Metabolomic approaches toward understanding nitrogen metabolism in plants

Abstract

Plants can assimilate inorganic nitrogen (N) sources to organic N such as amino acids. N is the most important of the mineral nutrients required by plants and its metabolism is tightly coordinated with carbon (C) metabolism in the fundamental processes that permit plant growth. Increased understanding of N regulation may provide important insights for plant growth and improvement of quality of crops and vegetables because N as well as C metabolism are fundamental components of plant life. Metabolomics is a global biochemical approach useful to study N metabolism because metabolites not only reflect the ultimate phenotypes (traits), but can mediate transcript levels as well as protein levels directly and/or indirectly under different N conditions. This review outlines analytical and bioinformatic techniques particularly used to perform metabolomics for studying N metabolism in higher plants. Examples are used to illustrate the application of metabolomic techniques to the model plants Arabidopsis and rice, as well as other crops and vegetables.

Arabidopsis, GC-MS, maize, metabolism, metabolite profiling, metabolomics, nitrogen, rice, tomato

Introduction

Nitrogen (N) is the most important mineral nutrient for plants. Since plants require the greatest quantity of N of all the mineral elements, deficiency of N is a limiting factor for plant growth (Coruzzi and Bush, 2001; Coruzzi, 2003; Miller et al., 2007; Schachtman and Shin, 2007; Krouk et al., 2010). Furthermore, carbon (C) and N metabolism are tightly coordinated in the fundamental processes that permit plant growth, for example photosynthesis and N uptake (Coruzzi and Bush, 2001; Thum et al., 2003; Urbanczyk-Wochniak and Fernie, 2005; Gutierrez et al., 2007). N is required for the synthesis of amino acids, proteins, chlorophyll, nucleic acids, lipids, and a variety of other metabolites containing N in their structure (Fig. 1). As the first step, plants assimilate nitrate or ammonium as the primary source of N (Lam et al., 1996; Stitt et al., 2002; Yamaya and Oaks, 2004; Tabuchi et al., 2007). Subsequently, assimilated N is utilized to produce amino acids, while carbon dioxide is fixed to synthesize sugars (Coruzzi and Zhou, 2001; Stitt et al., 2010). Complex interactions via biochemical networks of metabolite pathways exist in many aspects of N metabolism. N metabolism involves N uptake and regulation, N reduction and signalling, amino acid metabolism and transport, interactions between N and C metabolism, and the translocation and remobilization of N (Scheible et al., 2004; Diaz et al., 2008; Miller et al., 2008; Vidal and Gutierrez, 2008; Krouk et al., 2010). Primary metabolites (e.g. amino acids, sugars, sugar phosphates, and organic acids) as well as secondary metabolites (e.g. phenylpropanoids) are components of the complex networks of biochemical pathways in plant metabolism (Fig. 1). Metabolomics can be used to capture a snapshot of metabolic status in a cell in an untargeted manner (Weckwerth, 2003; Saito and Matsuda, 2010). Because N is the most crucial element for plant growth and metabolism, measuring metabolite levels using metabolomic techniques provides basic information about biological responses to physiological or environmental changes triggered by N status (Scheible et al., 1997; Tschoep et al., 2009).

Fig. 1.

Metabolite production through N and C metabolism in plants. Plants assimilate inorganic C and N to amino acids and sugars via N assimilation processes and photosynthesis, respectively. Green ellipses represent N-containing metabolites, while red ellipses indicate photosynthetic metabolites. Since N and C metabolism is tightly coordinated in metabolic pathways, levels of metabolites belonging to the metabolic pathway should be affected by N and C status. However, because there are many reversible and complex reactions, particularly in central metabolism, it is often difficult to observe major changes in metabolite levels.

Metabolites are not only catalytic products of enzymatic reactions but are also active regulators and the ultimate phenotypic representatives of homeostasis in highly complex biochemical networks. Metabolism allows optimized trade-offs among many requirements, such as reaction complexity, energy required for each reaction, and, particularly, adaptability via tolerance to various perturbations and evolvability in metabolite networks. These regulatory mechanisms can be observed as metabolite–metabolite correlations (Steuer et al., 2003; Camacho et al., 2005; Steuer, 2006). Metabolite–metabolite correlations can be calculated using a data matrix generated by metabolomic techniques such as mass spectrometry (MS). These correlations also provide a global snapshot of the biochemical mechanisms operating in plants and animals (Weckwerth et al., 2004; Noguchi et al., 2006; Kusano et al., 2007; Sato et al., 2008).

As with other ‘omics’ approaches, metabolomic data sets are too complex to curate manually. Therefore, it has been necessary to develop novel bioinformatic methods optimized for metabolomics, which involve data pre-treatment programs (Jonsson et al., 2006; Katajamaa et al., 2006; Smith et al., 2006; Izquierdo-Garcia et al., 2009; Lommen, 2009; Veselkov et al., 2009), statistical analysis, correlation network analysis, modelling, and metabolomic databases (Markley et al., 2007; Kind et al., 2009; Tohge and Fernie, 2009). Of these, statistical analysis (Bylesjo et al., 2009; Fukushima et al., 2009,b; Tolstikov, 2009), correlation networks, and modelling approaches (Rios-Estepa and Lange, 2007, 2008; Nagele et al., 2010; Ruffel et al., 2010) are useful for biological interpretation of metabolomic data sets when insights into the metabolomic status in N metabolism are sought.

In this review article, metabolomic techniques, particularly those applied for the study of N metabolism, are introduced first. As one of the strategies to increase the coverage of metabolomic detections, a novel method of integration of multiplatforms is also introduced. Bioinformatic approaches involving statistical analysis, correlation networks, and modelling to obtain biological information from metabolomics data sets are described. Finally, recent metabolomic investigations of N assimilation, and N and C metabolism, in model plants (Arabidopsis and rice) and crops are summarized.

Metabolomic analytical platforms for the study of N metabolism

Metabolite profiling as well as targeted metabolite analysis have been conducted to assess the metabolomic status in order to obtain insights into N metabolism involving a variety of metabolic events in plants (Table 1). When obtaining and analysing samples for metabolomics, it is highly recommended to follow the guidelines provided by the Metabolomics Standards Initiative (Fiehn et al., 2007; Sansone et al., 2007). A number of biological replicates should be analysed in order to achieve statistical reliability; six biological replicates have been advocated as the minimum for metabolomics (Roessner et al., 2001; Lisec et al., 2006). MS and nuclear magnetic resonance spectroscopy (NMR), which are the the most commonly used analytical methods, are used in metabolomic studies to provide qualitative or quantitative data. In addition, these techniques provide information on metabolite structure, which is necessary for metabolite identification (Saito and Matsuda, 2010).

Table 1.

Examples of plant metabolomic approaches using a variety of analytical techniques for investigation of N metabolism

Technique(s)	Species	Factor(s)	Metabolite detection	References
HPLC and enzymatic activities	Arabidopsis	Nitrate-deprivation and -rich conditions to wild-type plants	Targeted analysis (glucose, fructose, sucrose, starch, oxoglutarate, and amino acids)	Scheible et al. (2004)
FT-ICR-MS and HPLC	Arabidopsis	Sulphur (S)-deficient, nitrate-deficient, or S- and nitrate-deficient wild-type plants	Metabolome analysis and amino acid analysis	Hirai et al. (2004)
LC-q-TOF-MS	Arabidopsis	Nitrate-deprivation and -rich conditions to three overexpressing and knockout mutants of the LBD gene family	Targeted flavonoid profiling	Rubin et al. (2009)
GC-MS	Arabidopsis	Nitrate-limitation and -rich conditions to single knockout mutant of cytosolic fumarase 2	Metabolite profiling for detection of primary metabolites	Pracharoenwattana et al. (2010)
HPLC, enzymatic activities and GC-MS	Arabidopsis	Nitrate-limitation and -rich conditions to wild-type plants	Targeted analysis and metabolite profiling	Tschoep et al. (2009)
GC-TOF-MS	Arabidopsis and rice	Overexpression of rice full-length cDNA encoding LBD37	Metabolite profiling for detection of primary and other metabolites	Albinsky et al. (2010)
CE-MS	Arabidopsis and rice	Overexpression of Arabidopsis NAD kinase gene	Metabolite profiling for detection of ionic metabolites in primary metabolism	Takahashi et al. (2009); Takahara et al. (2010)
GC-TOF-MS	Rice	Ammonium supplied to RNAi lines of the chloroplastic PEPC gene	Metabolite profiling for detection of primary and other metabolites	Masumoto et al. (2010)
1H-NMR	Maize	Suboptimal nitrate or nitrate-limiting conditions to two single cytosolic GS knockout mutants and their double knockout mutants	Metabolite profiling for detection of primary and other metabolites	Broyart et al. (2010)
GC-MS	Tomato	Three different nitrate regimes under varied light intensity to tomato leaves	Metabolite profiling for detection of amino acids, organic acids, carbohydrates, and other metabolites	Urbanczyk-Wochniak and Fernie (2005)

MS-based metabolomic techniques

For MS-based metabolomics, there are two types of analysis: direct infusion (or flow-injection) analysis and MS analysis in combination with chromatographic separation of metabolites. The latter is often used for the study of N metabolism, because chromatographic separation-based MS analysis has a long history for identification and quantification of metabolites.

Direct infusion analysis requires no chromatographic separation. Instead, the technique enables performance of very high-throughput analysis for measurement of the total mass spectrum of the crude sample or extracts as a fingerprint, providing information on many different compounds. Ionization methods such as electron spray ionization (ESI), which is a gentle and wide-ranging ionization technique without the need for chemical derivatization, are utilized. The data quality of this approach relies greatly on the resolution specification of the mass analyser. Ion trap, time-of-flight (TOF), quadruple (q)-TOF, Orbitrap, and Fourier transform ion cyclotron resonance (FT-ICR) mass analysers have been used, among which Orbitrap and FT-ICR-MS are the two most powerful ultra-high resolution mass spectrometers. The combination of ESI with these mass analysers allows direct characterization of metabolite peaks within complex samples, as described by Hirai et al. (2004) (Table 1).

In MS analysis combined with chromatographic separation, gas chromatography (GC)-MS and liquid chromatography (LC)-MS are currently the standard MS methods for metabolite analyses. In addition to these techniques, capillary electrophoresis (CE)-MS and two ultra-high resolution mass spectrometers, Orbitrap-MS and FT-ICR-MS, have been used to perform plant metabolomics. Table 2 presents examples of the kinds of metabolites that can be detected by various analytical techniques. Because of the enormous diversity of metabolites with different physicochemical properties, it is practically impossible to analyse the entire plant metabolome using a single analytical method, as shown in Fig. 2.

Table 2.

Examples of metabolite detection using a variety of analytical techniques for metabolomics

Metabolite class	Typical metabolites	Instrument(s)
Amino acids and their derivatives	Twenty amino acids, β-alanine, GABA, etc.	CE-MS, GC-MS, LC-MS, NMR
Amines	Polyamines, etc.	CE-MS, GC-MS
Alkaloids	Polar alkaloids (e.g. pyrrolizidine alkaloids)	LC-MS, NMR
Fatty acids and their derivatives	Saturated and unsaturated aliphatic monocarboxylic acids and their derivatives	GC-MS
Isoprenoids	Terpenoids and their derivatives	GC-MS (non-polar), LC-MS (polar)
Nucleic acids and their derivatives	Purines, pyrimidines, mono-, di-, and triphosphate nucleosides	CE-MS, GC-MS (partially)
Organic acids in central metabolism	TCA cycle intermediates, etc.	CE-MS, GC-MS, NMR, LC-MS (partially)
Pigments	Carotenoids, chlorophylls, anthocyanins, etc.	HPLC, LC-MS
Polar lipids	Phospholipids, mono-, di-, and triacylglycerols	LC-MS
Sugars and their derivatives	Mono-, di-, and trisaccharides, sugar alcohols, and sugar mono- and diphosphates	GC-MS, CE-MS (sugar phosphates), CE-DAD (partially)
Volatiles	Phenylpropanoid volatiles, aliphatic alcohols, aldehydes, and ketones, etc.	GC-MS
Other secondary metabolites	Polar phenylpropanoids (e.g., flavonols), etc.	LC-MS

CE-DAD, capillary electrophoresis–diode array detection; GABA, γ-aminobutyrate; TCA, tricarboxylic acid cycle.

Fig. 2.

Conceptual coverage of metabolite detection using MS-based metabolite techniques.

GC-MS can detect not only volatiles but also small molecules involved in central metabolism, such as sugars, amino acids, organic acids, fatty acids, and steroids (Fig. 2, Table 2). The main advantage of this technique is that the identification of a wide range of primary metabolites is facilitated by the high efficiency of capillary GC, the reproducible fragmentation of metabolites in the mass spectrometer, and the ready availability of large mass spectral libraries (Halket et al., 2005). Although derivatization steps are required for detection of such polar metabolites, there are many examples of its application to investigate N metabolism using the technique (Table 1).

By choosing appropriate columns and solvents, LC-MS offers the possibility to cover a wider range of sample types and complexities than other techniques, although optimization of LC-MS conditions for metabolome analysis is not easy. For detection of secondary metabolites of economic importance for human health, LC-MS has been optimized in the plant metabolomics field (Tohge et al., 2005; Matsuda et al., 2009; Allwood and Goodacre, 2010). Recently, several research groups successfully detected metabolites belonging to central metabolism in Escherichia coli (Bennett et al., 2009; Yuan et al., 2009), yeast (Buscher et al., 2009), and plants (Arrivault et al., 2009) using LC-MS. Sawada and his colleagues established a widely targeted metabolome analysis approach for measurement of ∼100 metabolites in primary and secondary metabolism using LC-MS (Sawada et al., 2009). One of the drawbacks of LC-MS-based techniques is that they suffer from ion suppression which may cause matrix effects (Annesley, 2003). These may cause matrix components present in the biological sample to influence the response of the analyte (i.e. competition among co-eluting entities for ionization energy) under investigation.

CE-MS is a relatively new separation technique that combines the high-resolution separation power of CE and the superior detection ability of MS to detect ionic compounds (Monton and Soga, 2007; Soga, 2007; Ramautar et al., 2009). Coverage of metabolite detection for metabolite profiling using GC-MS and CE-MS is considered to overlap, as shown in Fig. 2. However, one of the advantages of CE-MS-based metabolite profiling is that the technique can detect metabolites (e.g. sugar diphosphates and nucleotides) that become unstable during derivatization steps required for GC-MS-based metabolite profiling.

Two types of ultra-high resolution analytical techniques, Orbitrap-MS and FT-ICR-MS, have great potential for metabolomics, especially for identification of unknown metabolites (Ohta et al., 2010). FT-ICR-MS displays ultra-high mass resolving power, high mass accuracy, and very high sensitivity in a wide dynamic range. It allows direct determination of the elemental composition of a metabolite in a complex mixture of extracts (Giavalisco et al., 2009; Miura et al., 2010). Orbitrap-MS, which was developed recently, performs more modestly compared with FT-ICR-MS (Makarov and Scigelova, 2010). It is a high-speed, high ion-transmission instrument, because of shorter accumulation times than those of FT-ICR-MS.

NMR-based metabolomics techniques

NMR is a spectroscopic technique that takes advantage of the spin properties of the nucleus of atoms. NMR data allow for a quantitative analysis of a mixture, with the area under the peak linearly correlated to the concentration of a particular molecule. Because NMR detects atoms with magnetic moments, such as ¹H, ¹³C, and ¹⁵N, the N atoms in N metabolites can be observed directly. However, the abundance of ¹⁵N is very low in nature and cannot be detected without ¹⁵N enrichment of samples (Kikuchi et al., 2004; Harada et al., 2006). While NMR spectroscopy is highly reproducible (Mesnard and Ratcliffe, 2005; Zulak et al., 2008; Kim et al., 2010), its low sensitivity and resolution have impeded the use of this technique for metabolomics. On the other hand, metabolic flux analysis using NMR techniques enables metabolite dynamics to be traced by a combination of ¹³C or ¹⁵N labelling approaches (Lundberg and Lundquist, 2004; Baxter et al., 2007,a; Kruger et al., 2007; Sekiyama and Kikuchi, 2007).

Multiplatform metabolomics approaches

Because no single analytical platform can assess all metabolites in a single run, data integration from multiple platforms such as GC-MS, LC-MS, CE-MS, Orbitrap-MS, FT-ICR-MS, and NMR enables achievement of a wider coverage of metabolite detection than using any single platform. Several studies have used this concept as a metabolomic approach, for example ¹H-NMR and LC-MS (Gipson et al., 2008), ¹H-NMR and GC-TOF-MS (Biais et al., 2009), and ¹H-NMR, GC-MS, and LC-MS (Ong et al., 2009). Plants possess huge chemical diversity compared with other organisms (Dixon and Strack, 2003). Thus, metabolite profiling by integration of multiple metabolomic platforms can be a powerful tool for the elucidation of N metabolism in plants. A multiplatform approach consisting of four different TOF-MS instruments (GC-TOF-MS, LC-q-TOF-MS, CE-TOF-MS, and LC-IT-TOF-MS) has been developed. The data sets obtained can be automatically unified by a novel computational approach involving automated identifier unification using the MetMask tool (Redestig et al., 2010) and feature summarization. The developed workflow has been applied to evaluate genetically modified tomatoes (Kusano et al., unpublished results) and to perform association analysis between metabolites and agrologically important traits in rice (Redestig et al., unpublished results).

Statistics, correlation network analysis, and modelling approaches for the study of N metabolism

N metabolism is tightly connected to other factors, for example light, temperature, cell type, and developmental stage. Metabolite contents as well as gene expression levels are regulated in response to such multiple input signals. Omic approaches provide snapshots of the regulatory events in the form of large data sets. Statistical data analysis is applied to survey and evaluate these. Many statistical methods are currently used in different fields. Here, statistical methods that are often used for evaluation of metabolomic data sets as well as other ‘omic’ data sets, such as transcriptomic and proteomic data, are introduced (Table 3).

Table 3.

Typical statistical methods for metabolomics

Univariate analysis

• t-tests (e.g., Student's t-test, Welch's t-test)

• Analysis of variance (ANOVA)

Multivariate analysis

Unsupervised method

• Principal component analysis (PCA)

• Independent component analysis (ICA)

• Cluster analysis

○ Hierarchical clustering (e.g. hierarchical cluster analysis)

○ Partitional clustering (e.g. k-means)

• Unsupervised neural networks (e.g. self-organizing maps)

• Correlation analysis

Supervised method

• Partial least squares (PLS)

• Artificial neural networks

• Discriminant analysis (e.g. PLS-DA, discriminant function analysis)

• Evolutionary-based algorithm (e.g. genetic algorithm)

• Regression analysis (e.g. multiple linear regression, PLS-regression)

• Regression trees (e.g. random forests, multivariate adaptive regression splines)

Statistical analysis based on GLM

The generalized linear model (GLM) is a large class of statistical models for relating responses to linear combinations of independent predictor variables. The GLM underlies most of the statistical analyses, including the t-test and analysis of variance (ANOVA). A t-test (Student's t-test and Welch's t-test), the most popular statistical method, assesses whether the means of two groups (samples or treatments) are statistically different at a given confidence level. The t-test is categorized as a special case of one-way ANOVA. For example, the t-test can be applied when different groups of N-treated samples are compared with non-treated samples (controls). An ANOVA allows comparison of the effects of two or more levels of factors. If the null hypothesis is rejected by ANOVA, post-hoc tests are conducted to assess which groups are significantly different from each other, based on their mean rank differences (Pavlidis et al., 2003; Kusano et al., 2010).

Combinatorial design is a testing approach that defines a small set of experiments that will effectively explore the effects of all possible combinations of multiple inputs. Shasha et al. (2001) introduced the use of a combinatorial design approach for the study of gene expression as the output generated by six inputs, namely light, starvation, C, N, and glutamate and glutamine levels. This approach reduced the total number of experiments to test how these six inputs interact to regulate gene expression, and provided smaller sizes of experimental data sets necessary to obtain a conclusive answer.

An excellent example study using the GLM concept and combinatorial design was reported by Gutierrez et al. (2007). The authors treated Arabidopsis roots with a combination of different concentrations of sucrose and nitrate to investigate C and N interactions (16 experiments in total), according to their hypothetical models of gene expression by C, N, and C×N interactions as multiple inputs. Correlation analysis and ANOVA were conducted using a transcript profiling data set to explore global gene expression in the 16 experiments. The genes identified as responding to C and/or N were then projected onto the multiple network in the VirtualPlant platform (Gutierrez et al., 2005; Katari et al., 2010). This multiple network allowed the identification of clusters consisting of specific interacting genes regulated by C, N, or C×N interactions, for example metabolism, protein synthesis and degradation, and cellular processes. Interestingly, many genes responsive to C, N, and C+N treatments were involved in metabolic pathways in the qualitative networks (Gutierrez et al., 2007). It will be of great interest to see whether a similar approach can be applied using metabolite profiling data sets together with transcript profiling data sets.

Multivariate analysis is one of the statistical techniques used to analyse data that arise from more than one variable. The use of multivariate analysis allows analysis of complex data sets, for example transcript profiles and metabolite profiles. Typical multivariate analyses applied in metabolomics include principal component analysis (PCA), hierarchical cluster analysis (HCA), and partial least squares (PLS) regression. PCA can be used to explore data in an unsupervised fashion by compressing the data to a small set of representative features. PLS regression is used both to explore and to test the correlation between the data and one or more dependent variables. There are many examples of the application of multivariate analysis using metabolite (Kusano et al., 2010), protein (Di Carli et al., 2009), and transcript profiles (Street et al., 2008) or a combination of these profiles (Bylesjo et al., 2009).

Correlation network analysis

Correlation network analysis is a statistical technique for the measurement of strengths of association between two variables (bivariates) without establishing a cause-and-effect relationship. In the case of a gene–gene correlation approach, when the expression of one gene shows significant correlation with the expression of another gene in a large correlation matrix, these two genes are likely to be co-regulated. This approach can be used for the identification of novel candidate genes in correlation networks based on the ‘guilt by association’ principle (Wolfe et al., 2005; Saito et al., 2008).

A metabolite–metabolite correlation approach allows insight to be gained into cellular metabolite status, although correlated metabolites are not always likely to be associated with a common biological function such as biochemical reactions in metabolic pathways (Camacho et al., 2005; Steuer, 2006; Muller-Linow et al., 2007; Saito et al., 2008). Metabolite–metabolite correlations can be observed across individual biological replications (Weckwerth et al., 2004; Kusano et al., 2007) and/or among samples regarding species-, time-, tissue-, or dose-dependent differences (Morgenthal et al., 2006; Noguchi et al., 2006; Muller-Linow et al., 2007). Metabolite–metabolite correlations can also arise under two different situations, namely global and specific perturbations (Steuer, 2006). Global metabolic perturbations are generated by environmental shifts (e.g. N treatment) and specific perturbations by overexpression or knockout of specific enzymes, for example (Fig. 3). By combining information about changes in metabolite levels, metabolite–metabolite correlations may provide information to generate novel hypotheses to understand underlying biochemical systems in the cell (Fukushima et al., in press).

Fig. 3.

Amplitude of global and specific perturbations triggered by N treatment in plants. The extent of global perturbations triggered by N treatment is much greater than that of specific perturbations (e.g. overexpression or knockout of genes), because N is one of the most important factors affecting different molecular levels (DNA, RNA, proteins, and metabolites) in plants.

Computational modelling approach

A computational modelling approach is a currently challenging field in systems biology to simulate and predict the dynamic responses of metabolic networks in plants (Morgan and Rhodes, 2002; Rios-Estepa and Lange, 2007; Schallau and Junker, 2010). Most examples of its application for understanding plant metabolic pathways are related to photosynthesis (Zhu et al., 2007; Luo et al., 2009) and sugar metabolism (Uys et al., 2007). Recently, Curien et al. (2009) constructed a large-scale kinetic model of the aspartate-derived amino acid pathway in Arabidopsis. The study identified allosteric interactions that are independently regulated among fluxes in competing pathways. In particular, the modelling approach demonstrated that the change of threonine level is the most sensitive variable in the system. Interestingly, the comparative approach of three methionine over-accumulation (mto) mutants based on metabolite profiling provided a similar conclusion (Kusano et al., 2010). Among the mto mutants, mto2 lacking threonine synthase (Bartlem et al., 2000) showed the most pronounced metabolite changes not only in the aspartate pathway but also in other pathways in central metabolism.

Applications

Since Fiehn et al. (2000) established a metabolite profiling method for Arabidopsis mutants, numerous examples of metabolomic studies have been published. These include plant stress responses (Sanchez et al., 2008; Lehmann et al., 2009; Urano et al., 2009), assistance with novel gene annotations (Bottcher et al., 2009; Okazaki et al., 2009), circadian clock function (Fukushima et al., 2009,a), and assessment of substantial equivalence of genetically modified crops (Catchpole et al., 2005; Beale et al., 2009). However, to our knowledge, there are fewer examples of the use of metabolomics for the study of N metabolism. In this section, an overview of some of the important studies to obtain insights into N metabolism in Arabidopsis and crops and vegetables using a metabolomic approach is presented (see also Table 1).

Arabidopsis

The most extensive metabolomic studies aiming to elucidate N metabolism have been conducted on Arabidopsis. Tschoep et al. (2009) performed metabolite profiling to study the response of phenotypic changes, biomass, major metabolites, and enzymes in central metabolism of Arabidopsis wild-type leaves in response to a mild and sustained N limitation. Metabolite measurement was achieved by a combination of authentic methods and GC-MS-based metabolite profiling. Short-term (diurnal) and long-term (age-dependent) changes in amino acid and organic acid contents under low and high N conditions were profiled. Interestingly, the authors found that the low N growth regime caused shoot growth reduction and increased levels of many amino acids (e.g. glutamate and glutamate-derived amino acids), although the total protein content remained unchanged.

A metabolomic approach can be used to characterize novel gene functions involved in N metabolism. As an excellent example, three N- and nitrate-induced members of the LATERAL ORGAN BOUNDARY DOMAIN (LBD)/ASYMMETRIC LEAVES2-like (ASL) gene family of transcription factors (LBD37/ASL39, LBD38/ALS40, and LBD39/ASL41) are dramatically induced in wild-type Arabidopsis when sufficient nitrate is supplied to N-limited seedlings, although the transcript levels of the genes were low under the N-limited condition (Rubin et al., 2009). N-limited LBD37/ASL39-overexpressing seedlings accumulate no anthocyanins, while N-limited wild-type and N-replete lbd37/ASL39 T-DNA mutant seedlings accumulate anthocyanins. Flavonoid profiling using LC-q-TOF-MS analysis of these overexpressors and knockout mutants revealed that the levels of two cyanidin glycosides were altered in response to N-limited and N-sufficient conditions, although other flavonol contents remained unchanged. By integrating these findings and transcript profiles of N-limited LBD37/ASL39 and LBD38/ASL40 overexpressors as well as N-limited and N- and nitrate-replete wild-type seedlings, the authors concluded that these three genes negatively regulate anthocyanin biosynthesis in Arabidopsis.

Another example is the characterization of a cytosolic fumarase in Arabidopsis (Pracharoenwattana et al., 2010). The Arabidopsis genome contains two genes encoding fumarases, FUMARASE1 (FUM1) and FUMARASE2 (FUM2). The green fluorescent protein (GFP)-based subcellular localization experiments revealed that the former encodes a mitochondrial fumarase, while the latter encodes a cytosolic fumarase and is abundantly expressed in leaves. GC-MS-based metabolite profiling of T-DNA knockout mutants captured the increase of malate, and dramatic decrease of fumarate, contents in the mutant leaves. Furthermore, the levels of many amino acids in the mutant leaves showed a significant decrease overnight, indicating that FUM2 affects these amino acid concentrations via fumarate/malate homeostasis. The mutant plants accumulate starch, suggesting that FUM2 may function to maintain C metabolism. This result, together with those from other experiments including N treatment experiments, revealed that FUM2 has crucial roles not only for the accumulation of fumarate from malate together with many amino acids and starch in leaves, but for effective assimilation of N and growth of Arabidopsis under high N conditions.

Rice

Rice is one of the most important crops in the world, as it feeds over half of the world's population. In addition, rice grown in paddy fields has a unique system for assimilation of N compared with N metabolism in rice in general. Metabolite profiling of rice plants may provide new insights for understanding this unique N assimilation mechanism. The characterization of novel gene functions in rice is also important, for example for achievement of breeding strategies. The combination of a full-length (FL) cDNA overexpressor gene hunting system (FOX hunting system) (Ichikawa et al., 2006) and metabolite profiling allows fast and reliable identification of candidate lines showing altered metabolite profiles. To identify and characterize novel gene functions in rice, Arabidopsis lines overexpressing rice FL cDNAs (rice FOX Arabidopsis lines) (Kondou et al., 2009) were screened using a GC-MS-based technique to identify rice genes that caused metabolic changes (Albinsky et al., 2010). Through screening, a rice FOX Arabidopsis line was identified that harboured the FL cDNA of the rice orthologue of the LBD/ASL gene of Arabidopsis, AtLBD37/ASL39. This line as well as Arabidopsis AtLBD37/ASL39-overexpressor plants showed prominent changes in the levels of metabolites related to N metabolism. Furthermore, OsLBD37/ASL39-overexpressing rice plants also showed altered levels of metabolites involved in N metabolism (Fig. 4A). The PCA loading plot demonstrated that many amino acids showed decreased levels in the transgenic lines when compared with those in the control samples (Fig. 4B). Combination of metabolite and transcript profiling using rice FOX Arabidopsis and OsLBD37/ASL39-overexpressing lines revealed that: (i) the function of LBD37/ASL39 is probably conserved in these dicot and monocot model plant species; and (ii) the gene is likely to be involved in the regulation of N metabolism. This hypothesis also supports the findings described by Rubin et al. (2009) (see the previous section). AtLBD37/ASL39, together with two homologous genes, have been characterized as negative regulators of anthocyanin biosynthesis and genes involved in nitrate uptake/assimilation in Arabidopsis.

Fig. 4.

PCA of metabolite profiles of the leaf blade (LB) samples obtained from three different rice OsLBD37/ASL39-overexpressor plants (K16331 lines) and control rice plants. The data matrix consisted of 31 samples and 109 annotated metabolite peaks including known and unknown metabolite peaks with mass spectral tags (Kopka, 2006). (A) PCA score scatter plot of metabolite profiles of LB samples of the three independent rice K16331 lines and two control rice plants, empty vector control and the wild-type. The control plants and the K16331 samples showed clear separation in PC1. The symbols in the PCA score scatter plot indicate as follows: red dot, the empty vector control (BIG3); pale-green asterisk, K16331_4; blue asterisk, K16331_13; orange asterisk, K16331_19; and pink box, the wild-type cultivar Nipponbare. (B) PCA loading scatter plot of metabolite profiles of LB samples of the K16331 lines and two control rice plants. To avoid complexity of the plot, only the metabolite names of the N-containing metabolites are presented in the plot. The plot shows that the levels of glutamine, asparagine, and some amino acids tended to be decreased in the K16331 lines when compared with those in the control samples.

Masumoto et al. (2010) provided another example of gene identification and characterization in rice. The main focus of their study was to characterize phosphoenolpyruvate carboxylase (PEPC) in rice. PEPC catalyses conversion of PEP into oxaloacetate (OAA). Among six putative PEPC genes found in the rice genome database, the authors focused on one of the plant-type PEPCs (Osppc4). Osppc4 is targeted to the chloroplast while PEPC genes and proteins in bacteria, algae, and vascular plants are believed to localize in the cytosol. A GC-TOF-MS-based metabolomics approach together with measurement of ammonium content in RNA interference (RNAi) knockdown mutants revealed that the novel PEPC localized in the chloroplast suppressed ammonium assimilation and subsequent amino acid synthesis by reducing levels of organic acids, which are C-skeleton donors for these processes.

A metabolomic approach is often conducted for elucidation of the functions of important enzymatic genes involved in central metabolism. Nicotinamide adenine dinucleotide (NAD) and its relative nicotinamide adenine dinucleotide phosphate (NADP) are two of the most important cofactors in many metabolic processes. Three genes encode nicotinamide adenine nucleotide kinase (NADK) in Arabidopsis (Berrin et al., 2005; Turner et al., 2005). Among these, a reverse genetic approach using knockout mutants revealed that the chloroplastic NADK (AtNADK2) gene in Arabidopsis is known to stimulate C and N metabolism (Chai et al., 2005; Takahashi et al., 2006). AtNADK2 was overexpressed in rice and Arabidopsis, and CE-MS-based metabolite profiling of the overexpressors was conducted to capture biochemical snapshots of primary metabolism, particularly of ionic metabolites (Takahashi et al., 2009; Takahara et al., 2010). Changes in amino acid and sugar phosphate contents in the overexpressor lines tended to be similar in both species (e.g. an increase in the levels of glutamine, glutamate, and threonine, and a decrease in the level of ribulose-5-phosphate), although the observed metabolic enhancement is considered to be a pleiotropic effect by enhancement of the AtNADK2 expression level using the cauliflower mosaic virus (CaMV) 35S promoter.

Plants grown under flooded conditions (e.g. rice in paddy fields) assimilate ammonium predominantly as inorganic N, while many other plants assimilate these ions as nitrates. Glutamine synthetase (GS) is essential for assimilation of ammonium in rice because GS catalyses glutamate and ammonium condensation to yield glutamine. GS primarily exists as two isoenzymes with different subcellular localizations: GS1 in the cytosol and GS2 in plastids (Hirel and Gadal, 1980). GS1 is important for normal growth and development (Yamaya and Oaks, 2004), and GS2 is required for the photorespiratory metabolism of N in chloroplasts (Wallsgrove et al., 1987). Of the three genes encoding GS1 in rice, OsGS1;1 is a critical regulator of normal growth and grain filling (Tabuchi et al., 2005). Using OsGS1;1 knockout mutants, the change in metabolic networks triggered by global and specific perturbations are being investigated (Fig. 5). The heatmaps show metabolite–metabolite correlation relationships in leaf blade samples of wild-type rice (cv. Nipponbare) and knockout mutants with either ammonium or water supply (Fig. 5B, C). The correlation heatmaps showed different patterns between the ammonium and water conditions, suggesting that environmental perturbation by ammonium treatment cause the changes in metabolic networks in each genotype. Specific perturbation caused by OsGS1;1 knockout was also observed in rice. The heatmaps of the wild type and the knockout mutants under ammonium and water conditions show similar correlation relationships in each condition. This result means that the extent of the change in metabolic networks by the specific perturbation is smaller than that caused by the global perturbation. Nevertheless, mutant-specific correlations in the heatmap of the knockout mutant with ammonium treatment were found (Fig. 5B). For example, a greater number of negative correlations were observed in the heatmap of the knockout mutant than that of the wild type. These results, together with the metabolite profiles of the knockout and wild-type samples, will be published elsewhere.

Fig. 5.

Metabolite–metabolite correlation heatmaps of LB of the rice OsGS1;1 knockout mutant (KO) and wild-type rice (WT; cv. Nipponbare) under two different N conditions. (A) A joint correlation heatmap that consists of two different correlation data sets, WT (the triangle under the diagonal line) and KO (that above the line). Pearson's correlation coefficient was calculated using individual biological replicates of WT (20 samples) and KO (16 samples), respectively. The joint correlation heatmaps of WT and KO under (B) ammonium and (C) water conditions. The colour of each cell depicts the Pearson's correlation coefficient value, with deeper colours indicating higher positive (red) or negative (blue) correlation coefficients. A total of 73 metabolite peaks that were commonly observed in both experiments were used.

Other crops and vegetables

Metabolomics approaches can be directly applied to other plants because measurement of metabolites needs no genomic information. Therefore, metabolite profile analysis has been performed for a variety of plant species (Jeong et al., 2004; Mercke et al., 2004; Huang et al., 2008; Galindo et al., 2009; Peluffo et al., 2010; Weichert et al., 2010; Zhang et al., 2010). Of these analyses, studies for which metabolite profiling was related to N metabolism are summarized.

Cytosolic GS has a crucial role not only in rice but also in maize (Martin et al., 2006). Maize mutants deficient for the expression of two genes encoding GS isoenzymes, GS1.3 and GS1.4, and the double knockout mutant show reduced grain yield, but these mutants exhibit normal growth. To obtain insights into N remobilization from shoots to kernels during the kernel filling period, Broyart et al. (2010) performed ¹H-NMR-based metabolite profiling of glutamine synthetase (gln)1.3, gln1.4, and the gln1.3/1.4 double mutant. Although the metabolite profiles of the three mutants did not show dramatic metabolite changes under high N or low N fertilization conditions, the levels of N-containing metabolites (e.g. asparagine, alanine, threonine, phenylalanine, tyrosine, and phophatidylcholine) tended to be elevated. Phylogenetic analysis revealed that OsGs1;1, gln1.3, and gln1.4 were classified in the same group (Nogueira et al., 2005; Bernard et al., 2008; Yajun et al., 2008). However, mutants lacking OsGs1;1 exhibited severe growth retardation (Tabuchi et al., 2005). This indicates the non-redundant function of the GS1 isoenzymes in rice and maize.

Inorganic N assimilation enables the production of amino acids, which is one of the most important metabolic processes in plants. Nitrate levels can affect carbohydrate metabolism in dicots (Scheible et al., 1997; Stitt et al., 2002). To characterize the response to altered N nutrition under varied light intensity, tomato plants were grown hydroponically in liquid culture under different nitrate and light regimes (Urbanczyk-Wochniak and Fernie, 2005). Although data interpretation for such a broad range of metabolite profiles is difficult without further experimentation, GC-MS-based metabolite profiling using leaf material revealed that the levels of amino acids and organic acids were tightly coordinated with response to nitrate regimes. On the other hand, carbohydrate levels tended to be more affected by the different light intensities than by the nutrient regime. Application of non-targeted metabolite profiling in this study revealed not only well-known metabolite changes, but also unexpected metabolic responses. Several metabolites, including chlorogenate, showed specific responses to nitrate stress. Metabolite profiles in this study also demonstrated that many miscellaneous metabolites can be produced under high light intensity with nitrate saturation.

Future perspectives for understanding N metabolism from the point of view of metabolomics

This review provides an overview of current analytical platforms and statistical approaches used in metabolomics, and examples of the application of a metabolomic concept for understanding N metabolism. Metabolomics is one of the most important members of the ‘omics’ fields. N regulatory mechanisms in plants are highly complex, and metabolomic data should be integrated with other ‘omic’ data sets, such as proteomic and transcriptomic data, as part of a systems biology approach. To extend measurement of steady-state levels of metabolites to that of metabolite dynamics, measurement of metabolic fluxes is required. Flux analysis can trace the dynamics of N status in metabolite structures. Current flux analysis has been demonstrated based on a combination of steady-state stable isotope labelling and NMR- or MS-based techniques (Mesnard and Ratcliffe, 2005; Baxter et al., 2007,b; Huege et al., 2007; Williams et al., 2010). It is believed that a systems biology approach by integration of multidimensional data (e.g. multiomic data, physical parameters, genotype data, and phenotypic traits) together with measurement of metabolic dynamics is a key to attaining a better understanding of the complex regulatory mechanisms of N metabolism.

We are grateful to M. Tabuchi and T. Yamaya at Tohoku University for providing plant materials for the study of OsGS1;1. We also thank D. Albinsky, N. Hayashi, and M. Kobayashi at RIKEN PSC for their technical assistance. We thank P. Jonsson, H. Stenlund, and T. Moritz at the Umea Plant Science Centre for sharing the data pre-treatment program for GC-TOF-MS data.