Ecometabolomics: optimized NMR-based method

Authors

  • Albert Rivas-Ubach,

    Corresponding author
    1. CSIC, Global Ecology Unit CREAF-CEAB-CSIC, Barcelona, Catalonia, Spain
    • CREAF, Catalonia, Spain
    Search for more papers by this author
    • Authors contributed equally to this manuscript.
  • Miriam Pérez-Trujillo,

    1. Servei de Ressonància Magnètica Nuclear, Universitat Autònoma de Barcelona, Barcelona, Catalonia, Spain
    Search for more papers by this author
    • Authors contributed equally to this manuscript.
  • Jordi Sardans,

    1. CREAF, Catalonia, Spain
    2. CSIC, Global Ecology Unit CREAF-CEAB-CSIC, Barcelona, Catalonia, Spain
    Search for more papers by this author
  • Albert Gargallo-Garriga,

    1. CREAF, Catalonia, Spain
    2. CSIC, Global Ecology Unit CREAF-CEAB-CSIC, Barcelona, Catalonia, Spain
    3. Servei de Ressonància Magnètica Nuclear, Universitat Autònoma de Barcelona, Barcelona, Catalonia, Spain
    Search for more papers by this author
  • Teodor Parella,

    1. Servei de Ressonància Magnètica Nuclear, Universitat Autònoma de Barcelona, Barcelona, Catalonia, Spain
    Search for more papers by this author
  • Josep Peñuelas

    1. CREAF, Catalonia, Spain
    2. CSIC, Global Ecology Unit CREAF-CEAB-CSIC, Barcelona, Catalonia, Spain
    Search for more papers by this author

Correspondence author. E-mail: a.rivas@creaf.uab.cat

Summary

  1. Metabolomics is allowing great advances in biological sciences. Recently, an increasing number of ecological studies are using a metabolomic approach to answer ecological questions (ecometabolomics). Ecometabolomics is becoming a powerful tool which allows following the responses of the metabolome of an organism environmental changes and the comparison of populations. Some Nuclear Magnetic Resonance (NMR) protocols have been published for metabolomics analyses oriented to other disciplines such as biomedicine, but there is a lack of a description of a detailed protocol applied to ecological studies.
  2. Here we propose a NMR-based protocol for ecometabolomic studies that provides an unbiased overview of the metabolome of an organism, including polar and nonpolar metabolites. This protocol is aimed to facilitate the analysis of many samples, as typically required in ecological studies. In addition to NMR fingerprinting, it identifies metabolites for generating metabolic profiles applying strategies of elucidation of small molecules typically used in natural-product research, and allowing the identification of secondary and unknown metabolites. We also provide a detailed description to obtain the numerical data from the 1H-NMR spectra needed to perform the statistical analyses.
  3. We tested and optimized this protocol by using two field plant species (Erica multiflora and Quercus ilex) sampled once per season. Both species showed high levels of polar compounds such as sugars and amino acids during the spring, the growing season. E. multiflora was also experimentally submitted to drought and the NMR analyses were sensitive enough to detect some compounds related to the avoidance of water loses.
  4. This protocol has been designed for ecometabolomic studies. It identifies changes in the compositions of metabolites between individuals and detects and identifies biological markers associated with environmental changes.

Introduction

Ecometabolomics

Metabolomics as a new research approach has been widely used in biomedicine (Nicholson, Lindon & Holmes 1999; Urban et al. 2010), ecotoxicology (Robertson 2005; Alam et al. 2010), animals (Deyrup et al. 2011) and plant biology (Fiehn et al. 2000; Weckwerth et al. 2004; Scott et al. 2010), and its application is increasing in ecological studies (Peñuelas & Sardans 2009a, b; Sardans, Peñuelas & Rivas-Ubach 2011). The metabolome is the entirety of molecules present in an organism as the final expression of its genotype at a particular moment (Fiehn 2002; Peñuelas & Sardans 2009a) and can be considered a molecular picture of biological diversity because each living species has its own metabolic profile (Gromova & Roby 2010). In ecological studies, metabolomics (ecometabolomics) has become a promising tool for following the responses of the metabolome of organisms to biotic and abiotic environmental changes (Sardans, Peñuelas & Rivas-Ubach 2011). The metabolome is the chemical phenotype of organisms and is the first to respond to internal and external stressors to maintain physiological homeostasis (Fiehn et al. 2000; Sterner & Elser 2002; Peñuelas & Sardans 2009a; Leiss et al. 2011; Sardans, Peñuelas & Rivas-Ubach 2011).

An optimized NMR-based protocol for ecometabolomics. Scope of application

Several protocols to explore the metabolome in humans and animals (Beckonert et al. 2007; Martineau et al. 2012) and in microorganisms (Smart et al. 2010; Roberts et al. 2012) have recently been published. Vascular plants have received less attention, but some methods for conducting metabolomic analyses of laboratory subjects based on Nuclear Magnetic Resonance (NMR) spectroscopy have been described (Kruger, Troncoso-Ponce & Ratcliffe 2008; Leiss et al. 2009; Gromova & Roby 2010; Kim, Choi & Verpoorte 2010; Kim & Verpoorte 2010). The field of ecology currently lacks a standard protocol for analysing the metabolomes of wild plants, which are sampled from the field under very heterogeneous environmental conditions. The objective of ecology is to explain general patterns of structure, function, and evolution in ecosystems. This task is complicated by the large number of factors interacting concurrently and by the resulting high variability at all levels. Wild individuals within the same species present large differences in elemental composition, metabolism, phenology, and life style, much more so than do entries in a laboratory model. This implies that the number of individuals to be analysed should be sufficiently large for a consistent statistical analysis. A protocol is thus needed to provide an overall analysis of the main metabolites in field samples, including secondary and nonpolar compounds, and which allows the detection and identification of those metabolites that play a key role in an organism's response to environmental change. This protocol must also be reproducible and amenable to robust statistical analyses.

Among the different analytical techniques, NMR spectroscopy has the advantage of providing an unbiased overview of all the small molecules in a solution. Its analyses require minimal sample preparation, are relatively quick, non-destructive and highly reproducible and robust. 1H NMR signals are directly and linearly correlated to metabolite abundance and it is the key technique for the elucidation of unknown metabolites, allowing the identification of most of the molecules detected and the differentiation between structural isomers, diastereoisomers (Pérez-Trujillo et al. 2010; Ellis et al. 2012), and even enantiomeric molecules (Pérez-Trujillo et al. 2012).

Here, we describe an optimized NMR-based protocol for ecometabolomic studies of wild plants, especially for those plants with robust structures such as sclerophyllous leaves. The samples are not part of laboratory models as described in other protocols (Leiss et al. 2009; Gromova & Roby 2010; Kim, Choi & Verpoorte 2010) but are typically taken from the field, where variability among individuals is high and a large number of samples are usually analysed, and where, therefore, high reproducibility and robustness of method are critical. For this, special stress is also put in the automation of the NMR fingerprinting and in the use of NMR spectrometers of a high magnetic field (600 MHz or higher), which provide high resolution NMR fingerprint spectra of the cohort of samples (Leiss et al. 2009; Kim, Choi & Verpoorte 2010). Besides, the method considers polar and nonpolar metabolites and it optimizes the sample preparation methodology for their extraction (carefully described in detail in the Materials and methods section). Even so, this method may also be applied to greenhouse and laboratory studies.

The following protocol covers from sample collection and storage to the NMR spectra processing and statistical analysis (Fig. 1), going through the metabolite extraction, the acquisition of NMR fingerprint data, and the identification of the detected metabolites to obtain the metabolic profile following the structural characterization of natural products (Robinette et al. 2012) that also includes the identification of the secondary and unknown metabolites, which may play a key role in an organism's response to an environmental change.

Figure 1.

General NMR-based procedures for an ecometabolomic study.

Even though this protocol has been optimized for the analysis of plant structures, these procedures can also be applied to the study of animals such as zooplankton, insects, annelids, and molluscs, among others, because the procedures have been tested for detecting the most abundant metabolites in the current extraction, making comparative analyses possible and rigorous.

Materials and methods

This protocol was optimized to obtain robust results in ecometabolomics studies. For this reason, we tested the grinding time of samples, the amount of sample, the sonication time, repeated extractions and the recovery (See supporting information for details and the step by step procedure is also described).

The complete procedure as explained below (Fig. 2) is divided into six main categories.

Figure 2.

Experimental procedure to obtain polar and nonpolar extracts for NMR analyses.

Sample collection and storage (Steps 1–3 of Fig. 2)

Field fresh plant material was collected and rapidly packed, labelled, and frozen in situ in a container of liquid nitrogen. It prevented the degradation of metabolites (Kim & Verpoorte 2010). The physiological processes of plants vary throughout the day, so individual subjects were sampled within a short period of time and under a constant environment to reduce any effects of diurnal rhythms. The frozen plant material was lyophilized and kept frozen in plastic cans. Lyophilization (freeze-drying) avoids the hydrolysis of metabolites by maintaining enzymes in an inactive state and is thus a crucial step in ecometabolomics by keeping the metabolomes intact. Samples were ground in a ball mill (Mikrodismembrator-U, B. Braun Biotech International) at 1600 rpm for 9 min for leaves of Erica multiflora and for 6 min for leaves of Quercus ilex. Keeping the samples lyophilized and frozen and using ball mills that allow the rapid grinding of several samples independently of their nature are highly recommended procedures for ecometabolomic studies with a large number of samples, often more than 100.

Metabolite extraction and NMR sample preparation (Steps 4–20.A/16.B of Fig. 2)

For each sample, a quantity of 100 mg of dried powder was added to a 50 mL centrifuge tube. 6 mL of a water-methanol (1/1) mixture and 6 mL of chloroform were added and all samples were vortexed (15 s) and sonicated (2 min). Samples were centrifuged at 1100 × g (15 min). The two liquid phases were collected separately.

For the aqueous extraction; for each sample, 4 mL of the aqueous extract was collected by micropipette in another centrifuge tube series. Then, the steps 5–10 of Fig. 2 were repeated in the same tubes. After the two extractions, 25 mL of water was added to each tube to reduce the methanol concentration and allowing lyophilization and were kept at −80°C. Once frozen, samples were lyophilized with caps loosened. A quantity of 4 mL of water was added to each sample and were vortexed (15 s) to resuspend all the content. Samples were centrifuged at 23 000 × g (3 min) to concentrate the content at the bottom and frozen again at −80°C. Again, the samples were lyophilized with caps loosened. Once totally dried, 1 mL of phosphate buffer in D2O + 0·01% TSP (trymethyl silane propionic acid sodium salt) was added to each sample and vortexed for 5 s. All the content was resuspended with a micropipette and transferred to Eppendorf tubes. Eppendorf tubes were centrifuged at 23 000 × g (3 min) and 0·6 mL of the supernatant was transferred to NMR tubes.

Finally, the recovery of two polar metabolites (glucose and alanine) was tested for validation.

For the organic extraction; for each sample, 4 mL of the organic extraction was collected by crystal syringes in crystal jars avoiding any collection of pellet. Then, the steps 5–10 of Fig. 2 were repeated in the same tubes. The organic fractions were placed into 25 mL round-bottomed evaporation flasks and were dried in a rotary vacuum evaporator. To each flask, 1 mL of chloroform D + 0·03% TMS (tetramethylsilyl) was added and closed. After 10 min of waiting, the content was resuspended and transferred into Eppendorf tubes, centrifuged at 23 000 × g (3 min) and 0·6 mL of the supernatant was transferred to NMR tubes. The use of Eppendorf tubes for organic solvents is recommended to avoid any interaction of chloroform with plastic polymers.

Acquisition of NMR fingerprint data

The NMR fingerprint of a sample consists of a quantitative NMR spectrum of it. The spectra are obtained under specific identical, defined conditions. This ecometabolomic protocol is based on 1H NMR, but the procedures described may be helpful when using other nuclei (Palomino-Schätzlein et al. 2011). All analyses are completely automatized to guarantee a high reproducibility and robustness. For this, an autosampler allowing a load of 60 samples is used (BACS; Bruker Biospin, Rheinstetten, Germany). A NMR spectrometer of high magnetic field generating high resolution spectra (600 MHz or higher) must be used, due to the typical high variability among samples and complexity of the spectra in ecometabolomic studies. A 600 MHz NMR spectrometer is used to assure high resolution spectra (Avance 600 equipped with a triple inverse 5-mm tube Z-gradient TBI (Triple resonance Broadband Inverse) probe and with a temperature control unit, Bruker Biospin). The temperature of the equipment must be previously calibrated and maintained constant for all the experiments (at 298 K). With this purpose an equilibration delay must be left once the tube is in the magnet and prior to the acquisition (2 min).

The automation must be configured (iconnmr software application, Bruker Biospin) to allow the automatic control of processes such as insertion/ejection of the sample into the magnet, waiting delay, automatic locking onto the signal of the deuterated solvent, tuning and/or adjusting the tune frequency to the Larmor frequency of the nuclei measured, homogenization of the magnetic field, adjustment of the receiver gain of the sample, and the execution of the experiment, which includes the acquisition of the FID (Free Induction Decay), its Fourier transformation, and the preprocessing of the spectrum. The automation of the aforementioned processes improves the reproducibility of the analysis, reducing errors from human source.

Samples of the polar and nonpolar fractions are analysed and compared separately. However, within each set of samples, it is recommended to randomize them (e.g. randomize treatments, populations etc.). Polar samples are analysed using a standard quantitative 90 pulse-acquisition 1H NMR experiment with solvent suppression (Zheng & Price 2010). The standard water presaturation experiment, a conventional composite 90° 1H pulse sequence with suppression of the residual water signal (Bax 1985), or a 90° 1H NOE enhanced pulse sequence commonly termed 1D NOESY-presat (Nicholson et al. 1995) is adequate (Table 1). Nonpolar samples are analysed with a standard 90° pulse-acquisition 1H NMR experiment. Acquisition parameters and processing parameters must be set up adequately (carefully detailed in the Supporting Information). All spectra must then be visually inspected, and those that are poorly phased or baseline corrected must be corrected manually. Finally, the spectra must be calibrated using the signal of the internal calibrating reference added to the sample.

Table 1. Most common NMR experiments for identification of metabolites. Standard versions and a brief description of their application for structural elucidation problems are indicated
ExperimentVersion (pulse sequencea)Descriptionb
  1. a

    According to Bruker nomenclature.

  2. b

    Extensive and updated description of the vast library of NMR experiments and their different versions is collected in the NMR Guide and Encyclopedia of Bruker.

  3. c

    Experiments for suppression of the signal of the residual water.

1D 1H

●Conventional pulse-acquisition (zg)

●With solvent presaturationc (zgpr or zgcppr), also 1D NOESY with presaturation (noesypr1d)

Standard experiment routinely used for fingerprinting, identification of metabolites, and determining chemical shifts (d) and coupling-constant (J) values. Also used for quantification.
2D 1H-1H COSY

●Gradient selection (cosygpqf) (Aue, Bartholdi & Ernst 1976; Nagayama et al. 1980)

●With solvent presaturation (cosypgqfpr)

Homonuclear Correlation Spectroscopy based on 1H-1H scalar coupling. Routinely used for the identification of metabolites, it correlates spin systems separated through chemical bonds.
2D 1H-1H TOCSY

●Conventional (mlevph)(Braunschweiler & Ernst 1983; Bax 1985)

●With solvent presaturation (mlevphpr)

●Selective 1D mode (selmlgp.2)(Bax 1985; Kessler et al. 1986; Stonehouse et al. 1994; Stott et al. 1995)

Total Correlation Spectroscopy. Based on homonuclear 1H-1H scalar coupling. It correlates spin subsystems within the same molecule.
2D 1H-1H NOESY

●Conventional (noesygpph) (Jeener et al. 1979; Wagner & Berger 1996)

●With solvent presaturation (noesygpphpr)

●Selective 1D mode (selnogp) (Kessler et al. 1986; Stonehouse et al. 1994; Stott et al. 1995)

Nuclear Overhauser Effect Spectroscopy. Based on homonuclear 1H-1H through-space interactions. Routinely used for the identification of metabolites, it provides information about which protons are close together in space (≤ 4Å).
2D 1H-13C HSQC●Conventional using adiabatic 13C pulses (hsqcetgpsisp) (Palmer et al. 1991; Kay, Keifer & Saarinen 1992; Schleucher et al. 1994)Heteronuclear Single Quantum Correlation. Based on heteronuclear one-bond 1H-13C scalar coupling. Routinely used for the identification of metabolites, it correlates protons to their directly bonded carbon atom.
2D 1H-13C HMBC●Conventional using low-pass J-filter (hmbcgplpndqf) (Bax & Summers 1986; Bax & Marion 1988)Heteronuclear Multiple Bond Correlation. Based on heteronuclear long-range 1H-13C scalar coupling. Routinely used for the identification of metabolites, it correlates protons to carbon atoms separated by multiple (usually 2, 3) bonds.

NMR-based metabolite identification. NMR metabolic profile

The NMR metabolic profile of a sample is obtained when each peak of the NMR spectrum is assigned to its corresponding metabolite. The profile gives the NMR signals a biomolecular meaning. This analysis is usually performed on a single representative sample. The differences observed among samples are mainly due to differences in metabolite concentrations; however, qualitative differences may occur. When the assignment of a specific peak is not possible (for example, when peaks overlap or when signals are of low intensity) the analysis of another sample can be helpful. A visual inspection of all spectra of the fingerprinting can help to find a better sample for the elucidation of a specific signal.

The assignment of the 1H NMR signals is conducted following two approaches. First, by comparison of the resonance frequencies (chemical shifts, δ) and line shapes (multiplicity and coupling constants, J) of the spectrum to bibliographical data and NMR spectral databases (Table S1). Second, by the structural elucidation of the mixture (sample) through the performance of a suite of 2D NMR experiments (Table 1) and the concerted analyses of the data obtained. Basic NMR strategies followed for the structural characterization of natural products are applied for the elucidation of complex mixtures of small biological molecules (metabolites), instead of to isolated molecules as in the case of natural products research (Robinette et al. 2012). Briefly, protons connected by three to five chemical bonds are identified using 2D 1H-NMR homonuclear COSY and TOCSY correlations. 1H-1H NOESY experiments determine connections between different parts of a same molecule, and heteronuclear 1H-13C HSQC and HMBC methods identify the carbon skeleton of a molecule. This approach, the structural elucidation of the mixture by NMR spectroscopy, is particularly helpful for the identification of secondary metabolites, since only less published NMR data of these are available (Table 1). These experiments are time consuming, but they only need to be performed once and just for one sample. 1D-selective 1H experiments can be complementary to the 2D experiments, depending on the problem requiring elucidation. They are less time consuming than the 2D correlations, but retain maximum resolution and are used to get specific information of a chosen NMR signal of the 1D 1H spectrum. The spectrum is much simpler to analyse and only shows the correlation information for the selected peak. 1D-selective 1H experiments are valuable tools for elucidating and confirming problematic molecules (Ellis et al. 2012). NMR experiments are performed at the same experimental temperature used for the NMR fingerprint spectra (298 K) and using the same spectrometer (detailed in the Supporting Information). However, to use the same NMR spectrometer is not necessary, since NMR data are fully comparable independently of the spectrometer used. The use of cryoprobes will also increase the sensitivity and considerably reduce the experimental time needed.

Each specific ecometabolomic study and assignment problem will require the performance of some or all of the experiments indicated in Table 1. The version of the experiment with water-signal presaturation is recommended for polar samples. In this protocol, we describe the most common and useful NMR experiments that provide structural information, but many other NMR experiments are available in the spectrometric libraries that can be useful for the elucidation of specific problems.

Spectra processing

The NMR data from fingerprint spectra were adequately processed before conducting the statistical analyses. The bucketing process consists in obtaining the integral numeric value of the selected regions of the spectra (buckets) directly correlated with the molar concentration by their relationship to the initial concentration of the internal standard (TMS or TSP). In our protocol for ecometabolomics, we used a variable-size bucketing that is highly recommended over regular-size bucketing (Leiss et al. 2009; Gromova & Roby 2010; Kim, Choi & Verpoorte 2010) for reducing the number of variables for statistical analyses. First, a pattern for each kind of spectrum (polar and nonpolar) was created. The pattern is determined from identifying exactly where an NMR signal (peak) begins and ends for all peaks in the spectrum, and then the bucketing process can be executed based on this pattern. All empty areas (without peaks) of spectra can also be introduced into the pattern to detect any qualitative differences between samples. We used the variable-size bucketing option of AMIX (Bruker Biospin), scaling the buckets relative to the internal standard (TMS or TSP), although other programmes can be used. The output was a data set containing the integral values for each 1H-NMR spectral peak accounted for in the described pattern.

The bucketed data sets from the NMR fingerprint spectra can be analysed directly (without a previous assignment of the metabolites), because rapidly classifying samples according to their origin or their ecological or ecophysiological relevance is sufficient (Sardans, Peñuelas & Rivas-Ubach 2011) (Figs 3 and S1). This last approach does not attempt to identify all metabolites, but provides the metabolomic signature of the organism and allows detection of any shift or anomaly in its metabolism (Figs 3 and S1).

Figure 3.

PCA plots conducted from 1H NMR fingerprinting data from Quercus ilex leaves. (a) panel of loadings of PC1 and (b) panel of loadings of PC2. Loadings of the different spectral regions are represented by different colours as indicated. (c) Panel of individuals categorized by season. Arrows outside the plot indicate the mean PC score for each season. The statistically significant differences between seasons are indicated by lowercase letters (< 0·05).

Statistical analyses of metabolomic profiles can be performed by two main ways when the 1H-NMR spectra have been treated by variable-size bucketing. (i) All 1H-NMR spectral buckets can be used as individual variables. Here the result is a data set where the number of variables is equivalent to the number of buckets in the bucketing process. (ii) The buckets corresponding to the same molecular compound can be added up. The final number of variables in the data set is reduced and the statistical results are easier to interpret. For our analyses we used this method adding up the buckets corresponding to same compound.

Statistical data analysis

Principal component analysis (PCA) and PLS-DA (Partial Least Squares - Discriminant Analysis) are the most common multivariate ordination analysis (MOA) used in metabolomic studies (Ramadan et al. 2006; Ebbels & Cavill 2009; Leiss et al. 2009; Kim, Choi & Verpoorte 2010; Rivas-Ubach et al. 2012). MOA do not provide any measure of significance and it is limited to show the relation of cases with the used variables. PCs loadings of cases provided by MOA can be used to detect any significant difference of the investigated treatments by statistical inference such as t-student test or anovas (See supplementary information for more details). To get measures of statistical significance in multifactorial designs, manova and permanova (Permutational manova) are the most suitable options for metabolomic studies (Johnson et al. 2007; Anderson, Gorley & Clarke 2008). permanova is a modern statistical multivariate method used when the data for all metabolites are non-normal or when a better accommodation of random effects and interaction terms is needed (Anderson, Gorley & Clarke 2008; Rivas-Ubach et al. 2012).

In addition, O-PLS (Orthogonal PLS) and GAs (Genetic Algorithms) are other statistical methods also used in metabolomic analyses (Ramadan et al. 2006; Ebbels & Cavill 2009).

Results and discussion

Optimization of the procedure

Different steps of the metabolite extraction procedure were tested to optimize the time and obtain reliable results for the statistical analyses.

Ball mills

Most metabolomic studies have used liquid nitrogen for grinding plant samples in mortars (Leiss et al. 2009; Gromova & Roby 2010; Kim, Choi & Verpoorte 2010). In our protocol, we preferred a ball mill, for three main reasons. First, the sampled plant materials frequently resist hand-grinding in mortars with liquid nitrogen due to the large proportion of cuticles and lignin (slerophyllous leaves, needles, wood, roots, among others), making the grinding difficult and time consuming. Second, lyophilization helps to maintain the metabolomes of sampled organisms intact for a long period of time (Kim, Choi & Verpoorte 2010), and grinding in mortars is laborious once the water is removed. Third, ecological studies often require a large number of samples, and ball mills allow a more rapid grinding of several samples than do mortars and produce also a homogeneous powder, with minimal variability in particle size between samples. Liquid nitrogen and mortars, though, may also be used with soft tissues by applying some modifications to the first steps of sample processing (detailed in supplementary information).

Sonication time

After testing different times of sonication, our experiments showed that the optimum extraction of metabolites was obtained with 2 min of sonication (Step 7 of Fig. 2) (Table S2 and Figs S2 and S3). The use of chloroform during extraction will dissolve cellular membranes and thereby reduce the time of sonication, being then an interesting step when processing different batches of samples in the same day. In other NMR protocols of polar metabolites (whithout chloroform extraction), sonication for 15–20 min has been recommended (Kim, Choi & Verpoorte 2010). In addition, longer times for sonication reduce the signal strength of metabolites in 1H NMR spectra, perhaps due to the heating of samples and the formation of metabolite artifacts (t'Kindt et al. 2008).

Repeated extractions

The importance of repeated extractions has been discussed in protocols based on LC-MS analyses (t'Kindt et al. 2008), but there is a lack of metabolomic studies showing the differences between different extractions. In our protocol we tested one, two and three extractions (See supporting information) and our tests showed that two extraction procedures were the optimum since the NMR samples presented the maximum concentration of metabolites in the extracts (Table S3). A first extraction from E. multiflora and Q. ilex leaves yielded 78·5% and 85·6% metabolites, respectively, relative to a second extraction, although the differences were not statistically significant (anova test of the global concentration of metabolites; = 0·32 for E. multiflora and = 0·26 for Q. ilex). Three extractions from these species showed no differences to two extractions (anova test of the global concentration of metabolites; = 0·97 for E. multiflora and = 0·89 for Q. ilex); more than 98% of metabolites from the third extraction had already been extracted after the second extraction (See supporting information for details).

Recovery

The recovery is an important factor to take into account when comparing different groups of samples by numerical data and not all metabolomic protocols took it into account. In our protocol, it was tested for two polar metabolites: glucose and alanine. Our results showed a recovery of 92·8% for alanine and of 86·4% for glucose (Figs S4 and S5).

The analysed samples were also used to determine the reproducibility of the method. For 12 independent extractions of the same sample powder, we obtained a mean NMR signal of 0·0913 ± 0·0026 for alanine and 0·3266 ± 0·0021 for glucose (mean ± SE). These results indicate that larger 1H-NMR spectral signals provide better reproducibility. Alanine usually gives very low signals in the 1H-NMR spectra of plants because its concentration in plants is very low compared to that of sugars. This high reproducibility greatly decreases methodological errors.

Anticipated results with wild plants

The 1H NMR metabolic profiles (polar and nonpolar) of the leaves of E. Multiflora and Q. ilex are shown in Fig. 4. A typical 1H NMR metabolic profile of a polar extract from a wild plant in ecometabolomics shows the presence of primary metabolites, such as sugars, amino acids, organic acids, hydroxyacids, alcohols, and nucleic acids, as well as secondary metabolites characteristic of the particular species or family. These molecules can be completely elucidated and identified as discrete molecules (Fan 1996; Fan & Lane 2008). Nonpolar extracts contain fatty molecules (such as free fatty acids; fatty alcohols; and mono-, di-, and triglyceraldehydes) and nonfatty molecules (such as polyphenols and terpenes). Nonfatty molecules, as with polar molecules, can be completely elucidated as discrete molecules. Fatty molecules, however, are qualitatively analysed as a group, identifying and quantifying the presence of mono- and polyunsaturated fatty chains; mono-, di-, or triglyceraldehydes; free fatty alcohols; and/or free fatty acids (Gunstone 1995; Vlahov 1999; Engelke 2007; Fan & Lane 2008).

Figure 4.

Typical 1H NMR metabolic-profile spectra of polar and nonpolar extracts of E. multiflora and Q. ilex leaves. Polar metabolites: α-glucose (αG); β-glucose (βG); sucrose (Suc); alanine (Ala); asparagine (Asp); glutamine (Gln); leucine (Leu); isoleucine (Ile); threonine (Thr); valine (Val); 6-deoxypyranose (10); 4-hydroxyphenylacetate (11); malate (12); maleate (13); citrate (14); 3-amino-4-hydroxybutyrate (15); N-acetyl group (N-Ac, 16); quinic acid (Q.ac); tartaric acid (T.ac); arbutin (Arb); choline (Ch); 1,2-propanediol (21); 22, γ-hydroxybutyrate (22); lactate (23); quercitol (Quer); formate (For); catechin derivative (U1). Nonpolar metabolites (assigned signals/regions indicated with letters): A, C, D, and F, fatty acid spectral regions; B, linoleyl fatty acid region; E and L, unsaturated fatty acid regions; G, free fatty acid region; H, polyunsaturated fatty acid region; I, diacylglycerid and triacylglycerid region; J, triacylglycerid 2 region; K, triacylglycerid 1 region; M, 1,2 diacylglycerid region; N, polyphenol region; O, aldehyde group region; Ac, acetyl group; DGA, 1,2-diacylglycerid; FAI, fatty alcohols; Lin, linolenyl chain; P1, polyphenol derivative of p-coumaric acid 1; P2, polyphenol derivative of p-coumaric acid 2; Ter, terpene compound 1; TGA1, triacylglycerid 1; TGA2, triacylglycerid 2; U1, unknown compound 1; U2, unknown compound 2 (data from Erica multiflora modified from Rivas-Ubach et al. 2012).

A multivariate analysis of the foliar metabolic fingerprint of E. multiflora throughout the seasons of the year is represented in Fig. 5, The PCA resulted in a first principal component (PC1) separating the foliar metabolome in the different seasons demonstrating the sensitivity of NMR to detect seasonal metabolic shifts. The leaves of spring (the growing season in Mediterranean climate) presented the highest concentrations of polar metabolites, such as alanine, glutamine, asparagine, threonine, α-glucose, β-glucose, and sucrose. In contrast, they had the lowest concentrations of lipids and secondary metabolites, such as terpene compound one and derivatives of p-coumaric acid (the results were in more detail in Rivas-Ubach et al. 2012).

Figure 5.

Plots of the first principal component (PC1) vs. the third (PC3) resulting from PCA conducted through the metabolomic profiles of E. multiflora leaves. (a) Panel of metabolomic variables. Variables are indicated by different colours: blue, polar metabolites from primary metabolism; red, glucose and sucrose; orange, polar metabolites from secondary metabolism; black, nonpolar compounds. Different metabolic families are separated by clusters in different colours: blue, sugars; yellow, amino acids; green, compounds related to amino acids and sugar metabolism in plants (RCAAS); violet, lipids. Variable labels are described in Fig. 4 and variables from 30 to 55 represent overlapped signals (Table S4). (b) Panel of samples categorized by season. Seasons are indicated by different colours (red, summer; yellow, autumn; blue, winter; and green, spring). Arrows outside the plots indicate the mean PC score for each season. The statistically significant differences between seasons are indicated by lowercase letters (< 0·05) (Adapted from Rivas-Ubach et al. 2012).

In addition, E. multiflora plants were experimentally stressed by conditions of drought throughout the year. A PCA was performed with only those variables presenting significant differences between control and droughted plants in summer (Fig. 6), even though differences among all seasons were detected (results explained in detail in Rivas-Ubach et al. 2012). Mainly the foliar metabolomes of droughted plants presented higher concentrations of quinic acid, tartaric acid, lipids and terpenes showing that our protocol was also sensitive enough to detect shifts in the metabolomes as a response to climatic changes (Rivas-Ubach et al. 2012). The increases of these compounds in accordance with the known increase of oxidative stress in plants that have endured drought (Peñuelas et al. 2004).

Figure 6.

Plots of the PCAs conducted on the 1H NMR metabolomic variables of the Erica multiflora analyses that presented different responses to experimental climatic treatments in summer. Variable labels are described in Fig. 4 and Table S4. Treatment is indicated by colour: green, control; yellow, drought. Arrows indicate the mean PC score for each treatment. The statistically significant differences are indicated by arrows with lowercase letters (< 0·05) (Adapted from Rivas-Ubach et al. 2012).

A NMR fingerprinting data analyses were performed for Q. ilex (Fig. 3). Data were classified in the different regions (Fig. S1). The results revealed, as in the case of E. multiflora, higher concentrations of polar compounds such as sugars and amino acids in spring leaves. Also, leaves of the summer (the warmest season) presented lower concentrations of nonpolar compounds than the rest of the seasons in agreement with other experimental warming studies on plants (Livonen et al. 2004), chlorophyta (Fuschino et al. 2011), and zooplankton (Gladyshev et al. 2011).

Conclusions

Here we have presented a new optimized NMR-based protocol for application of metabolomics to field ecology. It has been specially developed to reduce the experimental errors and to be applied to a large number of samples, as often required in ecology. It thus allows performing more accurate statistical analyses. It has demonstrated to be sensitive enough to detect the differences in metabolomes of plants across different seasons and among different experimental climatic treatments. This protocol has been designed for studying the metabolome of wild plants but can also be used with animals and it is effective both for targeted and untargeted studies. It will help to increase the knowledge in the shifts of the wild organism's metabolomes across environmental gradients and it will allow making a step forwards in the understanding of the role of metabolism driving the ecosystem structure and function.

Acknowledgements

We thank C. Mulder and two anonymous referees for their constructive comments on our manuscript, Gemma Montalvan, Sara Férez and Laia Mateu for laboratory and field support. This study was founded by the Spanish Government Projects CGL2010-17172/BOS, CTQ2009-08328, Consolider-Ingenio Montes CSD2008-00040 and by the Catalan Government Project SGR 2009-458.

Ancillary