protocol

NMR-based metabolomic analysis of plants

Hye Kyong Kim, Young Hae Choi & Robert Verpoorte
Division of Pharmacognosy, Section Metabolomics, Institute of Biology, Leiden University, Einsteinweg, Leiden, The Netherlands. Correspondence should be addressed to R.V. (verpoort@chem.leidenuniv.nl). Published online 25 February 2010; doi:10.1038/nprot.2009.237

2010 Nature Publishing Group http://www.nature.com/natureprotocols

nuclear magnetic resonance (nMr)-based metabolomics has many applications in plant science. Metabolomics can be used in functional genomics and to differentiate plants from different origin, or after different treatments. In this protocol, the following steps of plant metabolomics using nMr spectroscopy are described: sample preparation (freeze drying followed by extraction by ultrasonication with 1:1 cD3oD:KH2po4 buffer in D2o), nMr analysis (standard 1H, J-resolved, 1H1H correlation spectroscopy (cosY) and heteronuclear multiple bond correlation (HMBc)) and chemometric methods. the main advantage of nMr metabolomic analysis is the possibility of identifying metabolites by comparing nMr data with references or by structure elucidation using two-dimensional nMr. this protocol is particularly suited for the analysis of secondary metabolites such as phenolic compounds (usually abundant in plants), and for primary metabolites (e.g., sugars and amino acids). this procedure is rapid; it takes not more than 30 min for sample preparation (multiple parallel) and a further 10 min for nMr spectrum acquisition.

IntroDuctIon
Metabolomic analysis in plants by NMR spectroscopy In the past decade, metabolomics has been developed as an important field of plant sciences and natural products chemistry17. The ultimate goal of metabolomics is to measure all the metabolites in an organism both qualitatively and quantitatively, which can provide a clear metabolic picture of a living organism under certain conditions. This is a very ambitious goal, as the plant metabolome is very complex. For example, about 3,000 metabolites have already been reported just from one part of a single plant, such as tobacco leaves8. At the same time, these metabolites are all different regarding their polarity, chemical behavior, stability and concentration, which make the analysis of all metabolites in one single experiment extremely difficult. In view of this, instead of aiming to analyze all individual metabolites quantitatively and qualitatively, a more realistic and suitable approach will be to have a general view of all the metabolites present in an organism under certain conditions. In this context, NMR is a very suitable method to carry out such an analysis because it allows the simultaneous detection of diverse groups of secondary metabolites (flavonoids, alkaloids, terpenoids and so on) besides abundant primary metabolites (sugars, organic acids, amino acids and so on). Moreover, in an NMR spectrum, signals are proportional to their molar concentration, making the direct comparison of concentrations of all compounds possible, without the need for calibration curves of each individual compound. In other words, it reflects the real molar levels of metabolites present in a plant. In addition, NMR is a very useful technique for structure elucidation. Using various two-dimensional NMR measurements, many signals can be identified often without the need for further fractionation of the extract. Limitations Nevertheless, the application of NMR spectroscopy in the metabolomic analysis has several limitations, the major of which is perhaps its low sensitivity. Consequently, the amounts of samples required for NMR-based analyses are larger than those required when using other analytical methods. However, sensitivity has improved considerably because of the recent developments in NMR hardware. One of the greatest advances in NMR is the development of low temperature probes, CryoProbe (Bruker Biospin
536 | VOL.5 NO.3 | 2010 | nature protocols

GmbH, Rheinstetten, Germany) or Cold Probe (Varian, Palo Alto, CA, USA). In these probes, although the sample stays at room temperature, the electronics, one of the main sources of noise in NMR spectroscopic measurements, are cooled down to 20 K ( 253 C) by bathing them in a cryogen such as liquid helium9. Doing this, it is possible to achieve upto 16-fold increases in the signal-to-noise ratio per scan. Other sensitivity related issues, such as limited sample availability, can be partly solved by using reduced detection volume NMR probes. These special probes known as microprobes, allow the analysis of a few microliters of samples, e.g., 2 l samples can be analyzed using a 1 mm microliter probe such as MicroProbe (Bruker Biospin GmbH) or NanoProbe (Varian). The reported mass sensitivity is about 20 times that of a conventional 5 mm probe10. Commercially available probes are well-documented in a recent review11. The considerable signal overlap in the NMR spectra could be mentioned as another limitation in the metabolomic analysis. Typically, 1H NMR spectra contain hundreds of overlapping signals that may hamper signal identification and accurate peak integration. This overlapping can be solved to a great extent by using two-dimensional NMR, which has a much better resolution than one-dimensional 1H NMR. This will be discussed further on in this protocol. Several other methods, which include other combinations of analytical and/or extraction methods, have been applied in metabolomic analyses1215. Each method has its advantages, unique applications and also limitations. These are summarized in Table 1. Our method is not really suitable for the analysis of non-polar metabolites such as volatiles and lipids, as the extraction is aimed at compounds of moderate to high polarity. In addition, NMR spectroscopy is not the most adequate method for the identification of individual volatiles and lipids, for which gas chromatography mass spectrometry (GCMS) undoubtedly provides better results. A good example of this was the large-scale profiling of tomato fruit volatiles reported by Tikunov et al.16 Using GCMS analysis combined with solid phase microextraction (SPME), the variation of metabolites in different genotypes of tomato was observed and more than 300 distinguishing metabolites were identified.

protocol
taBle 1 | Comparison with other metabolomic protocols available from Nature Protocols. references Lisec et al.12 analytical method Gas chromatography mass spectrometer (GCTOFMS) Reverse-phase liquid chromatography coupled to high-resolution mass spectroscopy (LCQTOFMS) Nuclear magnetic resonance (NMR) spectroscopy characteristics + Relatively long history in metabolomic study. Subsequently posses stable protocols and large database for compound identification Limited to volatiles/difficult to detect secondary metabolites + Suitable to semi-polar metabolites. Can detect plant secondary metabolites (flavonoids and glucosinolates) Restricted compound identification (isomers cannot be differentiated) + Use perchloric acid as extraction solvent. Suitable to isotopic (13C) study Restricted to primary metabolism of plants Not suitable for metabolites unstable in acids + Analysis of urine, plasma and animal tissue. Relatively long history in metabolomic analysis. Subsequently, well-developed protocols and database + Simple sample preparation step (several biofluids such as urine and cerebrospinal fluids can be directly analyzed without any preparation steps) Metabolites are limited to organic acid/amino acid/sugars and their derivatives + Simple sample preparation step. Detect many different groups of primary/secondary metabolites. Strong structure elucidation power Lower sensitivity

De Vos et al.13

2010 Nature Publishing Group http://www.nature.com/natureprotocols

Kruger et al.14

Beckonert et al.15

NMR spectroscopy

This protocol

NMR spectroscopy

+ , advantages or distinctive in the protocols; , limitations.

Applications of this protocol Nuclear magnetic resonance (NMR)based metabolomic analysis has been successfully applied in many different fields. Among the numerous applications, several interesting examples are summarized below. Those that are highlighted in bold print were carried out according to the protocol described in this paper. (1) Classification and characterization of different species of (medicinal) plants (e.g., Cannabis17, Ephedra18, Strychnos19, Ginseng20, Ilex21 and Hypericum22). For this application, NMR-based metabolomics is a very powerful tool especially in terms of the quality control of medicinal plants. At present, quality control is limited to the detection and/or quantification of certain compounds considered to be, possibly, the main active compounds. However, many medicinal plants are used as an extract or a plant powder without knowing what the active compounds are, hence the assessment of all the metabolites is necessary to guarantee reproducible pharmacological efficacy. (2) Monitoring the response to stress, wounding, herbivory or infection of plants (e.g., Catharanthus23, Brassica2426, Arabidopsis27, Senecio28 and tobacco29) and plant cell cultures (e.g., after cryopreservation30 and elicitation31). Changes induced by the infection of a plant can be monitored not only by analyzing their physiological response, but also by following changes in their metabolomic profile. In the case of the interaction with other organisms such as in the Senecioinsect interaction, it can be observed that the changes in metabolism have an important role in the resistance of plants28.

(3) Discrimination of wild/transgenic plants or different genotypes (e.g., tobacco32, Arabidopsis33, tomato34, maize35, Brassica36 and Senecio37). Unexpected metabolic changes can often be detected in transgenic plants, and NMR metabolomic analysis proved to be quite promising, allowing the detection of changes after metabolic engineering not only in the targeted pathways but in others as well. Many applications are well documented in some recent reviews38,39. Before carrying out plant metabolomic analysis, there are several steps that are required to be established and validated. To begin with, a reliable sample preparation procedure has to be developed. This step must be simple, rapid and at the same time reproducible and suitable for high throughput analysis. Metabolomic analysis generates huge datasets that are required to be handled by adequate chemometric methods. In most cases, multivariate data analysis such as principal component analysis (PCA) and partial least squares-discriminant analysis (PLS-DA) are employed routinely to extract information from very large datasets. Practical considerations A good experimental design is very important in metabolomic analysis. The level of metabolites present in plants are dependant on the developmental stages36,40, so the developmental stage of the plant to be used in experiments has to be clearly defined. Even in a given plant, metabolites in leaves of different age are not the same. Thus, all these factors should be taken into consideration for the experiments. Any metabolomic study must start with mapping
nature protocols | VOL.5 NO.3 | 2010 | 537

protocol
the biological variation, i.e., measuring different samples from one genotype, harvested at different times of the day and during different stages of development. Because of good reproducibility, these data can be used during many years as controls in various experiments, to monitor any major changes occurring in the plants over the years, or in a specific experiment. Thus, proper documentation of all samples measured (metadata) is of great importance. In the early stages of plant development (e.g., seedlings), one single plant often cannot yield sufficient amount of extract for NMR measurement, thus several plants (seedlings) have to be mixed for each group. The number of samples is also an important aspect to be considered. Considering biological variations, an appropriate number of samples have to be analyzed to extract valid information. In the case of pilot studies, such as the tracking of changes in metabolites at different time points after infection or treatments, smaller number of samples per group (~45) will be sufficient to detect the trends of plant response. If the study is aimed at discriminating several different groups of plants, more samples per group (~ > 6) will be required. As an NMR spectrum is very reproducible, in most cases, one measurement per sample is sufficient41. In dealing with experimental variation or uncertainty, the so-called technical variation or source of error is always much lower than the biological variation of plants, so biological replicates are more important than replicate analyses of the same sample. Harvesting plants Harvesting fresh plant material is a crucial step in the analysis. Undesirable chemical or enzymatic reactions of metabolites can occur during harvesting and sample preparation42. To avoid/reduce degradation of compounds, the rapid cooling of harvested samples is strongly recommended. Plant tissue can be transferred immediately after removal from the original plant into a container, which is filled with liquid nitrogen. It is always better to separate different organs in this step, such as roots, flowers and leaves. The metabolites in the vein of the leaf are quite different from those of the other parts of the leaf 32, so the possibility of separating veins from leaves should be considered. In general, one may state that higher the level of specialization of harvested tissue (cells), the better, as in principle each single cell has its own specific metabolome. Another factor to be kept in mind is that leaves of different ages do have considerable differences in their metabolome36. The time of day for harvesting should also be considered as the level of some primary metabolites (e.g., sugars, malic acid and citric acid) vary throughout the daily cycle43,44. Thus, it is recommended to prepare samples at the same time of the day in experiments that last several days, weeks or even months. Harvested samples can be kept in a freezer ( 80 C) for weeks or months before extraction. Sample preparation The sample preparation step includes the following procedures: drying, weighing and extraction. If samples are ground before drying, their handling will be easier throughout the rest of the procedure apart from increasing the efficiency of the extraction. Small-size samples or soft tissue samples can be ground with a pre-cooled pestle and mortar under liquid nitrogen, whereas a blender can be used for bigger or woody tissue samples. Drying can be carried out in many ways, but freeze-drying is the most commonly used method. The presence of water in samples increases the probability of enzymatic reactions
538 | VOL.5 NO.3 | 2010 | nature protocols

and also affects the results of NMR analysis either because of shifted resonances caused by pH differences or because of the interference/overlapping of the H2O signal. In freeze-drying, frozen water in the material sublimes under low pressure, and is thus a relatively gentle and mild drying method45. As samples are kept at low temperatures before complete drying, enzyme activity is inhibited, thus reducing enzyme-catalyzed degradation. It is important to note that even with freeze-drying, the water cannot be completely removed, as water molecules that are tightly bound to the hydroxyl groups of polysaccharides or to the carbonyl and amino groups of proteins, e.g., are more difficult to remove. However, in these conditions, microbial growth and chemical and enzymatic reactions in the materials no longer occur46. It is necessary to bear in mind that in spite of its rapidity, some metabolites could eventually de degraded or even lost during the freeze-drying process. This is the case of volatiles, e.g., but compared with other drying methods, the loss of these compounds is not a major concern47. Weighing samples exactly and precisely before the preparation of the extracts is very important. The desirable amount of plant material is between 50100 mg as dry weight, but depending on the sample type, much less can be used. The homogeneity of the material is crucial, particularly when analyzing very small amounts ( < 20 mg). Solvent choice for the extraction is a critical step. One single solvent is unlikely to extract all groups of metabolites as desired for meeting the objective of metabolomics. Perchloric acid extraction is the most suitable for polar compounds as described by Kruger et al.14, but it is not suited for acid-labile compounds. It is particularly effective for the extraction of primary metabolites such as sugars and amino acids14,48. Aqueous methanol is commonly used as an extraction solvent alone20,27,28, and/or in combination with chloroform in a two-phase system (yielding a non-polar and a polar extract)17,21,23. However, a two-phase solvent system requires an additional separating step, the evaporation of solvents and redissolving of the residue in appropriate NMR solvent. For the analysis of a large number of samples, the use of NMR solvent itself as extraction solvent produces a considerable reduction of analysis time. Another factor to be considered is the pH of the extracts, which should be controlled to avoid shifts of NMR signals. Several buffers are available for NMR measurements, the most common of which are phosphate buffers44. NMR analysis In most cases, 1H-NMR is sufficient to generate metabolomic data of a sample within a relatively short time (510 min for 64128 scans). A typical 1H-NMR spectrum of plant material contains several hundreds of signals as shown in Figures 1 and 2. It reflects the amount of all metabolites in the extract because the NMR signals are directly proportional to the molar concentration independent of the characteristic of a compound. The absolute concentration of metabolites can be calculated by comparison of the peak intensity with an internal standard. In Figure 1, 3-(trimethylsilyl) propionic-2,2,3,3-d4 acid (TSP) was used as an internal standard. TSP is a reference for the calibration of NMR shifts, and can also serve as a reference (internal standard) in the quantitation of metabolites. Depending on the solvent, different standards should be used as is described in the review by Pauli49. In a sample prepared by direct extraction with CD3OD-phosphate buffer, macromolecular compounds are not expected to be present, thus standard NMR pulse sequences provide sufficient resolution

2010 Nature Publishing Group http://www.nature.com/natureprotocols

protocol
a
Sugar region

Aromatic region

Amino acid region IS K

SN

2010 Nature Publishing Group http://www.nature.com/natureprotocols

8.0 7.5 7.0 6.5 6.0 5.5 5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 p.p.m.

b
S N N

8.4

8.2

8.0
S

7.8

7.6

7.4

7.2

7.0

6.8
COOH

6.6

6.4

6.2

6.0
N

5.8

p.p.m.

-D-Glc H N OSO3K N N H O

CH3

N OCH3 Neoglucobrassicin

H Indole-3-acetic acid OH

OCH3

Catharanthine

S I

N H HO HO

O S OH N SO3K+

1 8. 0 7. 9 7. 8 7. 7 7. 6 7. 5 7. 4 7. 3 7. 2 7. 1 7. 0 6. 9 6. 8 6. 7 6. 6 6. 5 6. 4 6. 3 6. 2

8.

8.

8.

8.

p.p.m.

H3CO

N CH3

CH3 H COOCH3 OH

Sinigrin

c
C C

Vindoline O

C
OH

HO HO OH Gluconapin O O O OH HO S N SO3K
+

V V

C V

HO H3C

O OH O O OH

Kaempferol-3,7-O--L-dirhamnoside OH OH CH3

8.4

8.2

8.0

7.8

7.6

7.4

7.2

7.0

6.8

6.6

6.4

6.2

6.0

5.8 p.p.m.

Figure 1 | Representative 1H-nuclear magnetic resonance (NMR) spectra of several plant extracts. (a) 1H-NMR spectrum of Brassica rapa leaves 14 d after methyl jasmonate treatment. (b) Expended area of a in the range of 8.5 6.0. (c) Catharanthus roseus leaves in the range of 8.5 6.0. (d) Arabidopsis thaliana (Columbia) in the range of 8.5 6.0. IS: internal standard (3-(trimethylsilyl) propionic-2,2,3,3-d4 acid (TSP)). (e) Chemical structures of metabolites, C: catharanthine, G: gluconapin, I: indole-3- acetic acid, K: kaempferol-3,7-O--L-dirhamnopyranoside, N: neoglucobrassicin, S: trans-sinapoylmalate, SN: sinigrin and V: vindoline (adapted Fig. 1a from ref. 95).

for further analysis. The presence of macromoleculespolysaccharides or proteinsin samples may cause NMR peak broadening. In this case, other NMR pulse sequences such as the MeiboomGill modification of the CarrPurcell (CPMG) spin-echo sequence method should be considered to increase spectral resolution by selectively emphasizing the signals from the low molecular weight metabolites50. To remove the undesired signal caused by residual water, suppression methods are often used. Popular water suppression techniques include weak radiofrequency irradiation (presaturation (pre-sat))51, gradient-based methods such as WATERGATE (water suppression by gradient tailored excitation)52 and its variants53, and combinations of gradient and weak radiofrequency pulse such as WET (water suppression enhanced through T1 effects)54 or its variants. Among these, pre-sat method is widely used because of its simplicity and robustness. The addition of selective irradiation of

the water resonance before data acquisition will equalize the spin populations of the water protons, avoiding their contribution to the NMR spectrum. It is the preferred method for small molecule samples. However, this technique requires a well-shimmed environment, thus when good shimming is not achieved, other techniques have to be considered55. The one-dimensional version of the two-dimensional-1H, 1H-NOESY sequence is also a widely used suppression scheme56. The pulse sequence depends upon the T1 relaxation time of the solvent being much greater than the solute. With proper phase cycling and mixing time, the broad components from the water signal can be cancelled out. This technique is often combined with pre-sat and seems to be the present method of choice for metabonomic studies56,57. Water suppression by gradient tailored excitation technique, which uses pulsed field gradients, yields better suppression of the water signal, but has the risk of suppressing signals which are very close to the water resonance53. PURGE (pre-sat utilizing relaxation
nature protocols | VOL.5 NO.3 | 2010 | 539

protocol
a
if the signals on the chemical shift axes are projected, simplified skyline projection spectra can be generated because in this projection, multiplet signals appear as singlets60. This projected spectrum can be used for further multivariate analysis29,61. Other two-dimensional techniques which can be applied for identification include correlation methods such as correlation spectroscopy (COSY), total correlation spectroscopy (TOCSY), heteronuclear multiple bond correlation (HMBC) and heteronuclear single quantum coherence spectroscopy (HSQC). Although COSY shows correlation of protons, which have mutual spinspin couplings, TOCSY shows spin systems of coupled protons in the same molecule. TOCSY is very useful for carbohydrate or amino acid identification because once typical signals are assigned (e.g., the anomeric proton in sugar), the remaining signals even those in crowded areas, can be assigned using this correlation62. In HMBC spectra, long-range correlations between carbon and protons can be assigned, which helps to identify resonances from the same molecules. An example of a two-dimensional spectrum for identification is shown in Figure 4. Recently, HSQC has been shown to be particularly useful in the metabolomics field. The presence of certain metabolites within an extract (a mixture) can be clearly distinguished by comparison of the HSQC spectra of mixtures of known reference compounds and those of the extracts63,64. Moreover, the 13C chemical shift is relatively insensitive to the change of pH and concentration of samples65. In addition, it can also be used to compare different samples by subtracting/ superimposing spectra from/with others. Figure 5 shows the signals of different terpenoids of local and systemic leaves of tobacco plant, after tobacco mosaic virus (TMV) infection. The superposition of these spectra with that of the control plants clearly showed that terpenoids (sesquiterpenes) are greatly increased in infected plants29. Many new strategies to facilitate fast two-dimensional NMR experiments (e.g., within 12 min) have been proposed; therefore, HSQC has become a feasible alternative to one-dimensional NMR for metabolomic analysis63,64. Minor compounds are not so easy to detect and to identify because the signals are weak, often appearing in crowded areas, which might even hamper their two-dimensional NMR detection. To identify all signals would be extremely time-consuming. Therefore, among all the NMR signals, one may focus on the signals that according to multivariate analysis discriminate samples. A simple fractionation step to remove major primary metabolite signals may facilitate identification. Solid phase extraction (SPE) using C18 or silica gel cartridges (1 or 3 ml) are strongly recommended for this purpose66. Even with two-dimensional NMR, the complete identification of all metabolites in a mixture is not realistic. Hyphenating NMR and high-performance liquid chromatography (HPLC) may help in further identification as it provides NMR data of individual metabolites after separation by HPLC67. At this stage, the liquid chromatographymass spectrometry (LCMS) platform can also be combined for the targeted isolation of minor metabolites. A nice example was recently published by Wolfender and co-workers68. LCMS monitored micro-isolation procedure provided low quantities of pure stress biomarkers in Arabidopsis and their structures were further identified by NMR. Direct coupling of NMR and HPLC, however, is limited to major compounds because of lack of sensitivity of the NMR. Instead HPLCSPENMR, in which the LC-peaks are concentrated on SPE cartridges and subsequently eluted with the NMR solvent is more effective69,70.

7.5 7.0 6.5

p.p.m.

p.p.m.

7.5 7.0 6.5

p.p.m.

2010 Nature Publishing Group http://www.nature.com/natureprotocols

p.p.m.

Figure 2 | 1H-nuclear magnetic resonance (NMR) spectra of two different accessions of Arabidopsis thaliana. Two Arabidopsis were grown in the same condition, harvested at the same developmental stages (4-week-old seedlings) and extracted using the same method. Six plants were combined when harvested. (a) Arabidopsis thaliana Col-0. (b) Arabidopsis thaliana C24 (unpublished data).

gradients and echoes) is a relatively new technique relying on both pre-sat and repeated refocusing cassettes. This has less effect on the neighboring resonances58. Other possible techniques are well documented in some excellent reviews56,59. Usually, in our experiments, the standard setup 1H NMR protocol with pre-sat can efficiently suppress water signals and generate spectra with good resolution. In addition, this region of the spectra can be removed in data processing before multivariate analysis, so it does not influence the results44. Metabolite identification and structure elucidation of metabolites Identification of metabolites in metabolomic analysis is the most important and at the same time the most time-consuming step. The first step for identification of metabolites is the assignment of resonances and the comparison of NMR chemical shifts and coupling constants with those of authentic samples, which are measured in the same conditions. There are about 20 common metabolites which are detected in most plant materials5, all of which are listed in the Table 2. A database of the spectra of authentic substances including common primary metabolites and secondary metabolites such as phenolic compounds (benzoic acid, salicylic acid, gallic acid, cinnamic acid derivatives and so on) measured under the same conditions as described by the protocol is the basis for the rapid identification of known metabolites. A library with one-dimensional and two-dimensional NMR spectra of more than 300 compounds is available in our laboratory, as well as more than 20,000 spectra of plant extracts from different experiments, allowing direct comparison of the data from new experiments with those of previous experiments. For the following stepthe identification of signals of unknown (new or less common) compoundsit is necessary to apply twodimensional measurements. Among the various two-dimensional NMR techniques, the 1H1H J-resolved spectrum has proved to be very useful. It provides two different types of information: a plot of the chemical shifts versus the coupling constants60 (Fig. 3). This is especially useful for the signals in crowded areas. Furthermore,
540 | VOL.5 NO.3 | 2010 | nature protocols

protocol
Pre-processing for data analysis Once NMR data have been acquired, it has to be digitalized to numeric values for further statistical analysis. In this procedure, the NMR spectrum is divided into a series of small bins (buckets). The sum of intensities of signals in each bin is calculated either by relative intensities to reference areas or to the sum of total intensities, after the removal of unwanted signals from residual solvents or water. Scaling to the sum of the total intensities helps to minimize the effect of variation between samples regarding the amount of tissue extracted71. If the buckets are made every 0.04 p.p.m., which is the most commonly used size, then around 250 variables are generated in the range of 10.0 0.5. This bucketing can be carried out by commercial programs such as AMIXTOOLS (Bruker Biospin GmbH) and ACD NMR Manager (Advanced Chemistry Development, Toronto, Ontario, Canada). Bucketing may cause a loss of spectral resolution, with the resulting difficulty in the interpretation of data (e.g., loading plot). Alternatively, it is also possible to use full-resolution NMR data for the statistical analysis as shown by Rasmussen et al.72. Full-resolution NMR dataset generates about 30,000 variables (0.15 Hz per data point), much bigger than the integrated binned NMR dataset (0.04 p.p.m. bucketing resulting in 200 variables). Consequently, it can be more informative to find significant signals in PCA analysis. In fact, in this way, a flavonoid glycoside was found to be the discriminating metabolite from different preparations of the herbal medicine St. Johns Wort72. This method seems to be particularly useful to study plants containing closely related flavonoid and flavonoid glycosides, in which only small chemical shift changes are observed for the flavonoid moieties. Otherwise, the peak alignments method can be considered to overcome the problem of shifting signals7375.
taBle 2 | Most common metabolites found in plants by nuclear magnetic resonance (NMR) analysis. Group Sugar Metabolite -Glucose -Glucose Sucrose Amino acid Alanine Glutamate Glutamine Proline Threonine Tryptophan selected characteristic signals in nMr 5.18 (d, J = 3.8 Hz) 4.58 (d, J = 7.8 Hz) 5.40 (d, J = 3.8 Hz), 4.17 (d, J = 8.5 Hz) 1.48 (d, J = 7.2 Hz ) 2.07 (m), 2.36 (m) 2.15 (m), 2.47 (m) 4.08 (dd, J = 8.6 Hz, 6.4 Hz), 2.34 (m) 1.32 (d, J = 6.6 Hz) 7.20 (t, J = 7.4 Hz), 7.29 (t, J = 7.5 Hz), 7.32 (s), 7.54 (d, J = 8.1 Hz), 7.73 (d, J = 7.9 Hz) 7.19 (t, J = 8.5 Hz), 6.86 (t, J = 7.5 Hz) 1.00 (d, J = 6.8 Hz), 1.05 (d, J = 6.8 Hz) 2.74 (d, J = 17.6 Hz), 2.56 (d, J = 17.6 Hz) 8.46 (s) 6.56 (s) 1.90 (m), 2.30 (t, J = 7.2 Hz), 3.01 (dd, J = 8.4 Hz, 6.3 Hz) 4.34 (dd, J = 6.6 Hz, 4.7 Hz), 2.74 (dd, J = 16.6 Hz, 4.7 Hz), 2.68 (dd, J = 16.6 Hz, 6.6 Hz) 2.56 (s) 8.2 (s), 8.21 (s) 3.24 (s) 4.00 (t, J = 2.8 Hz), 3.61 (t, J = 9.9 Hz), 3.44 (dd, J = 9.9 Hz, 2.9 Hz), 3.24 (t, J = 9.3 Hz)

2010 Nature Publishing Group http://www.nature.com/natureprotocols

Tyrosine Valine Organic acid Citric acid Formic acid Fumaric acid -Amino-butyrate (GABA) Malic acid

Succinic acid Others Adenine Choline Inositol

Multivariate data analysis To handle the huge dataset of the metabolomic analysis of hundreds of samples and each spectrum comprising hundreds of signals, suitable chemometric methods are required. Among multivariate data analysis methods, PCA and PLS-DA are nowadays routinely used methods. PCA is an unsupervised pattern recognition method, in which all samples are grouped with the maximum separation of all samples based on the signals in the spectra representing the metabolomes. Chemical shifts of discriminating signals involved in distinguishing the samples can be identified by so-called loading plots76,77. PLS-DA is a

supervised method that discriminates between groups of samples that are defined by the analyst76,78,79, e.g., asking the question to show the signals that give maximum separation of infected and healthy plants26. The most important information obtained from PLS-DA is the correlation between two data sets, for instance, the measured 1H NMR signals (metabolites) and the sample classification (group information). In addition, an orthogonal filter incorporated into the PLS algorithim (O-PLS) has been used for NMR data of several materials, which allows the removal of spectral components unrelated to the
nature protocols | VOL.5 NO.3 | 2010 | 541

protocol
Figure 3 | J-resolved nuclear magnetic resonance (NMR) spectra of methyl jasmonate (MJ)-treated Brassica rapa leaves. (a,b) In the region of 8.5 6.0 (a) and their chemical structures (b) 1: neoglucobrassicin, 2: hydroxycinnamates malate, 3: indole-3-acetic acid, 4: trans-sinapoylmalate, 5: cis-sinapoylmalate, 6: cis-feruloylmalate, 7: cis-coumaroylmalate, 8: trans-feruloylmalate and 9: trans-coumaroylmalate (adapted from ref. 95).

a
Hz 20 15 2 1 1 3 13 3 1 1 3 57 4 4 9 10 5 0 5 10 6 8 15 20 8.4 8.2 8.0 7.8 7.6 7.4 7.2 7.0 6.8 6.6 O 4 1 6.4 6.2 p.p.m.

sample classes chosen, i.e., filtering the markers from the biological varieties80. PCA and PLS-DA (or PLS) are primarily classification methods, exploring class differences and highlighting explanatory metabolites. Similar methods such as LDA (linear discriminant analysis)81, ICA (independent component analysis)82 and MSD (multidimensional scaling)83 have also been applied to NMR data of various materials. More information on the background and application of those methods can be found in several excellent reviews8486. Other techniques including more exploratory data mining techniques such as self organizing maps, artificial neuronal networks and other non-linear tools that have the potential to be applied to NMR analysis of plants have been reviewed87,88. Before multivariate data analysis, the raw data need to be scaled89. Among several scaling methods, mean-centering method, unit variance method and Pareto scaling methods are mostly applied. In the unit variance method, the s.d. and the scaling weight as the inverse s.d. are calculated. Subsequently, each variable is divided by the s.d.. Each scaled variable then has equal variance. It is particularly suitable to use this when the classification should also include the differences between the minor compounds71,76. However, the interpretation of the loading plots is more difficult as they are different from the original spectra, as all signals will show the same maximum intensity, independent of the absolute intensity as observed in the original spectra. In contrast, with the meancentering method, each variable resembles the original spectra, and the interpretability of loading plots improves90. However, the influence of minor signals might be overlooked. The Pareto scaling method is the intermediate compromise between the two extremes of mean-centering and unit variance scaling, and has become more common76,90. Several commercial software programs for multivariate analysis are available such as AMIX-TOOLS (Bruker Biospin GmbH), Matlab with statistics toolbox (MathWorks, Natick, MA, USA), Minitab (Minitab, State College, PA, USA), Pirouette (Infometrix, Bothell, WA, USA), SIMCA-P (Umetrics, Ume, Sweden), SPSS (SPSS, Chicago, IL, USA) or Unscrambler (CAMO software, Woodbridge, NJ, USA).

2010 Nature Publishing Group http://www.nature.com/natureprotocols

b
O R1 3 HO 4 6 5 R2 R1 = R2 = H, R1 = OCH3, R2 = H, R1 = R2 = OCH3, 2 7 1 8 O 2 3

OH OH

Coumaroyl malate Feruloyl malate Sinapoyl malate

More detailed information of theory and the background of these methods can be obtained in some excellent reviews76,79,90. Data storage One of the important points in metabolomics is the long-term data storage, as this should be the strength of metabolomics, data-mining in all results obtained over the years. Similarly to what occurs in molecular biology, in which it is relatively easy to make a database containing all gene sequences to share and store forever 91, metabolomics should also strive to construct such a database. NMR-generated data are more robust in the sense of its reproducibility when compared with other platforms such as LCMS. As it measures physical characteristics of compounds, NMR spectra will be always independent of the place or time of measurement if recorded in the same field strength. By using the two-dimensional J-resolved projected spectra, even spectra from different field strength can be directly compared. This makes NMR-based metabolomics
Figure 4 | Heteronuclear multiple bond correlation (HMBC) spectrum of aromatic moiety of phenylpropanoids of Brassica rapa leaves in the range of 6.60 7.30 of 1H and 100 150 of 13C. 1: H-2/C-6 of feruloyl malate, 2: H-2/C-7 of feruloyl malate, 3: H-2/C-3 of feruloyl malate, 4: H-2/C-4 of feruloyl malate, 5: H-6/C-2 of feruloyl malate, 6: H-2/C-6 of caffeoyl malate, 7: H-2/C-3 of caffeoyl malate, 8: H-6/C-7 of feruloyl malate, 9: H-2/C-7 of caffeoyl malate, 10: H-2/C-4 of caffeoyl malate, 11: H-6/C-4 of feruloyl malate, 12: H-2 and 6/C-2 and 6 of sinapoyl malate, 13: H-6/C-2 of caffeoyl malate, 14: H-2 and 6/C-1 of sinapoyl malate, 15: H-2 and 6/C-4 of sinapoyl malate, 16: H-6/C7 of caffeoyl malate, 17: H-6/C-4 of caffeoyl malate, 18: H-2 and 6/C-7 of sinapoyl malate, 19: H-2 and 6/C-3 and 5 of sinapoyl malate, 20: H-6/C-2 of 5-hydroxyferuloyl malate, 21: H-2/C-6 of 5-hydroxyferuloyl malate, 22: H-5/C-1 of feruloyl malate, 23: H-2/C-4 of 5-hydroxyferuloyl malate, 24: H-6/C-4 of 5-hydroxyferuloyl malate, 25: H-2/C-7 of 5-hydroxyferuloyl malate, 26: H-6/C5 of 5-hydroxyferuloyl malate, 27: H-6/C-7 of 5-hydroxyferuloyl malate, 28: H-5/C-3 of feruloyl malate, 29: H-5/C-4 of feruloyl malate and 30: H-2/C-3 of 5-hydroxyferuloyl malate (adapted from ref. 96).

p.p.m. 12 5 13 6 120 14 23 130 24 18 19 6.9 26 31 6.7 25 27 28 22 21 110

15 2 3 4 7.2 8 11 7.1 7 9 10 7.0 16 17

140

29 30

150 p.p.m.

6.8

542 | VOL.5 NO.3 | 2010 | nature protocols

protocol
Figure 5 | Heteronuclear single quantum coherence spectroscopy (HSQC) spectra of Nicotiana tabacum leaves. The spectra of the local leaves infected with tobacco mosaic virus and systemic acquired resistant leaves were superimposed with corresponding healthy leaves (lower and upper leaves). (a) Local infected leaves (red) and healthy lower leaves (blue). (b) Systemic-acquired resistant leaves (red) and healthy upper leaves (blue). Arrow indicates increased signals in systemic-required resistant leaves after tobacco mosaic virus infection.

p.p.m.

20

25

30

2010 Nature Publishing Group http://www.nature.com/natureprotocols

the method of choice for a long-term metabolomics database. Whenever new data are generated, they can be added as such without any aligning or formatting efforts. However, it is necessary to define the most suitable format for the long-term storage of data and for the efficient communication between research groups. According to the requirements of most current journals for the NMR data of compounds, all chemical shifts and coupling constant have to be shown in a table with full assignments of all signals92,93. However, this information is not very useful for metabolomics, and showing raw data (the NMR spectra itself ) of individual metabolites in the same way (same scale) as an image is more practical. One can directly compare the NMR spectra of the extract with that of compounds in the database. In addition, previously obtained data of the same organism can be used for further data mining. Obviously, this also requires that the same extraction procedure be used and that the spectra are measured in the same solvent, at the same pH and temperature, to avoid shifts of the resonances of the compounds present. This is probably the most difficult step to reach a consensus for. However, for metabolomics, a public database is crucial to eventually do data-mining using sustainable data. As mentioned before, for a public database, all steps of metabolomic analysis have to be defined and standardized. Most aspects and parameters have been discussed in the Metabolomics

35

1.7

1.6

1.5

1.4

1.3

1.2

1.1

1.0

0.9

0.8

40 p.p.m. p.p.m.

20

25

30

35

1.7

1.6

1.5

1.4

1.3

1.2

1.1

1.0

0.9

0.8

40 p.p.m.

Standards Initiative94. Important aspects in NMR-based metabolomics are NMR instruments (e.g., magnetic filed strength) and data acquisition parameters. Certainly higher magnetic field will generate data with better resolution; however, the use of 600 MHz seems to be the current trend in NMR-based metabolomic studies.

MaterIals
REAGENTS KH2PO4, ACS reagent (Riedel-de Han, cat. no. 30407) 3-(trimethylsilyl)propionic-2,2,3,3-d4 acid sodium salt (TSP), 99 atom % D (Sigma-Aldrich, cat. no. 26991-3) Deuterium oxide (D2O) > 99.9 atom % D (Spectra Stable Isotopes, cat. no. 5150) Methanold4 (CD3OD) 99.8% (Cambridge Isotope Laboratories, cat. no. DLM-24-50) Sodium deuteroxide (NaOD) 99.5% (40% in D2O, Cambridge Isotope Laboratories, cat. no. DLM-45-50) Liquid nitrogen ! cautIon It should be handled carefully and gloves and glasses should be used for protection. EQUIPMENT Freeze-dryer for sample drying (Edwards Freeze Dryer Modulyo) Freezer ( 80 C) for sample storage (RevcoScientific, BV) Ultrasonicator (Branson 5510E-MT, Branson Ultrasonics) Microcentrifuge (MC-13, Amicon) Eppendorf tubes, 2.0 ml (VWR International) pH meter (Ankersmit) with electrode (Spintrode, Hamilton) 500 MHz Bruker NMR spectrometer (DMX500, Bruker) equipped with a 5 mm TXI probe and a z-gradient system or similar instrument. AMIX (Bruker) for bucketing NMR data SIMCA-P version 12 (Umetrics AB) or comparable software for multivariate analysis. REAGENT SETUP Phosphate buffer Prepare phosphate buffer (90 mM, pH 6.0) by adding 1.232 g of KH2PO4 and 10 mg of TSP (0.01 %) to 100 ml of D2O. After stirring until total dissolution, adjust the pH using 1.0 M NaOD. 1.0 M NaOD Prepare by adding 1 ml of NaOD (40%, 10 M) to 9 ml of D2O and mix them well. EQUIPMENT SETUP NMR spectrometer An NMR spectrometer of 500 MHz NMR or higher field strength is suitable for the metabolomic analysis of plants. In general, the MeOD-phosphate buffer extract does not contain those macromolecules that could cause signal broadening in the spectra. Consequently, a 1H NMR measurement protocol with water suppression and a standard pulse sequence is used in metabolomics studies. For the identification of signals, two-dimensional measurements such as J-resolved, 1H1H COSY and HMBC are necessary. Detailed parameters are described in the PROCEDURE.

proceDure Harvesting of plant 1| Prepare liquid nitrogen in the container (e.g., Dewar barrels) (Fig. 6).

nature protocols | VOL.5 NO.3 | 2010 | 543

protocol
2| Remove the leaves from the plant and, when possible, separate the main veins from the remainder of the leaves (this might not be possible in many plants owing to their small size). Transfer to the liquid nitrogen container in tubes. pause poInt Frozen tissue can be stored at 80 C for several weeks before drying. However, degradation of metabolites might occur during storage. The comparison with fresh material will help to determine whether any degradation takes place or not. preparation of freeze-dried samples 3| Grind the frozen leaves using a pre-cooled pestle and mortar under liquid nitrogen (Fig. 6).
2010 Nature Publishing Group http://www.nature.com/natureprotocols

4| Transfer the ground leaves into a plastic tube using a spatula. 5| Keep it in the deep-freezer before freeze-drying. 6| Place samples in the freeze-dryer for 12 d. ! cautIon Do not leave samples in a freeze-dyer for longer time. Dried sample can absorb moisture easily if new non-dried samples are placed in the dryer. pause poInt Completely dried samples can be stored at room temperature for several weeks before extraction. A desiccator can be used for the storage of dried samples. sample preparation for nMr analysis 7| Weigh the freeze-dried sample in an Eppendorf tube (Fig. 6). 8| Add 0.75 ml of CH3OH-d4 and 0.75 ml of KH2PO4 buffer in D2O (pH 6.0) containing 0.1% (wt/wt) TSP. 9| Vortex for 1 min at room temperature (2025 C). 10| Ultrasonicate for 1020 min at room temperature. 11| Centrifuge at room temperature for 510 min using a microtube centrifugue (17,000*g, room temperature; *variable speed (14,00017,000g) can be used to obtain a clear supernatant. For lower-speed centrifugation, more time is required to achieve a clear supernatant). 12| Transfer supernatant (more than 1 ml) to a 1.5 ml Eppendorf tube. If a clear supernatant is not obtained, repeat centrifugation (17,000g, 1 min, room temperature). 13| Transfer 800 l of supernatant to a 5 mm NMR tube. pause poInt Extract can be kept for few days in the cold room (04 C) before NMR analysis. However, it is recommended to place at room temperature at least half an hour before NMR measurement to avoid bad shimming owing to the temperature difference in samples.
544 | VOL.5 NO.3 | 2010 | nature protocols

Leaf 2 1

Liquid N2 Harvest plant Grind leaves with pestle and mortar under liquid nitrogen

Cap Transfer frozen powder to tubes

Keep in deep freezer (80 C) or directly dry by freeze-drying

Dry sample 12 d

Weigh material and add solvent

Sonicate for 1520 min

k
NMR measurements

Centrifuge for 1020 min

Take supernatant

Transfer to NMR tube

Figure 6 | Experimental procedures for sample preparation. (a) Harvest plant. (b) Pre-cool pestle and mortar by adding liquid nitrogen first (1) and place the material in the pestle (2). Grind the materials under liquid nitrogen. (c) Transfer the frozen powder to plastic tubes. (d) Keep frozen samples at 80 C or directly dry using freeze-dryer. (e) Dry material for 12 d. (f) Weigh dry samples (50 mg) and add extraction solvents (750 l CD3OD + 750 l KH2PO4 buffer in D2O). (g) Extract using ultrasonicator for 1020 min. (h) Centrifuge at 17,000g at room temperature for 510 min to obtain clear supernatant. (i) Collect the supernatant. (j) Transfer 800 l of supernatant to an NMR tube. (k) Analyze by NMR.

protocol
nMr data acquisition* 14| Load the NMR tube into the spectrometer. 15| Set the sample temperature to 298 K (25 C) and leave a few minutes for thermal equilibration. 16| Tune and match the NMR tube. 17| Lock the spectrometer frequency to the deuterium resonance arising from the NMR solvents (either MeOD or D2O, preferably MeOD).
2010 Nature Publishing Group http://www.nature.com/natureprotocols

18| Shim the sample using either manual or automated method. 19| Determine the frequency of the water resonance and set the center of the spectrum to this frequency. *All these steps are set up in the automation system, but it is recommended to do the first sample manually to obtain good resolved spectra. 20| Select suitable experiments, the options below are for the most frequently used experiments. (a) standard 1H nMr spectroscopy (i) Set up pulse sequence comprising (relaxation delay-60-acquire), where pulse power is set to achieve a 60 flip angle, 10 kHz spectral width and water pre-sat applied during 1.5-s relaxation delay. Processing parameters: zero-fill to 64k data points, apply exponential line broadening of 0.3 Hz. After Fourier transformation, manually phase spectrum (zero and first phase), correct baseline and calibrate the spectrum by setting TSP peak at 0.00 p.p.m. (B) J-resolved spectroscopy (i) Setup the J-resolved pulse sequence, two-pulse echo sequence (relaxation delay-90 [t1/2] 180 [t1/2]-acquire) with water pre-sat during a relaxation delay of 1.5 s. Acquire FID using data matrix of 64 4,096 points covering 66 6,361 Hz, with 16 scans for each increment. Zero-fill the data to 128 4,096 and apply a sine bell-shaped window function in both dimensions before magnitude mode two-dimensional Fourier transformation. Tilt the resulting spectra along the rows by 45 relative to the frequency axis and symmetrize about the central line along F2. Manually correct baseline and calibrate to the internal standard (TSP = 0.0 p.p.m.). (c) 1H1H cosY (i) For COSY, use a phase sensitive/magnitude mode standard three pulse sequence with pre-sat during relaxation delay of 1 s. A data matrix of 512 4,096 points covering 6,361 6,361 Hz, record with 8 scans for each increment. Zero fill data to 4,096 4,096 points and apply a sine2 bell-shaped window function shifted by /2 in the F1 and /4 in the F2 dimension before States-TPPI type two-dimensional Fourier transformation. Manually phase all spectra, correct baseline and calibrate to the internal standard (TSP = 0.0 p.p.m.). (D) 1H13c HMBc (i) For HMBC spectra, use a data matrix of 254 4,096 points covering 27,164 6,361 Hz with 256 scans for each increment with a relaxation delay of 1 s. The data should be linear predicted to 512 4,096 points using 32 coefficients before magnitude type two-dimensional Fourier transformation and apply a sine bell-shaped window function shifted by /2 in the F1 dimension and /6 in the F2 dimension. Calibrate all spectra according to the internal standard (1H: TSP = 0 p.p.m. and 13C: CD3OD = 49.0 p.p.m.). Data analysis 21| Convert NMR spectra to a suitable form for further multivariate analysis. AMIX software is commonly used for converting spectra to an ASCII file. In this step, the peak is integrated into a small bin (bucket) the size of which is defined by the user. The size is preferably 0.04 p.p.m. to avoid the effect of signal fluctuation because of pH or concentration. At this stage, signals of remaining solvents have to be removed for the statistical analysis. 22| Carry out PCA using SIMCA-P software (or equivalent softwares, see multivariate data analysis section) as described in the user guides available from Umetrics homepage. 23| Identify as many of the metabolites as possible, either by comparison with NMR signals to reference compounds or by two-dimensional NMR spectra. For more information on the structure elucidation of compounds in mixture, refer to these good reviews38,39,62.

nature protocols | VOL.5 NO.3 | 2010 | 545

protocol
tIMInG
In general, sample preparation steps take more time than NMR measurements, and the latter are usually done in an automated system.
PC2 (19%)

a
14 20.00 10.00 2 3 1 7 1 33 2 1 0.5 7 1 0.5 14 14 7 14 14 14 3 27 2 0 1 2 0 7 13 33 4 2 7 7

2010 Nature Publishing Group http://www.nature.com/natureprotocols

Sample preparation Harvesting and drying steps depend on the number of samples, as once harvested, homogenization of the sample can be done separately. Freeze-drying will take 2436 h, but many samples can be handled simultaneously. In the case of extraction, parallel preparation is only limited by the amount and capacity of the equipment involved (i.e., centrifugation); in our case, 24 can be processed at the same time. Extraction itself will take 2030 min including centrifugation and transfer to NMR tubes. NMR analysis Once the NMR spectrometer is set up for the experiment, 1H-NMR measurement will take 510 min, depending on the concentration of samples. But usually, when the extract is obtained from 50100 mg of dry material, 64128 scans are enough to obtain a good spectrum. However, for the first sample, manual shimming and tuning may take up to 20 min. NMR samples can be loaded in automated systems, which allow, e.g., 24 samples for each run or five times 96 samples if a Samplejet is available (Bruker).

0 10.00 20.00

1 0 0.5 0 0 0 0 0.5 0.5 0.5

30.00

20.00

10.00

b
0.30 0.20 0.10 PC1 0

0 10.00 PC1 (32%)

20.00

30.00

Neoglucobrassicin IAA

Alanine 0.10 Sucrose 0.20 0.30 Glucose Glutamic acid

Figure 7 | Metabolites changes of Brassica rapa leaves after methyl ? trouBlesHootInG jasmonate (MJ). (a,b) Score plot of principal component analysis (PCA) of The slight shift of certain NMR signals owing to pH or concen(a) J-resolved nuclear magnetic resonance (NMR) data of Brassica rapa and tration effects can be problematic. Particularly the signals of (b) loading plot of PC1. Control plants are shown as open triangles () and fumaric acid (singlet, 6.50 6.60) and malic acid ( 2.50 MJ-treated plants are shown as solid box (). The number after the symbol shows the time (d) after MJ treatments (adapted from ref. 95). 3.00) are well known to vary considerably owing to pH and/or concentration effect. Even in a buffer, which ensures minimum pH shifts, concentration difference in samples can still cause a signal shift. This can be a problem in PCA analysis, where a signal can be recognized as a different metabolite as it bucketed in different bins in different samples. This can be overcome by making a bigger bucket, e.g., 0.1 p.p.m. instead of 0.04 p.p.m. in this area or by removing this area before data analysis60.

antIcIpateD results A typical 1H NMR spectra of a plant extract resulting from the above mentioned extraction protocol is shown in the Figures 1ad. Signals of most primary metabolites are detected in the 5.5 0.5 region; amino acids appear around 2.0 0.5, organic acids at 3.0 2.0 and sugars at 5.0 3.0 (Fig. 1a; refer also to a table 2). The aromatic region comprises many characteristic signals of secondary metabolites. Some examples are as follows: (1) Indole compounds such as indole glucosinolate (neoglucobrassicin) and indole acetic acid in Brassica (Fig. 1b), indole alkaloids such as catharanthine and vindoline in Catharanthus roseus (Fig. 1c) (2) Phenylpropanoids, flavonoids and aliphatic glucosinolates in Arabidopsis (Fig. 1d). Different plant extracts showed different profiles. Even for the same plants, considerable metabolite differences can be found depending on their environment and developmental stage. An example is shown in Figure 2. Two Arabidopsis thaliana of different accessions, grown in the same conditions, showed large difference in their metabolite profiles, not only in their concentrations of individual metabolites but even in the types of metabolites (unpublished data). Congested signals in the 1H-NMR spectra can be simplified using J-resolved spectra as shown in the Figure 3a.

p.p.m.

8
1

p.p.m.

Figure 8 | H-nuclear magnetic resonance (NMR) spectra of methyl jasmonate (MJ)-treated Brassica rapa leaves (a,b). (a) The fifth fraction containing phenylpropanoids before Sephadex LH-20 column chromatography and (b) isolated fraction by eluting with methanol. The heteronuclear multiple bond correlation (HMBC) spectrum of B is shown in Figure 4. The sugars and other metabolites are considerably removed by Sephadex LH-20 column. 546 | VOL.5 NO.3 | 2010 | nature protocols

protocol
a
PC2 (22.0%)

30 20
10 7 7 1 3 7 1 3 10 13 3 3 7 1 1 3 3 1 1 1 1 10 10

b 20
15
PC3 (12.9%)

c 30
20
1 1 1 1 1 3 3 7 3 7 3 1010 10 3 7 7 10 7

0 10 20 30

10 10 7 3 7

7 10 10 3 7 7 10

5 0 5 10 15 20

PC2 (19.3%)

10

10

10

10 0 10 20

7 1

10 10 10

3 3 1 1 1

3 10

13

35 30 25 20 15 10 5

10 15 20 25 30 35

35 30 25 20 15 10 5 0

10 15 20 25 30 35

35 30 25 20 15 10 5

10 15 20 25 30 35

PC1 (38.8%)

PC1 (39.6%)

PC1 (34.3%)

2010 Nature Publishing Group http://www.nature.com/natureprotocols

Figure 9 | Metabolites changes of Nicotiana tabacum leaves after tobacco mosaic virus (TMV) infection. (a,b) Score plot of principal component analysis of (a) healthy (lower and upper) and (b) infected (local-infected and systemic acquired resistant) Nicotiana tabacum leaves by tobacco mosaic virus, and (c) loading plot (PC2) of infected leaves. , Lower leaves of healthy plants; , upper leaves of healthy plants; , local-infected leaves of TMV-infected plants; , systemic-acquired resistance leaves of TMV-infected plants. (a) Principal component analysis (PCA) for lower and upper leaves in healthy plants and inoculated and systemic acquired resistant leaves, (b) PCA for lower leaves in healthy plants and inoculated leaves in the infected plants, (c) PCA for upper leaves in healthy plants and systemic-acquired resistant leaves in the infected plants. Numbering in plot A and B are the dates after infection. The number of 1 and 4 on plot C are nicotine and 5-caffeoyl quinic acid signals, respectively (adapted from ref. 29).

By making a projection on the spectral axis of these two-dimensional spectra, multiplet signals become singlets, giving much clearer and sharper signals and thus considerably facilitating peak identification (compare Fig. 1b and Fig. 3a). In one of our studies on Brassica, a metabolomic approach was carried out to obtain a holistic picture of metabolic changes in Brassica after methyl jasmonate (MJ) treatment95. Samples were collected at different time points after MJ treatment and their NMR signals were compared with corresponding control samples. PCA results showed that MJ treatment produced metabolite changes after 2 d and further changes continued for the 14 d of experiment. Control groups showed different metabolite profiles to MJ-treated groups (Fig. 7). In MJ-treated Brassica, several phenolic compounds were observed to increase if compared with control groups. Considerable signal overlapping hampered the identification of individual compounds in the crude extract. To isolate increased metabolites, the crude extract was Inoculated and SAR leaves submitted to further column chromaNicotine tography. The first separation on HP20 yielded five fractions, the last of which containing phenolic compounds ? Inositol Dehydroascorbic acid Ascorbic acid Putrescine Nicotinic acid (Fig. 8a) was further subjected to Inoculated leaves Sephadex LH-20 column chromatograOrnithine Tryptophan phy. In the isolated fraction, ten difGlucose Glucose-6-P Steroids Triterpenoids ferent malate-conjugated phenylpropaSucrose noids were detected96 (Fig. 8b). Their Fructose GGPP Cembranoids structures were identified using their Inoculated and SAR leaves Rishitin (?) two-dimensional NMR spectra, and the Capsidiol FPP DMAPP ()-Jasmonic acid HMBC spectrum is shown in Figure 4. Lubumin Hydrosylubumin GPP Another interesting example was the (+)-7-Isojasmonic acid SAR leaves determination of metabolites at differIPP ent time points in tobacco leaves (local Mevalonic acid -Linolenic acid leaves) after infection with TMV and sysCinnamic acid Phenylalanine Proline Shikimic acid D-Erythrose-4-P temic leaves. Leaves above the infected Naringenin leaves showed an increased resistance Pyruvic acid Malonic acid Glutamine 3-O-caffeoyl quinic acid ? against TMV virus after the infection Kaempferol Quercetin Alanine ? Acetyl-CoA (systemic acquired resistance (SAR)). 4-O-caffeoyl quinic acid Inoculated and SAR leaves SAR leaves ? Moreover, different metabolic changes 5-O-caffeoyl quinic acid Citric acid were observed in local and systemic Inoculated and SAR leaves leaves after leaf infection (Fig. 9)29. Thus, Isocitric acid by measuring many metabolites simultaMalic acid -Ketoglutaric acid neously in one analysis, both quantitatively and qualitatively, the metabolomic Inoculated and SAR leaves approach provides a snapshot of the plant Figure 10 | Proposed metabolomic alterations in the Nicotiana tabacum leaves infected by tobacco metabolism after infection. The metabolic mosaic virus. Blue ( ): decreased; red ( ): increased; gray ( ): transient increased; purple (): previous changes are summarized in the biosynresults from other Nicotiana species (e.g. Nicotiana undulata or Nicotiana rustica and Nicotiana thetic pathways as shown in Figure 10. glutinosa); and black ( ): based on general plant biosynthesis (adapted from ref. 29).
nature protocols | VOL.5 NO.3 | 2010 | 547

protocol
acKnowleDGMents We thank Ms. E.G. Wilson for reviewing the manuscript and providing helpful comments. We also thank Dr. A. Meissner, Mr. C. Erkelens and Mr. A.W.M. Lefeber for their kind help in setting up NMR parameters. This research has received funding from the European Communitys Seventh Framework Programme [FP7/2007-2013] under Grant Agreement No 217895. autHor contrIButIons All authors discussed all the steps of the protocol, its implications and applications. H.K.K. wrote the manuscript, and Y.H.C. and R.V. revised it. Published online at http://www.natureprotocols.com/. Reprints and permissions information is available online at http://npg.nature. com/reprintsandpermissions/. 1. Fiehn, O. et al. Metabolic profiling for plant functional genomics. Nat. Biotechnol. 18, 11571161 (2000). Sumner, L.W., Mendes, P. & Dixon, R.A. Plant metabolomics: large-scale phytochemistry in the functional genomics era. Phytochemistry 62, 817836 (2003). Rochfort, S. Metabolomics reviewed: A new omic platform technology for systems biology and implications for natural products research. J. Nat. Prod. 68, 18131820 (2005). Hall, R.D. Plant metabolomics: from holistic hope, to hype, to hot topic. New Phytol. 169, 453468 (2006). Verpoorte, R., Choi, Y.H. & Kim, H.K. NMR-based metabolomics at work in phytochemistry. Phytochem. Rev. 6, 314 (2007). Ward, J.L., Baker, J.M. & Beale, M.H. Recent applications of NMR spectroscopy in plant metabolomics. FEBS J. 274, 1261131 (2007). Seger, C. & Sturm, S. Analytical aspects of plant metabolic profiling platforms: current standings and future aims. J. Proteome Res. 6, 480497 (2007). Wahlberg, I. & Enzell, C.R. Tobacco isoprenoids. Nat. Prod. Rep. 4, 237276 (1987). Kovacs, H., Moskau, D. & Spraul, M. Cryogenically cooled probes-a leap in NMR technology. Prog. Nucl. Magn. Reson. Spectrosc. 46, 131155 (2005). Grivet, J.-P. & Delort, A.-M. NMR for microbiology: in vivo and in situ applications. Prog. Nucl. Magn. Reson. Spectrosc. 54, 153 (2009). Mukhopadhyay, R. Liquid NMR probes: oh so many choices. Anal. Chem. 79597964 (2007). Lisec, J., Schauer, N., Kopka, J., Willmitzer, L. & Fernie, A.R. Gas chromatography mass spectrometry-based metabolite profiling in plants. Nat. Protoc. 1, 387396 (2006). De Vos, R.C.H. et al. Untargeted large-scale plant metabolomics using liquid chromatography coupled to mass spectrometry. Nat. Protoc. 2, 778791 (2007). Kruger, N.J., Troncoso-Ponce, A.T. & Ratcliffe, R.G. 1H NMR metabolite fingerprinting and metabolomic analysis of perchloric acid extracts from plant tissues. Nat. Protoc. 3, 10011012 (2008). Beckonert, O. et al. Metabolic profiling, metabolomic and metabonomic procedures for NMR spectroscopy of urine, plasma, serum and tissue extracts. Nat. Protoc. 2, 26922703 (2007). Tikunov, Y. et al. A novel approach for non-targeted data analysis for metabolomics. Large-scale profiling of tomato fruit volatiles. Plant Physiol. 139, 11251137 (2005). Choi, Y.H. et al. Metabolomic differentiation of Cannabis sativa cultivars using 1H NMR spectroscopy and principal component analysis. J. Nat. Prod. 67, 953957 (2004). Kim, H.K. et al. Metabolic fingerprinting of Ephedra species using 1H-NMR spectroscopy and principal component analysis. Chem. Pharm. Bull. 53, 105109 (2005). Frdrich, M. et al. Metabolomic analysis of Strychnos nux-vomica, icaja and ignatii extracts by 1H nuclear magnetic resonance spectrometry and multivariate analysis techniques. Phytochemistry 65, 19932001 (2004). Yang, S.Y. et al. Application of two dimensional nuclear magnetic resonance spectroscopy to quality control of ginseng commercial products. Planta Med. 72, 364369 (2006). Choi, Y.H. et al. Classification of Ilex species based on metabolomic fingerprinting using NMR and multivariate data analysis. J. Agric. Food Chem. 53, 12371245 (2005). Roos, G., Rseler, C., Berger-Bter, K. & Simmen, U. Classification and correction of St. Johns wort extracts by nuclear magnetic resonance spectroscopy, multivariate data analysis and pharmacological activity. Planta Med. 70, 771777 (2004). Choi, Y.H. et al. Metabolic discrimination of Catharanthus roseus leaves infected by phytoplasma using 1H-NMR spectroscopy and multivariate data analysis. Plant Physiol. 135, 23982410 (2004). 24. Widarto, H.T. et al. Metabolomic differentiation of Brassica rapa leaves attacked by herbivore using two dimensional nuclear magnetic resonance spectroscopy. J. Chem. Ecol. 32, 24172428 (2006). 25. Jahangir, M., Kim, H.K., Choi, Y.H. & Verpoorte, R. Metabolomic response of Brassica rapa submitted to pre-harvest bacterial contamination. Food Chem. 107, 362368 (2008). 26. Simoh, S. et al. Metabolic changes in Agrobacterium tumefaciensinfected Brassica rapa. J. Plant Physiol. 166, 10051014 (2009). 27. Hendrawati, O. et al. Metabolic differentiation of Arabidopsis treated with methyl jasmonate using nuclear magnetic resonance spectroscopy. Plant Sci. 170, 11181124 (2006). 28. Leiss, K.A. et al. NMR Metabolomics of thrips (Frankliniella occidentalis) resistance in senecio hybrids. J. Chem. Ecol. 35, 219229 (2009). 29. Choi, Y.H. et al. NMR metabolomics to revisit the tobacco mosaic virus infection in Nicotiana tabacum leaves. J. Nat. Prod. 69, 742748 (2006). 30. Suhartono, L. et al. Metabolic comparison of cryopreserved and normal cells from Tabernaemontana divaricata suspension cultures. Plant Cell Tissue Organ Cult. 83, 5966 (2005). 31. Snchez-Sampedro, A. et al. Metabolomic alterations in elicitor treated Silybum marianum suspension cultures monitored by nuclear magnetic resonance spectroscopy. J. Biotechnol. 130, 133142 (2007). 32. Choi, H.K. et al. Metabolic fingerprinting of wild type and transgenic tobacco plants by 1H NMR and multivariate analysis technique. Phytochemistry 65, 857864 (2004). 33. Le Gall, G. et al. Metabolite profiling of Arabidopsis thaliana (L.) plants transformed with an antisense chalcone synthase gene. Metabolomics 1, 181198 (2005). 34. Le Gall, G. et al. Metabolite profiling of tomato (Lycopericon esculentum) using 1H NMR spectroscopy as a tool to detect potential unintended effects following a genetic modification. J. Agric. Food Chem. 51, 24472456 (2003). 35. Manetti, C. et al. NMR-based metabonomic study of transgenic maize. Phytochemistry 65, 31873198 (2004). 36. Abdel-Farid, I.B., Kim, H.K., Choi, Y.H. & Verpoorte, R. Metabolic characterization of Brassica rapa leaves by NMR spectroscopy. J. Agric. Food Chem. 55, 79367943 (2007). 37. Kirk, H. et al. Comparing metabolomes: the chemical consequences of hybridization in plants. New Phytol. 167, 613622 (2005). 38. Holmes, E., Tang, H., Wang, Y. & Seger, C. The assessment of plant metabolite profiles by NMR-based methodologies. Planta Med. 72, 771785 (2007). 39. Van der Kooy, F. et al. Quality control of herbal material and phytopharmaceuticals with the use of MS and NMR based metabolic fingerprinting. Planta Med. 75, 763775 (2009). 40. Keurentjes, J.J.B. et al. The genetics of plant metabolism. Nat. Genet. 38, 842849 (2006). 41. Keun, H.C. et al. Analytical reproducibility in 1H NMR-based metabonomic urinalysis. Chem. Res. Toxicol. 15, 13801386 (2002). 42. Maltese, F., van der Kooy, F. & Verpoorte, R. Solvent derived artifacts in natural products chemistry. Nat. Prod. Commun. 4, 447454 (2009). 43. Queiroz, O. Circadian-rhythms and metabolic patterns. Annu. Rev. Plant Physiol. Plant Mol. Biol. 25, 115134 (1974). 44. Kim, H.K., Choi, Y.H. & Verpoorte, R. Metabolomic analysis of Catharanthus roseus using NMR and principal component analysis. In Biotechnology in Agriculture and Forestry 57 Plant Metabolomics. (eds. Saito, K., Dixon, R.A. & Willmitzer, L.) 261276 (Springer, Leipzig, Germany, 2006). 45. Verpoorte, R., Choi, Y.H., Mustafa, N.R. & Kim, H.K. Metabolomics: back to basics. Phytochem. Rev. 7, 525537 (2008). 46. Maltini, E., Torreggiani, D., Venir, E. & Bertolo, G. Water activity and the preservation of plant foods. Food Chem. 82, 7986 (2003). 47. Venskutonis, P.R. Effect of drying on the volatile constituents of thyme (Thymus vulgaris L.) and sage (Salvia oficinalis L.). Food Chem. 59, 219227 (1997). 48. Kruger, N.J. et al. Network flux analysis: impact of 13C-substrates on metabolism in Arabidopsis thaliana cell suspension cultures. Phytochemistry 68, 21762188 (2007). 49. Pauli, G.F., Jaki, B.U. & Lankin, D.C. Quantitative 1H NMR: Development and potential of a method for natural products analysis. J. Nat. Prod. 68, 133149 (2005). 50. Phalaraksh, C. et al. NMR spectroscopic studies on the hemolymph of the tobacco hornworm, Manduca sexta: assignment of 1H and 13C NMR spectra. Insect Biochem. Mol. Biol. 29, 795805 (1999). 51. Hoult, D.I. Solvent peak saturation with single phase and quadrature Fourier transformation. J. Magn. Reson. 21, 337347 (1976).

2010 Nature Publishing Group http://www.nature.com/natureprotocols

2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22.

23.

548 | VOL.5 NO.3 | 2010 | nature protocols

protocol
52. Sklenar, V., Piotto, M., Leppik, R. & Saudek, V. Gradient-tailored water suppression for 1H-15N HSQC experiments optimized to retain full sensitivity. J. Magn. Reson. A 102, 241245 (1993). 53. Liu, M. et al. Improved WATERGATE pulse sequences for solvent suppression in NMR spectroscopy. J. Magn. Reson. 132, 125129 (1998). 54. Ogg, R.J., Kingsley, P.B. & Taylor, J.S. WET, a T1- and B1-insensitive watersuppression method for in vivo localized 1H NMR spectroscopy. J. Magn. Reson. B 104, 110 (1994). 55. Mo, H. & Raftery, D. Pre-SAT180, a simple and effective method for residual water suppression. J. Magn. Reson. 190, 16 (2008). 56. McKay, R.T. Recent advances in solvent suppression for solution NMR: a practical reference. Annu. Rep. NMR Spectr. 66, 3376 (2009). 57. Reily, M.D. & Lindon, J.C. NMR spectroscopy: principles and instrumentation. In Metabonomics in Toxicity Assessment. (eds. Robertson, D.G., Lindon, J., Nicholson, J.K. & Holmes, E.) 75104 (CRC Press, Boca Raton, USA, 2005). 58. Simpson, A.J. & Brown, S.A. Purge NMR: effective and easy solvent suppression. J. Magn. Reson. 175, 340346 (2005). 59. Price, W.S. Water signal suppression in NMR spectroscopy. Annu. Rep. NMR Spectr. 38, 289354 (1999). 60. Viant, M.R. Improved methods for the acquisition and interpretation of NMR metabolomic data. Biochem. Biophys. Res. Commun. 310, 943948 (2003). 61. Tiziani, S. et al. Effects of the application of different window functions and projection methods on processing of 1H J-resolved nuclear magnetic resonance spectra for metabolomics. Anal. Chim. Acta 610, 8088 (2008). 62. Fan, T.W.M. Metabolite profiling by one- and two-dimensional NMR analysis of complex mixtures. Prog. Nucl. Magn. Reson. Spectrosc. 28, 161219 (1996). 63. Xi, Y., De Ropp, J.S., Viant, M.R., Woodruff, D.L. & Yu, P. Improved identification of metabolites in complex mixtures using HSQC NMR spectroscopy. Anal. Chim. Acta 614, 127133 (2008). 64. Lewis, I.A. et al. Method for determining molar concentrations of metabolites in complex solutions from two dimensional 1H-13C NMR spectra. Anal. Chem. 79, 93859390 (2007). 65. Xia, J., Bjordahl, T.C., Tang, P. & Wishart, D.S. MetaboMiner-semiautomated identification of metabolites from 2D NMR spectra of complex biofluids. BMC Bioinformatics 9, 507 (2008). 66. Fraccaroli, M. et al. Pre-analytical method for metabolic profiling of plant cell cultures of Passiflora garckei. Biotechnol. Lett. 30, 20312036 (2008). 67. Bobzin, S.C., Yang, S. & Kasten, T.P. Application of liquid chromatographynuclear magnetic resonance spectroscopy to the identification of natural products. J. Chromatogr. B 748, 259267 (2000). 68. Glauser, G. et al. Optimized liquid chromatography-mass spectrometry approach for the isolation of minor stress biomarkers in plant extracts and their identification by capillary nuclear magnetic resonance. J. Chromatogr. A 1180, 9098 (2008). 69. Lambert, M. et al. Identification of natural products using HPLC-SPE combined with CapNMR. Anal. Chem. 79, 727735 (2007). 70. Jaroszewski, J.W. Hyphenated NMR methods in natural products research, part 2: HPLC-SPE-NMR and other new trends in NMR hyphenation. Planta Med. 71, 795802 (2005). 71. Craig, A. et al. Scaling and normalization effects in NMR spectroscopic metabonomic data sets. Anal. Chem. 78, 22622267 (2006). 72. Rasmussen, B., Cloarec, O., Tang, H., Staerk, D. & Jaroszewski, J. Multivariate analysis of integrated and full-resolution 1H-NMR spectral data from complex pharmaceutical preparations: St. Johns wort. Planta Med. 72, 556563 (2006). 73. Forshed, J., Schuppe-Koistinen, I. & Jacobsson, S.P. Peak alignment of NMR signals by means of a genetic algorithm. Anal. Chim. Acta 487, 189199 (2003). 74. Forshed, J. et al. A comparison of methods for alignment of NMR peaks in the context of cluster analysis. J. Pharm. Biomed. Anal. 38, 824832 (2005). 75. Lee, G.-C. & Woodruff, D.L. Beam search for peak alignment of NMR signals. Anal. Chim. Acta. 513, 413416 (2004). 76. Eriksson, L., Johansson, E., Kettaneh-Wold, N. & Wold, S. Multi- and Megavariate Data Analysis. Principles and Applications. (Umetrics AB, Ume, Sweden, 2001). 77. Trygg, J. & Lundstedt, T. Chemometrics techniques for metabonomics. In The Handbook of Metabonomics and Metabolomics (eds. Lindon, J.C., Nicholson, J.K. & Holmes, E.) 171200 (Elsevier, Amsterdam, The Netherlands, 2007). 78. Kemsley, E.K. Discriminant Analysis and Class Modeling of Spectroscopic Data. (John Wiley & Sons, Chichester, UK, 1998). 79. Holmes, E. et al. Chemometric models for toxicity classification based on NMR spectra of biofluids. Chem. Res. Toxicol. 13, 471478 (2000). 80. Trygg, J. & Wold, S. O2-PLS, a two-block (X-Y) latent variable regression (LVR) method with an integrated OSC filter. J. Chemom. 17, 5364 (2003). 81. Rezzi, S. et al. Classification of olive oils using high throughput flow 1H NMR fingerprinting with principal component analysis, linear discriminant analysis and probabilistic neural networks. Anal. Chim. Acta. 552, 1324 (2005). 82. Cuny, M. et al. Fruit juice authentication by 1H NMR spectroscopy in combination with different chemometrics tools. Anal. Bioanal. Chem. 390, 419427 (2008). 83. Pierens, G.K. et al. A robust clustering approach for NMR spectra of natural product extracts. Magn. Reson. Chem. 43, 359365 (2005). 84. Lindon, J.C., Holmes, E. & Nicholson, J.K. Pattern recognition methods and applications in biomedical magnetic resonance. Prog. Nucl. Magn. Reson. Spectrosc. 39, 140 (2001). 85. Berrueta, L.A., Alonso-Scales, R.M. & Hberger, K. Supervised pattern recognition in food analysis. J Chromatogr. A 1158, 196214 (2007). 86. Ebbels, T.M.D. & Cavill, R. Bioinformatic methods in NMR-based metabolic profiling. Prog. Nucl. Magn. Reson. Spectros. 55, 361374 (2009). 87. Yin, H. Nonlinear dimensionality reduction and data visualization: a review. Int. J. Autom. Comput. 4, 294303 (2007). 88. Steuer, R., Morgenthal, K., Weckwerth, W. & Selbig, J. A gentle guide to the analysis of metabolomic data. In Metabolomics-Methods and Protocols (ed. Weckwerth, W.) 105126 (Human Press, Totowa, New Jersey, USA, 2007). 89. van den Berg, R.A. et al. Centering, scaling, and transformations: improving the biological information content of metabolomics data. BMC Genomics 7, 142 (2006). 90. Colquhoun, I.J. Use of NMR for metabolic profiling in plant systems. J. Pestic. Sci. 32, 200212 (2007). 91. Benson, D.A. et al. GenBank. Nucleic Acids Res. 37, D26D31 (2009). 92. Journal of Natural Products, guide for authors, pubs.acs.org/userimages/ ContentEditor/1218551109887/jnprdf_authguide.pdf. 93. Phytochemistry guide for authors www.elsevier.com/wps/find/ journaldescription.cws_home/273/authorinstructions. 94. Sumner, L.W. et al. Proposed mimimum reporting standards for chemical analysis. Metabolomics 3, 211221 (2007). 95. Liang, Y.-S. et al. Metabolomic analysis of methyl jasmonate treated Brassica rapa leaves by two dimensional NMR spectroscopy and multivariate analysis. Phytochemistry 67, 25032511 (2006). 96. Liang, Y.-S. et al. Identification of phenylpropanoids in methyl jasmonate treated Brassica rapa leaves using two dimensional nuclear magnetic resonance spectroscopy. J. Chromatogr. A 1112, 148155 (2006).

2010 Nature Publishing Group http://www.nature.com/natureprotocols

nature protocols | VOL.5 NO.3 | 2010 | 549

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.