 transients collected into 16 384 data points requiring a 4.5 min acquisition time. 3376
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 9, 2007
Datasets were zero-filled to 32 768 points, and exponential line-broadenings of 0.5 Hz were applied before Fourier transformation. 1-D 1H NMR spectra of mantle samples from Southampton were obtained on the same NMR spectrometer (at 300 K) using excitation sculpting to suppress the water resonance (25). All parameters were as stated previously except that 80 transients were collected, requiring 6.5 min. All spectra were phased manually, baseline-corrected using a quadratic function, and calibrated (TMSP at 0.0 ppm) using TopSpin (version 1.3, Bruker). Spectral Pre-processing and Statistical Analysis. NMR spectra were converted to a format for multivariate analysis using custom-written ProMetab software in Matlab (version 7.1; The MathsWorks, Natick, MA). Each spectrum was segmented into 0.005 ppm bins between 0.6 and 10.0 ppm with bins from 4.70 to 5.15 ppm (water) and 7.60 to 7.76 ppm (residual chloroform) excluded from all spectra. Additional bins between 7.28 to 7.32 and 7.44 to 7.49 ppm were excluded from the Port Quin mantle spectra due to resonances detected in the buffer. The total area of each spectrum was normalized to 1, and for adductor muscle spectra, bins between 7.08 to 7.10 and 7.84 to 7.88 ppm containing pH-sensitive resonances were each compressed into a single bin resulting in a matrix of 56 samples × 1956 bins. Similarly, for mantle spectra, bins between 7.76 to 7.79, 7.84 to 7.87, 8.26 to 8.29, 8.61 to 8.63, and 8.70 to 8.74 ppm were each compressed into a single bin resulting in a matrix of 52 samples × 1956 bins. Each matrix was subject to the generalized log transformation (26) to stabilize the technical variance across the bins. The transformation parameter, λ, was determined by recording NMR measurements on one tissue that was subdivided into six fractions to ascertain the technical variance associated with sample preparation and NMR analysis: λ ) 5.255 × 10-9 and 1.043 × 10-7 for adductor muscle and mantle, respectively. Data were mean-centered before principal components analysis (PCA) and partial least-squares discriminant analysis (PLS-DA) using PLS_Toolbox (version 3.53, Eigenvector Research, Manson, WA). PCA allowed the differences and similarities between NMR metabolic fingerprints to be visualized in the form of scores plots, where samples that are metabolically similar cluster together. The corresponding PCA loadings plots were used to identify the metabolic basis of the clustering. Initial PCA results identified two adductor muscle spectra and four mantle spectra for exclusion based upon Q residuals and T2 Hotelling statistics that exceeded 95% confidence limits (inspection of these spectra revealed poor line-shapes and baseline artifacts). There is a fundamental difference between PCA and PLS-DA in that PCA is unsupervised, and so the clustering of samples in the scores plot is based upon the metabolites with the greatest variability, irrespective of whether these metabolites are involved in hypoxia. PLS-DA is a supervised analysis that identifies which metabolites discriminate between specific classes of samples (e.g., between hypoxia and normoxia and between male and female). The quality of PLS-DA models was assessed using cross-validation with five-way split Venetian blinds (27). Peaks were identified and quantified using Chenomx NMR Suite with associated libraries (version 4.5; Chenomx Inc., Edmonton, Canada), and homarine was identified using information from a previous study (28). The metabolite concentrations were normalized to total spectral area, and Student’s t tests were used to test for significant changes between groups (Microsoft Excel). Species Determination by PCR. To identify the species of mussels, DNA was extracted (DNeasy kit, Qiagen, Crawley, U.K.) from 25 mg of gill tissue. The 5′ untranslated nonrepetitive region of the polyphenolic adhesive protein gene (Glu-5′) was amplified using the PCR protocol of Wood et al. (29) and the Me-15 and Me-16 primers from Inoue et al. (30).
FIGURE 1. (a) Representative 1-D 1H NMR spectrum of adductor muscle from M. galloprovincialis. Key: (1) unidentified metabolite, (2) homarine, (3) tyrosine, (4) glycine-betaine, (5) glycine, (6) aspartate, (7) glutamate, (8) succinate, (9) acetoacetate, (10) glutamine, (11) arginine/phosphoarginine, (12) alanine, (13) lactate, (14) leucine, (15) valine, and (16) isoleucine. (b) PCA scores plot showing the metabolic variability in adductor muscle arising from sampling protocol and hypoxic stress. Group classifications: field hypoxic (2), field normoxic (9), laboratory hypoxic (∆), and laboratory normoxic (0). Ellipses represent (SD about each group mean, and dashed lines are for hypoxic animals. The resulting PCR products were analyzed on a 2% agarose gel alongside a DNA ladder (Hyperladder V, Bioline, London, U.K.). Species-specific bands were expected at 180 base pairs (bp) for M. edulis and 126 bp for M. galloprovincialis, with both bands present for heterozygous individuals of the viable hybrid species.
Results and Discussion Effect of Sampling on the Overall Metabolic Variability in Adductor Muscle. A representative 1H NMR spectrum of an adductor muscle extract is shown in Figure 1a. Several metabolite classes were observed including amino acids (e.g., alanine), organic osmolytes (e.g., glycine-betaine), phosphagens (e.g., phosphoarginine), and Krebs cycle intermediates (e.g., succinate). The PCA scores plot of the NMR metabolic fingerprints of the adductor muscles from Port Quin mussels is shown in Figure 1b. Clearly, the metabolomes of the 54 adductor muscle samples overlap extensively between the laboratory and the field groups and between the hypoxic and the normoxic groups. The major source of variation (28.85% along PC1) is unassociated with both the hypoxic stress and the sampling method. The metabolites contributing to PC2, however, are related to the hypoxic stress, with hypoxic samples having somewhat larger PC2 scores. The overall metabolic variability for each group (calculated as the area of the ellipse that surrounds the PC1 and PC2 error bars) is 2.78 and 5.47 for hypoxic and normoxic animals from
the field and 5.38 and 6.92 for hypoxic and normoxic animals from the laboratory. To assess if the differences between the variances of the four groups were statistically significant, Bartlett’s test was applied to the PC1 and PC2 scores. PC1 showed no significant difference between the variances of the four groups (p ) 0.966), yet the differences between the metabolic variabilities along PC2 did approach significance (p ) 0.098). This indicates that, contrary to our hypothesis, the adductor muscles sampled directly from the field show a somewhat lower (although not statistically significant) metabolic variability than animals stabilized in the laboratory, especially for the stressed hypoxic phenotype. Effect of Sampling on the Visibility of Hypoxia in Adductor Muscle. Although the effects of hypoxia are barely visible when considering all 54 adductor muscle samples (Figure 1b), PCA of the field-exposed animals alone shows a clear separation between normoxia and hypoxia with the stress response characterized by an increase in PC2 scores (Figure 2a). This distinctive response is surprisingly not mirrored in the equivalent PCA scores plot for laboratoryexposed animals (Figure 2b). To determine if hypoxia did in fact occur in the laboratory animals, we applied PLS-DA. The scores plot shows a clear distinction between the hypoxic and the normoxic groups (Figure 2c) that, from consideration of the weightings, was confirmed to arise from hypoxia. This strongly suggests that the lack of a visible hypoxic stress in the PCA scores plot (Figure 2b) is due to greater metabolic variability in the laboratory-exposed mussels as compared to the field mussels, which masked the effects of the applied stressor. Originally, we hypothesized that hypoxia would be more readily detectable in the laboratory animals due to their lower metabolic variability following 60 h in a controlled environment. Several reasons could explain why this is not the case: first, although animals were transported using a standard ICES protocol (22), they would still have undergone a significant metabolic stress, and the stabilization period may not have been of optimal duration. In addition, although holding conditions were deemed standard for in vivo experimentation with mussels, the experimental environment (artificial seawater, no food, and no tidal cycle) is significantly different from the field. These factors apparently induced sufficient metabolic variability in the laboratory-exposed mussels to mask the effect of hypoxia in the PCA model, although the supervised PLS-DA method could still delineate the applied stress. Metabolic Changes during Acute Hypoxia. The PCA loadings for field-exposed mussels (Figure 3a) and PLS weightings for laboratory-exposed animals (Figure 3b) can be used to determine the metabolic changes that correlate with hypoxia. On the basis of the corresponding scores plots (Figure 2a,c), the axes providing the greatest discrimination between normal and stressed phenotypes are PC2 and LV1 + 0.539 LV2 for field and laboratory mussels, respectively. These plots represent biomarker profiles of hypoxia, where positive peaks correspond to metabolites that are elevated in hypoxic animals and negative peaks to metabolites that are elevated in normoxic mussels. Peaks with the largest loadings were identified and quantified, and fold changes following hypoxia were calculated (Table 1). For both field and laboratory mussels, the most pronounced change during hypoxia was a large increase in succinate (p < 0.0001 and p < 0.001, respectively) and, for the field animals, an increase in an unidentified metabolite at 8.60 ppm (p < 0.0001). A slight increase in alanine (not significant) was also observed. These changes can be rationalized as succinate and alanine, not lactate, are the major end-products of anaerobic metabolism in marine bivalves (31-33). Effect of Sampling on the Overall Metabolic Variability in Mantle. The PCA scores plot of the NMR metabolic VOL. 41, NO. 9, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
3377
FIGURE 3. (a) PC2 loadings and (b) LV1 + 0.539 LV2 weightings plots showing the metabolic effect of hypoxia in adductor muscle for field and laboratory environments, respectively. Positive peaks correspond to metabolites that are at higher concentrations in hypoxic mussels. Metabolites showing significant changes are labeled (1) succinate, (2) unidentified singlet at 8.60 ppm, and (3) homarine.
TABLE 1. Metabolites in M. galloprovincialis Adductor Muscle Identified from PCA Loadings and PLS Weightings Plots that Change Concentration Following a 2 h Hypoxic Stress in Field and Laboratorya field exposure
FIGURE 2. Scores plots showing effect of hypoxia (2) vs normoxia (9) in adductor muscle: (a) PCA of field-exposed animals, (b) PCA of laboratory-exposed animals, and (c) PLS-DA of laboratory-exposed animals. The arrows show the mean metabolic trajectory from a normoxic to hypoxic phenotype. fingerprints of mantle from Port Quin mussels is shown in Figure 4a. Unlike in adductor muscle, two distinct clusters formed in PCA space, but the separation was unrelated to either the sampling strategy or the hypoxic stress. The corresponding loadings plot revealed several metabolites that contribute to this separation, in particular, glycine (Figure 4b). We sought to rationalize this unexpected clustering by considering both the mussel species and the phenotype. Metabolomics experiments have defined species variability 3378
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 9, 2007
laboratory exposure
metabolite
fold-change
p
fold-change
p
succinate unidentifiedb homarine glycine-betaine aspartate alanine
7.402 2.598 1.021 0.848 0.902 1.029
