'Systems toxicology' approach identifies coordinated metabolic responses to copper in a terrestrial non-model invertebrate, the earthworm Lumbricus rubellus
- 15k Downloads
New methods are needed for research into non-model organisms, to monitor the effects of toxic disruption at both the molecular and functional organism level. We exposed earthworms (Lumbricus rubellus Hoffmeister) to sub-lethal levels of copper (10–480 mg/kg soil) for 70 days as a real-world situation, and monitored both molecular (cDNA transcript microarrays and nuclear magnetic resonance-based metabolic profiling: metabolomics) and ecological/functional endpoints (reproduction rate and weight change, which have direct relevance to population-level impacts).
Both of the molecular endpoints, metabolomics and transcriptomics, were highly sensitive, with clear copper-induced differences even at levels below those that caused a reduction in reproductive parameters. The microarray and metabolomic data provided evidence that the copper exposure led to a disruption of energy metabolism: transcripts of enzymes from oxidative phosphorylation were significantly over-represented, and increases in transcripts of carbohydrate metabolising enzymes (maltase-glucoamylase, mannosidase) had corresponding decreases in small-molecule metabolites (glucose, mannose). Treating both enzymes and metabolites as functional cohorts led to clear inferences about changes in energetic metabolism (carbohydrate use and oxidative phosphorylation), which would not have been possible by taking a 'biomarker' approach to data analysis.
Multiple post-genomic techniques can be combined to provide mechanistic information about the toxic effects of chemical contaminants, even for non-model organisms with few additional mechanistic toxicological data. With 70-day no-observed-effect and lowest-observed-effect concentrations (NOEC and LOEC) of 10 and 40 mg kg-1 for metabolomic and microarray profiles, copper is shown to interfere with energy metabolism in an important soil organism at an ecologically and functionally relevant level.
KeywordsNuclear Magnetic Resonance Nuclear Magnetic Resonance Spectroscopy Citric Acid Cycle Metabolomic Analysis Copper Exposure
Understanding biological responses to individual toxic chemicals and chemical classes is clearly of key importance for pollution assessment, both for monitoring exposure to existing environmental contamination and for informing the risk assessment of off-target effects. However, ecotoxicological research frequently focuses only on easily measurable endpoints, typically mortality, although more sensitive tests on effect endpoints such as reproduction and growth are also used widely. Thus, a major challenge for ecotoxicology is understanding toxic mechanisms at a molecular level, and how these molecular changes relate to functional changes at the organism and population level . The 'ecotoxicogenomic' post-genomic approach has clear benefits, and is currently generating interest from end users such as regulatory authorities as well as from research scientists [2, 3]. In order for this potential to be realised, a solid bedrock of research is needed to characterise the fundamental responses of important test organisms to a range of model toxins covering a wide chemical space. It will be important to determine just how specific omic fingerprints of toxicity are, and whether they can be used successfully to distinguish between different modes of toxic action, and hence yield novel information on mechanistic toxicology. This 'systems toxicology' approach has been applied in widely used model organisms such as the laboratory rat and other vertebrates [4, 5, 6, 7]. However, these animal models have the benefit of many more existing data [8, 9]. In addition, it is often easier to perform manipulative experiments, and there is a much greater scope for complementary mechanistic cell-based work, such as histopathology. In contrast, the situation with non-model, ecologically relevant species is quite different.
The term 'ecologically relevant' is not precisely defined: clearly the most relevant level for studying the effects of chemicals is the community and/or ecosystem, and there are approaches which aim to understand, or at least quantify, responses to pollution at this level (see, for example, [10, 11, 12, 13, 14]). Here, however, we refer to controlled studies on single species that may already be widely studied but are not classic model organisms; for example, animals used in regulatory ecotoxicity tests fall into this category, such as the earthworm Eisenia fetida, the enchytraeid Enchytraeus albidus, and collembolans Folsomia candida and Orchesella cincta for terrestrial, and Daphnia magna, Gammarus pulex, chironomid larvae and Mytilus species for aquatic testing. Working with these animals presents some common challenges: none has a fully sequenced genome; it is not generally possible to obtain antibodies against specific molecular targets; they are often so small as to preclude ready dissection of internal organs or tissues; it is impossible or extremely difficult to modulate gene activity, for example by creating knockout strains; and there is in general much less knowledge about fundamental biological systems, such as signalling pathways or gene regulation, in these organisms. Modern omic approaches offer a potential opportunity to circumvent some of these drawbacks [15, 16, 17, 18, 19, 20, 21, 22]. In particular, metabolomics and metabonomics have one great advantage for work with non-model organisms: because metabolites are detected directly, and primary metabolites at least are identical across different species, samples can trivially be analysed with no need for prior knowledge of the gene and protein sequences . Metabolomics also reports on the final integrated phenotype of an organism, as metabolism is the final downstream product of gene and enzyme regulation [24, 25, 26, 27]. As a consequence, we decided to carry out an integrative study of the metabolic response of Lumbricus rubellus to copper, using both nuclear magnetic resonance (NMR)-based metabolic profiling and cDNA microarrays for transcript profiling.
The earthworm L. rubellus is a common species with a worldwide distribution [28, 29, 30, 31]. It is found even in shallow and contaminated soils, so it is appropriate for both laboratory and field studies . It has also been the subject of an expressed sequence tag (EST) sequencing project, permitting the construction of cDNA microarrays . Copper is an essential element that is also highly toxic to soil invertebrates in high concentrations. Hence, as well as inducing general toxic-response pathways, there will also be specific biological mechanisms for copper handling that may be expected to be perturbed, and thus copper is an excellent model toxin for demonstrating an integrative ecotoxicogenomics approach. We exposed worms to sub-lethal levels of copper in a semi-field situation using buried mesocosms, and monitored the dose response using transcriptomics (using a cDNA microarray fabricated with 8,129 EST reporters representative of all consensus gene objects generated from 17,225 high-quality ESTs) and metabolomics (using proton NMR spectroscopy). We also measured changes in reproduction and general condition as measured by body weight and by a well-characterised cellular bioassay; monitoring these functional endpoints is important for phenotypic anchoring of the omic data [15, 34]. Our aim was to look for links between the different levels of information (metabolism, transcription and functional) to obtain stronger inferences about the mechanistic effects of copper than could be gathered from the individual datasets alone. A secondary aim was to test the hypothesis that copper exposure up-regulates histidine metabolism in L. rubellus [35, 36].
Results and discussion
Ecological and functional endpoints (survival, weight change, reproduction rate and neutral red retention by coelomocytes) have been reported in a previous study . In brief, there was no effect on mortality, confirming that the exposure was appropriate for probing sub-lethal molecular responses, but the other functional assays all showed a response at medium to high levels of copper (40 to 160 mg/kg and above).
List of NMR-visible metabolites assigned in earthworm extracts
There is currently no 'metabolite ontology' database that allows one to systematically annotate individual compounds into different groups; approaches based on network topology for the definition of elementary modes are exciting, but are still under development, and cannot be applied at the whole-organism level for metazoans . Hence, we performed a simple annotation based on existing biochemical knowledge. We note that our annotations were not performed systematically; it is also important to realise that many metabolites should be annotated to multiple categories, which we did not attempt to do. Nevertheless, this simple approach may be a useful first step in interpreting metabolomics data.
Several metabolite groups exhibited a coordinated response; this was especially clear for the lipophilic amino acids (Figure 2). Several other possible coordinated responses were identified, including sugars, nucleotides, organic acids (including several synthesised through the citric acid cycle) and, possibly most interestingly, 'cell membrane compounds' (Figure 2). It should be noted that attempting to assign 'metabolite ontologies', at however limited a level, offers an option that simply does not exist to any great extent for genes: that of chemical similarity. (Some kind of chemical classification of genes might also be made, for example, on the basis of percentage occurrence of DNA bases, but this would be of very limited applicability.) Of the groups we have labelled in Figure 2, the 'cell membrane' group does not include any compounds that directly form cell membranes, but, instead, those that could be involved in either the anabolism or catabolism of lipid compounds. It should be noted that this is the most structurally diverse group that we labelled, containing an amino acid (glycine, which is a direct precursor to serine and, hence, sphingolipid metabolism, and is a breakdown product of choline and betaine), as well as metabolites that form polar lipid head groups (inositol compounds and phosphoethanolamine). In addition, we assigned the earthworm metabolite HEFS to this group: the very high tissue concentration of HEFS implies it has some kind of structural or stabilising role. As amphiphilic compounds are known to stabilise cell membranes during desiccation stress , and desiccation is a common and severe hazard faced by earthworms, we argue that the high HEFS concentrations probably reflect a role in membrane stabilisation.
Non-targeted metabolite profiling has often been explicitly categorised as a biomarker discovery tool [48, 49], with the implication that the aim of a metabolomic study is to discover a small number of biomarkers in amongst the biological noise. This approach may be advantageous when the desired outcome is a screening tool (for instance, for disease, or for environmental pollution), when a single (or few) robust biomarker(s) will enable the use of simpler and more robust predictive models (for example, classical linear discriminant analysis ) and targeted detection. An additional advantage not often made explicit is that this greatly enhances the transferability between different laboratories and, hence, the overall potential use and scientific value of the biomarkers. However, this approach is limited when seeking to integrate large datasets and to relate these to biological functions. In the current study, considering the coordinated responses of functionally related metabolites showed clear group responses, which would not have been obvious if only the most strongly changing metabolites had been investigated.
Histidine compound metabolism
We used a K-means clustering approach to show that the transcripts fell into four response groups (clusters), which displayed similar dose-dependent profiles with respect to copper exposure. One group comprised of genes showing a general dose-dependent increase with respect to copper (Figure 6D), and two groups displayed the converse relationship being separated by the degree of down-regulation observed in the two lower copper doses (Figure 6E). The most intriguing cluster, however, were those transcripts that were up-regulated when exposed to a low dose of copper with a subsequent down-regulation at higher doses (Figure 6F). This profile indicates that the organisms' hormetic response to copper is also manifest at the level of transcription .
To complement the purely statistical approaches that provide insight into the coordination of transcript response, we also analysed the bias in gene function within transcripts (selected by unsophisticated filtering) to identify those that displayed more than twofold changes in transcript level in response to any of the copper exposures. We chose this group to be 'inclusive' because, although we recognise that it includes a small quantity of noise due to the quantity of transcripts analysed, it provided a more complete representation of functional group effects. This was especially desirable given that, because functional assignment was based on the top human orthologues, only about one-third of the reporters could be assigned confidently. We used a background created by those human orthologues assigned to the complete reporter set to calculate the representation bias for gene ontology terms for those orthologues assigned to transcripts exhibiting copper-induced twofold change in expression . The ontology groups that were significantly overrepresented (p < 0.1) are summarised in Additional file 7. The clear disruption in genes associated with mitochondrial electron transport strongly implies a copper-induced mitochondrial dysfunction, consistent with previous observations in other organisms [59, 60, 61]. Even a slight reduction in the capacity for adenosine triphosphate (ATP) production by aerobic respiration would result in a redistribution of energy production through anaerobic processes, and this is indeed evident through the changes in genes involved with sugar mobilisation for energy production (Additional file 7). We also observed transcript changes that we interpret as representing functional interactions between copper and other essential metal ions, in particular changes in expression of calcium- and iron-binding proteins. Finally, there were clear alterations in lipid metabolism at the transcript level, complementing the observations made at the metabolite level.
Samples for both microarray and metabolomic analysis were chosen by selecting worms with the best condition from each independent mesocosm, thus avoiding obviously diseased and/or parasitised worms. This has two major advantages for the molecular data. First, it avoids major confounding factors, for example, from chance-affected worms. Secondly, even in the case that there would be an interaction with copper treatment, for example, through copper toxicity decreasing the resistance to external factors, one would expect the molecular mechanisms induced to be general and stress-related. Thus, even though we may have selected individuals that did not provide the most accurate representation of ecological fitness under metal stress, the molecular endpoints will be more closely related to specific copper-induced mechanisms.
Integration of the different datasets
There are multiple possibilities for integrating omic datasets, but these can be considered essentially to fall into two categories: statistical, that is, relying wholly on data-driven associations between variables (for example, [4, 72]); and knowledge-based, that is, using prior knowledge about biological organisation and pathways/networks (for example, [73, 74]). Clearly, an optimal solution would use information from both of these approaches; but then this, in the limit, effectively becomes a full model of metabolism, and this is not currently achievable, even for highly controllable unicellular model organisms. Here we have chosen a separate analysis of datasets and subsequent combination of observations.
These observations are directly reflected by the metabolomic data, where a set of organic acids that are citric acid cycle intermediates were altered as a functional group, and there was a decrease in free sugar concentrations (Figure 3). In addition, when these observations are considered together with the links to specific transcripts and functional endpoints (Figure 7), and the changes in metabolites directly involved in maintaining cellular energy levels (vide infra), it is clear that energy metabolism in L. rubellus was profoundly impacted by copper treatment.
Interpretation of energy reserve data
Some copper-responsive metabolites were those related to maintenance of cellular energy levels, in particular adenosine phosphates. In general, adenosine monophosphate (AMP) and adenosine diphosphate (ADP) levels were both high and correlated to each other, and also showed some positive correlation to copper. This, combined with the lack of any detectable 1H signal for phospholombricine, implies that the worms' cellular energy reserves were exhausted, and thus not truly representative of in vivo metabolism. This is not a simple artefact of the extraction procedure: for comparison, a neutralised 6% perchloric acid extract tells a similar story about the energetic state, that is, AMP and ADP both higher in concentration than ATP, and also completely lacking a signal from phospholombricine (Additional file 9). The worms were sampled under conditions that would have preserved metabolic integrity, by flash-freezing into liquid nitrogen, storage at -80°C, and lyophilised tissues extracted directly into ice-cold solvent/water mixtures. In addition, the high quality of the mRNA extracted for the microarray analyses confirms the preservation of the biological integrity of the samples. Thus, we conclude that the general trend of the adenosine phosphates (AXP) data (ATP decreasing, and AMP and ADP increasing, with increased soil copper) is a reflection of a general disturbance in energy metabolism, as also shown by decreases in free sugars, and up-regulation of transcripts for breakdown of these sugars and catabolism of energy reserves, even if it may not be a true 'snapshot' of in vivo metabolism. Indeed, any extraction procedure is necessarily selective and different protocols will give different windows on the metabolome. Techniques such as high-resolution magic-angle-spinning of tissue biopsies (and even larger samples) are increasingly used for biochemical and biomedical investigations , even though this will lead to far more extensive enzymatic changes than the extractions used here.
There was a clear impact on both metabolomic and microarray profiles at the second-lowest additional copper concentration of 40 mg/kg. This is either as sensitive or more sensitive than the functional parameters measured, such as weight change, reproduction and the neutral red retention time (NRRT) bioassay, all of which had lowest-observed-effect concentrations (LOECs) of 160 mg/kg or higher, and much more sensitive than an endpoint based on mortality for which no significant effect was found at any of the tested exposure concentrations (that is, no-observed-effect concentration (NOEC) greater than 480 mg/kg) . This sensitivity seems a fairly general rule, at least for laboratory experiments, where metabolomics has been shown to be extremely sensitive to individual chemicals compared with traditional endpoints . Such sensitivity provides an interesting insight into the likely value of such analyses for regulatory testing regimes. Here omic approaches may provide insights on low effect level metabolic dysfunctions which may have implication over a life-time exposure, but may not be revealed in the time-limited chronic bioassays that form a cornerstone of modern chemicals policy .
Like some of the observed phenotypic responses (for example, reproduction rate and weight change), the omic data (both transcripts and metabolites) very clearly demonstrate a non-linear response to copper. This was also true for the weight change data, with an initial increase in body weight followed by a clear decrease at higher concentrations. It has been argued that hormesis is a very general response to potential toxins, with low-level exposure either being beneficial or stimulating a compensatory response . While the response seen for copper may merely represent the fact that this metal is both essential and toxic for earthworms, it is currently the case more generally that a mechanistic basis for understanding such 'hormetic' responses is lacking. While initial suggestions have been made to explain hormesis in terms of an overcompensation of homeostasis to disruption that is mediated by the different affinities of stimulatory and inhibitory regulatory pathways , such suggestions have yet to be confirmed or refuted in empirical studies. Our data clearly show that the hormetic response in this case is manifest at a molecular level, indicating a possible role for metabolomics in understanding the mechanistic nature of the response.
Drawing reliable inferences from omic data is often difficult, especially for non-model organisms which may be less well annotated than standard laboratory models. Here we have used a dual metabolomic and transcriptomic strategy, with each element providing complementary data, thus reinforcing the inferences made from the independent datasets. Interpreting the data in terms of functional groups, for both transcript and metabolite data, showed clear molecular group responses to copper. The integrated dataset indicated a clear alteration of energy metabolism as a result of sub-lethal copper exposure, with an increased switch to metabolism of stored carbohydrates, presumably as a consequence of copper interfering with mitochondrial function and reducing the amount of energy available from oxidative phosphorylation. These molecular impacts resulted in higher-level endpoints, including a reduction in body weight at high copper levels, reflecting the changes in energy metabolism, and a decrease in lysosomal integrity, indicating effects on membranes that could parallel the observed impacts on mitochondrial function.
Exposure and sampling
A full description of the earthworm exposure and sampling conditions is given by Spurgeon et al . Briefly, worms were exposed for 70 days under field conditions in mesocosms to controlled levels of soil copper (0, 10, 40, 160 and 480 mg additional copper per kilogram dry weight soil; original copper content of the soil was 16.1 mg/kg), with four replicate mesocosms per dose level. A number of functional endpoints were measured including survival, weight change (average value per mesocosm), NRRT and cocoon production rate. Fifteen adult worms were exposed per mesocosm, and at the end of the experiment, three worms were used to measure tissue copper concentrations; four worms were used to measure NRRT and the three worms maintaining highest condition in each replicate were pooled and used for both transcriptomic and metabolomic analysis. This avoided inclusion of diseased or heavily parasitised worms within the omic experiments that may have otherwise confounded results. The worms were flash-frozen in liquid nitrogen, and ground to powder in a mortar and pestle under liquid nitrogen to obtain a single sample for each mesocosm. Total mRNAs from the samples were then isolated using established protocols [79, 80]. The remaining sample was lyophilised without allowing the sample to thaw, and stored at -80°C until extracted for metabolomic analysis.
The L. rubellus EST project  has established a database of over 17,000 ESTs from unexposed and chemically exposed earthworms (see  for details). All sequences were assembled into clusters and annotated using the PartiGene pipeline . To fabricate the glass slide cDNA microarray, a representative EST (usually the longest) was selected from each of the 8,029 clusters assembled from the ESTs (see  for full details). This sequence was polymerase chain reaction (PCR)-amplified and aliquots (5 μl) of concentrated products mixed in 384 well plates with an equal volume of dimethyl sulphoxide (DMSO). These were then printed onto Ultra-GAP glass slides (Corning) using 48 SMP3 pins (Telecham) mounted in a Spotarray 72 (Perkin-Elmer). Landmarks were introduced at the left-hand corner of each sub-array by the introduction of five replicates of the Lucida Scorecard (Amersham) gene reporters, which show no cross-reactivity to earthworm transcripts. All reporters were cross-linked to the slide by baking at 80°C for 2 hours, and UV cross-linking.
Lucida Scorecard test spike (Amersham Life Sciences) was added to 10 μg of total RNA prior to oligo-dT reverse transcription and coupling to Cy3 using an indirect amino amyl procedure. Clean-up of labelled targets, yield and integrity were all measured according to Owen et al . A reference design was used in which approximately 30 pmol of Cy3 labelled target RNA was hybridised against the common oligonucleotide reference (representing 30 pmol of Cy5). The reference used was a 65–70 mer oligonucleotide designed against the vector sequence between the amplification primer binding site and cDNA insert. Use of this universal reference design allowed a comparison of slides to be made both within and between experiments. After hybridisation (18 hours), slides were washed and imaged according to Owen et al . Array images were subjectively quality controlled for artefacts that would compromise quantification such as background effects and spot morphology prior to image analysis with Imagene 5.0 (Biodiscovery). Subsequently, the calibration standards from the Lucida Scorecard were analysed to objectively assess the sensitivity range and to define both saturation and background readings (Additional file 10). The microarray data can be accessed through the NEBC file store  (EnvBase accession number [EGCAT:4024]).
Statistical analysis of microarray data
Numeric data were imported into LimmaGUI [85, 86] which allowed the subtraction of background measurements, generation of additional quality control plots and subsequent normalisation using a within-array Tiplowess manipulation. Abnormally distributed samples were excluded from further analysis. Normalised data were subsequently imported into GeneSpring 7.3 (Agilent Technologies, Stockport, UK) and represented relative to the median expression within the control group. Final, quantity assessment was performed by visualising a box plot of the normalised data together with the generation of MA plots for the average data from each dose (Additional file 11). Abnormally distributed samples were excluded from the analysis.
Using GeneSpring, the three control slides and three slides for the exposure concentration that most closely matched the cocoon production EC10 for each chemical were used to identify chemically responsive genes. The data were first filtered to include only spots flagged as present in 3 out of the 15 slides. This filtered dataset was then used to generate three gene lists for each dataset. These were: (1) genes with a more than twofold difference in mean expression between control and exposed samples; (2) genes with significantly different (p < 0.05) expression between control and exposed samples after correction for multiple sample testing .
Tissue samples (20–30 mg) were homogenised into 3 ml of extraction solvent (Heidolph SilentCrusher S) using a modified Bligh and Dyer protocol [87, 88], in which a monophasic chloroform/methanol/water extraction is followed by the addition of water and chloroform, splitting the sample into two phases. The proportion of water was adjusted for the fact that lyophilised tissue was used, assuming a 90% water content in normal tissue. Both fractions were dried at 40°C using a rotary vacuum concentrator. The polar fraction was then resuspended in 0.65 ml of NMR buffer (100 mM phosphate buffer pH 7.0 in 2H2O, also containing 0.98 mM sodium trimethylsilyl-2,2,3,3-2H4-propionate (TSP); the 2H2O provided a field frequency lock for the spectrometer and reduced the signal from water protons) and centrifuged (5 minutes, 16,000 g). The supernatant (0.6 ml) was then transferred to 5 mm NMR tubes. The aqueous samples were analysed at 300 K on a 14.1 T DRX600 Avance NMR spectrometer (Bruker BioSpin, Rheinstetten, Germany) with a 600 MHz proton resonance frequency, equipped with a 5 mm broadband inverse probe. The samples were run using an automatic tube changer, over a period of about 12 hours, during which time they were kept at room temperature; the samples were loaded onto the tube changer in randomised blocks. A one-dimensional NOESY sequence with a mixing time of 100 ms was used for water suppression of the residual HOD in the NMR buffer, using a 30 Hz presaturation pulse. The spectra were acquired for 128 transients, with four dummy scans, into 32 K data points over a 12 kHz spectral width. A 3.5-second longitudinal relaxation recovery delay was added for each transient, giving a recycle time of 5 seconds. The chloroform fraction was resuspended in 0.65 ml of CDCl3 containing 0.03% tetramethylsilane (TMS) and transferred to 5 mm NMR tubes. The lipid samples were then analysed using a Carr-Purcell-Meiboom-Gill (CPMG) sequence, with a 8.65-second longitudinal relaxation delay giving an approximately 10-second recycle time (the CPMG sequence gave improved baselines and slight reduction of broad resonances compared to a simple pulse-acquire experiment). The raw data (free induction decays (FIDs)) for all spectra, and the fitted compound data for the polar extracts (vide infra), are available through the NEBC file store  (EnvBase accession number [EGCAT:4024]). The fitted compound data are also provided as Additional file 12.
NMR processing and data analysis
The spectra were initially processed using iNMR 2.2.7 (Nucleomatica, Molfetta, Italy). The summed FIDs were multiplied by an exponential window function equivalent to 0.5 or 1 Hz line broadening (aqueous and lipid spectra, respectively). They were then zero-filled by a factor of 1.5, and Fourier transformed. Phasing was carried out using the automatic 'metabolomic phase correction' option, and adjusted manually where necessary; baseline correction was performed using an automatic first-order polynomial fit. All spectra were referenced to TSP/TMS at 0 ppm. The spectra from polar (aqueous) samples were further analysed using Chenomx NMR Suite 4.6 (Chenomx, Edmonton, AB, Canada). This software fits idealised spectra made up of combinations of Lorentzian line shapes, based on spectra of authentic standards, and estimates compound concentrations using TSP as an internal quantitation standard. Compounds were fitted using the proprietary Chenomx 600 MHz library, to which we added standards for glucose-6-phosphate, HEFS and lombricine. HEFS and lombricine are not commercially available; we purified HEFS from earthworm tissues using solid-phase extraction with a mixed-mode C18/anion exchange phase. Only a single peak for lombricine was fitted, based on a spectrum of an existing tissue extract acquired under fully relaxed conditions, and concentration estimated by comparing this with the integral of a known concentration of TSP. Thus, it is likely that the absolute accuracy of the lombricine concentrations will be lower than for the other compounds reported in this study, although the relative levels (precision) will be comparable. The fitted compound concentrations are available in Additional file 12.
For pattern recognition, data were normalised following the method of Dieterle et al  in which each profile is compared with a representative reference (in our case, a median of all sample profiles). The relative fold change for each variable in turn is calculated, and all values for a spectrum are then divided by the median fold change for that spectrum. Data were log-transformed by log(n i + 0.018) for concentrations (expressed as mM); the value of 0.018 was chosen because it effectively removed the correlation between intensity and standard deviation for a series of five technical replicates, that is, increasing homoscedasticity (the principle is discussed elsewhere [90, 91]). Factor analysis and hierarchical cluster analysis on non-centred data was carried out using Aabel 2.2 (Gigawiz, Tulsa, OH, USA). (Note that the factor analysis here is exactly equivalent to PCA carried out using the correlation matrix, but we have retained the term 'factor analysis' for consistency with the terms used in the software package.) The lipid spectra were integrated within selected regions using iNMR (integral boundaries given in Table 2).
We thank Chenomx, Inc., for access to and assistance with Chenomx NMR Suite 4.6 software. Funding was provided by NERC (grant references NER/T/S/2002/0021 and NE/D007755/1). Jennifer Owen is thanked for the fabrication of the L. rubellus cDNA microarrays. Hector Keun and Tim Ebbels are thanked for valuable discussions.
- 13.Newman MC, Ownby DR, Mezin LCA, Powell DC, Christensen TRL, Lerberg SB, Anderson BA: Applying species-sensitivity distributions in ecological risk assessment: assumptions of distribution type and sufficient numbers of species. Environ Toxicol Chem. 2000, 19: 508-515.Google Scholar
- 19.Soetaert A, Moens LN, Ven Van der K, Van Leemput K, Naudts B, Blust R, De Coen WM: Molecular impact of propiconazole on Daphnia magna using a reproduction-related cDNA array. Comp Biochem Physiol C. 2006, 142: 66-76.Google Scholar
- 21.Knigge T, Monsinjon T, Andersen OK: Surface-enhanced laser desorption/ionization-time of flight-mass spectrometry approach to biomarker discovery in blue mussels (Mytilus edulis) exposed to polyaromatic hydrocarbons and heavy metals under field conditions. Proteomics. 2004, 4: 2722-2727.CrossRefPubMedGoogle Scholar
- 22.Kuperman RG, Checkai RT, Ruth LM, Henry T, Simini M, Kimmel DG, Phillips CT, Bradley BP: A proteome-based assessment of the earthworm Eisenia fetida : response to chemical warfare agents in a sandy loam soil. Pedobiologia. 2003, 4: 617-621.Google Scholar
- 24.Raamsdonk LM, Teusink B, Broadhurst D, Zhang NS, Hayes A, Walsh MC, Berden JA, Brindle KM, Kell DB, Rowland JJ, Westerhoff HV, van Dam K, Oliver SG: A functional genomics strategy that uses metabolome data to reveal the phenotype of silent mutations. Nat Biotechnol. 2001, 19: 45-50.CrossRefPubMedGoogle Scholar
- 30.Deibert EJ, Utter RA: Earthworm (Lumbricidae) survey of North Dakota fields placed in the US Conservation Reserve Program. J Soil Water Conserv. 2003, 58: 39-45.Google Scholar
- 32.Spurgeon DJ, Weeks JM, Van Gestel CAM: A summary of eleven years progress in earthworm ecotoxicology. Pedobiologia. 2003, 47: 588-606.Google Scholar
- 33.Sturzenbaum SR, Parkinson J, Blaxter M, Morgan AJ, Kille P, Georgiev O: The earthworm Expressed Sequence Tag project. Pedobiologia. 2003, 47: 447-451.Google Scholar
- 35.Gibb JOT, Svendsen C, Weeks JM, Nicholson JK: 1H NMR spectroscopic investigations of tissue metabolite biomarker response to Cu(II) exposure in terrestrial invertebrates: identification of free histidine as a novel biomarker of exposure to copper in earthworms. Biomarkers. 1997, 2: 295-302.CrossRefGoogle Scholar
- 40.Biological Magnetic Resonance Data Bank – Metabolomics. [http://www.bmrb.wisc.edu/metabolomics]
- 41.Bundy JG, Lenz EM, Bailey NJ, Gavaghan CL, Svendsen C, Spurgeon D, Hankard PK, Osborn D, Weeks JA, Trauger SA, Spier P, Sanders I, Lindon JC, Nicholson JK, Tang H: Metabonomic assessment of toxicity of 4-fluoroaniline, 3,5-difluoroaniline and 2-fluoro-4-methylaniline to the earthworm Eisenia veneta (Rosa): identification of new endogenous biomarkers. Environ Toxicol Chem. 2002, 21: 1966-1972.CrossRefPubMedGoogle Scholar
- 48.Kenny LC, Dunn WB, Ellis DI, Myers JE, Robinson AE, Kell DB, Baker PN: Novel biomarkers for preeclampsia detected using metabolomics and machine learning. J Soc Gynecol Investig. 2005, 12: 355-Google Scholar
- 49.Holmes E, Tsang TM, Huang JTJ, Leweke FM, Koethe D, Gerth CW, Nolden BM, Gross S, Schreiber D, Nicholson JK, Bahn S: Metabolic profiling of CSF: evidence that early intervention may impact on disease progression and outcome in schizophrenia. PLoS Med. 2006, 3: e327-PubMedCentralCrossRefPubMedGoogle Scholar
- 56.Benjamini Y, Hochberg Y: Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B. 1995, 57: 289-300.Google Scholar
- 58.Dennis G, Sherman BT, Hosack DA, Yang J, Gao W, Lane HC, Lempicki RA: DAVID: Database for annotation, visualization, and integrated discovery. Genome Biol. 2003, 4 (9):Google Scholar
- 63.Spurgeon DJ, Sturzenbaum SR, Svendsen C, Hankard PK, Morgan AJ, Weeks JM, Kille P: Toxicological, cellular and gene expression responses in earthworms exposed to copper and cadmium. Comp Biochem Physiol C. 2004, 138: 11-21.Google Scholar
- 65.Marino F, Winters C, Morgan AJ: Heat shock protein (hsp60, hsp70, hsp90) expression in earthworms exposed to metal stressors in the field and laboratory. Pedobiologia. 1999, 43: 615-624.Google Scholar
- 68.Sagripanti JL: DNA damage mediated by metal ions with special reference to copper and iron. Metal Ions in Biological Systems. Edited by: Sigel A, Sigel H. 1999, New York: Marcel Dekker, 36: 179-209.Google Scholar
- 73.Hirai MY, Yano M, Goodenowe DB, Kanaya S, Kimura T, Awazuhara M, Arita M, Fujiwara T, Saito K: Integration of transcriptomics and metabolomics for understanding of global responses to nutritional stresses in Arabidopsis thaliana. Proc Natl Acad Sci USA. 2004, 101: 10205-10210.PubMedCentralCrossRefPubMedGoogle Scholar
- 77.Ankley GT, Miracle AL, Perkins EJ, Daston GP: Genomics in Regulatory Ecotoxicology: Applications and Challenges. 2007, Boca Raton, FL: CRC PressGoogle Scholar
- 81.Environmental genomics of earthworms. [http://www.earthworms.org]
- 82.Owen J, Hedley BA, Svendsen C, Wren J, Jonker MJ, Hankard PK, Lister LJ, Stürzenbaum SR, Morgan AJ, Spurgeon DJ, Blaxter ML, Kille P: Transcriptome profiling of developmental and xenobiotic responses in a keystone soil animal, the oligochaete annelid Lumbricus rubellus. BMC Genomics. 2008, 9: 226-CrossRefGoogle Scholar
- 84.NEBC file store. [http://nebc.nox.ac.uk/]
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.