High-resolution plant metabolomics: from mass spectral features to metabolites and from whole-cell analysis to subcellular metabolite distributions


(fax +49 331 5678236; e-mail giavalisco@mpimp-golm.mpg.de).


The main goal of metabolomics is the comprehensive qualitative and quantitative analysis of the time- and space-resolved distribution of all metabolites present in a given biological system. Because metabolite structures, in contrast to transcript and protein sequences, are not directly deducible from the genomic DNA sequence, the massive increase in genomic information is only indirectly of use to metabolomics, leaving compound annotation as a key problem to be solved by the available analytical techniques. Furthermore, as metabolites vary widely in both concentration and chemical behavior, there is no single analytical procedure allowing the unbiased and comprehensive structural elucidation and determination of all metabolites present in a given biological system. In this review the different approaches for targeted and non-targeted metabolomics analysis will be described with special emphasis on mass spectrometry-based techniques. Particular attention is given to approaches which can be employed for the annotation of unknown compounds. In the second part, the different experimental approaches aimed at tissue-specific or subcellular analysis of metabolites are discussed including a range of non-mass spectrometry based technologies.


Every biological organism relies for its proper function on the interaction of a multitude of cellular molecules. The interplay between these molecules is complex and needs to be coordinated with respect to both time and space. Over the years, the study of both small molecules and macromolecules has progressed from the analysis of single compounds to the study of complex mixtures, enabling the detection and identification of several hundred to several thousand different entities (Feng et al., 2008; Hsu et al., 2011). Due to technological developments the major molecular classes studied comprise transcripts, proteins, and metabolites. The most comprehensive analytical techniques have been developed for RNA molecules, and present-day technologies allow a full analysis of all RNAs present in a given biological system (Gresham et al., 2008; Forrest and Carninci, 2009). In contrast, there is yet no single technique that allows monitoring of all proteins, including their post-translational modified derivatives (Wade et al., 2004; Gstaiger and Aebersold, 2009; Yates et al., 2009; Walther and Mann, 2010). This is because proteins, in contrast to RNA, differ significantly in their chemical properties and thus also their analytical behavior. Even more difficult is the study of small molecules and metabolites, where analytical tools face three challenges: (i) metabolites differ vastly in their concentration; (ii) they comprise compounds of extremely different chemical properties (e.g. a hydrophilic organic acid versus a highly hydrophobic triacylglyceride); and, most importantly, (iii) metabolomics benefits only indirectly from completed or ongoing genome sequencing projects. The reason for this is obvious: whereas RNA and protein sequences are essentially collinear with genomic information, metabolic structures represent a completely different level of realization of genome information which does not necessarily display collinearity with the genome sequence (Saito and Matsuda, 2010).

Two main technologies, nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry (MS), have been employed in the field of metabolomics research. Both technologies are highly complex and provide different advantages and limitations (Dettmer et al., 2007; Kim et al., 2011). Compared with the MS-based approaches, NMR requires almost no sample preparation and, in principle, allows structural determination and quantitative analysis of the most abundant signals (Kim et al., 2011). Unfortunately, the lack of sensitivity and the associated lack of brought compound annotation represents a serious drawback when using NMR for metabolomics on complex mixtures (Kim et al., 2011).

In contrast, MS-based metabolomics approaches in combination with different separation techniques (gas or liquid chromatography but also capillary electrophoresis) provide a much higher sensitivity and flexibility (Villas-Boas et al., 2005; Dunn, 2008; Allwood and Goodacre, 2010). As a consequence, MS-based approaches provide deeper and more detailed insight into the metabolic inventory of a biological sample (Kim et al., 2011), even though precise compound identification still represents a major bottleneck (Saito and Matsuda, 2010; Lei et al., 2011).

In this review article we focus on MS-based metabolomic analysis in plants, since only the high sensitivity achieved with this technology provides the means to potentially tackle the whole metabolome, defined as the complete set of small molecules present in a single cell/organism (Oliver et al., 1998). In the second part we further review several strategies used for the analysis of subcellular metabolite distributions within a single cell or different cell types.

Targeted and Non-Targeted Analysis in MS-Based Metabolomics

Hyphenated MS-based metabolomics, meaning the combination of, for example, chromatographic separation techniques and MS, are available in various different flavors (Villas-Boas et al., 2005; da Silva et al., 2006; Dettmer et al., 2007; Dunn, 2008; Feng et al., 2008; Lei et al., 2011). The main parameter for the choice of a given method depends on the initial biological question and on the nature of the compounds to be measured (Figure 1). Restriction to a single or several analytical methods has to be taken, since currently neither an all-inclusive extraction method nor a technological platform able to measure all possible metabolites in a single analysis is available or in sight of being developed soon (Saito and Matsuda, 2010; Lei et al., 2011). This means that selection of the technology should be matched to the specific set of metabolites or based on the function of the target gene of interest. If more than one method is available, the most sensitive and robust analytical technology should normally be selected.

Figure 1.

 Schematic diagram indicating the work flow from biological sample to metabolite functional assignment.
Two main routes are given, which either follow a targeted or a non-targeted metabolome analysis.
NMR, nuclear magnetic resonance; MS, mass spectrometry; PC, principal component.

An approach aiming to measure and profile metabolites with known chemical structures is commonly called targeted metabolomics (Albinsky et al., 2010; Dudley et al., 2010). The best established systems for such a strategy are gas chromatographic (GC) separation systems, coupled either to time-of-flight (TOF) or quadrupole mass spectrometers (Lisec et al., 2006) or liquid chromatography (LC) systems coupled to, for example, triple quadrupole (QqQ) mass spectrometers (Lu et al., 2008). In plant science the GC-TOF MS system has already been in use for several years as one of the workhorses for the analysis of primary metabolism (Schauer et al., 2005; Lisec et al., 2006; Williams et al., 2007; Lytovchenko et al., 2009; Shuman et al., 2011) but to some extent also for secondary compounds like triterpenes (Field and Osbourn, 2008) or sterols (Klahre et al., 1998). The LC-QqQ systems can, depending on the chromatography used, cover several compound classes ranging from highly polar compounds of the Calvin cycle (Arrivault et al., 2009) to semi-polar plant hormones (Pan et al., 2010) and non-polar compounds like plant lipids (Markham and Jaworski, 2007).

The major limitation of targeted metabolic profiling techniques is that they require authentic reference compounds for the identification and quantification of the measured metabolic profiles, limiting the analysis to compounds that are either commercially available (Last et al., 2007; Nakabayashi et al., 2009) or already identified and validated by orthogonal technologies like NMR, for example (Kim et al., 2011).

In many cases, the class of metabolites of interest might not be certain at the beginning of the study. For example, studies might seek to understand the metabolomic influence of a gene with a more generic function, such as a kinase or a transcription factor, or even a gene of unknown function. As there is no a priori knowledge or idea about which metabolites will be influenced by this gene, a targeted metabolome analysis runs a high risk of missing the most significant changed metabolites. Obviously in this case a consequent step of the study is to expand the analysis beyond the known targets and include signals of unknown identity. Of course the strategy of broadening the analysis beyond the targeted pathway of interest is also useful if the primary function of the gene is known. This approach is commonly called non-targeted, or unbiased, analysis (De Vos et al., 2007; Hanhineva et al., 2008; Tianniam et al., 2009; Oliver et al., 2011). Non-targeted analysis can be performed in two ways. One approach is so-called metabolic fingerprinting, where no chromatographic separation of the samples prior to the MS-based measurement is performed (Last et al., 2007). The second approach, called metabolite profiling, usually includes a chromatographic separation step prior to the mass spectrometric measurement (Last et al., 2007; Allwood and Goodacre, 2010). Both methods provide complex metabolic signatures which can be analyzed by multivariate statistical methods (Bylesjo et al., 2007; Sumner et al., 2007b; Eliasson et al., 2011). This provides information on metabolic features that are significantly different between the analyzed samples. At this stage of analysis it is not important if these features correspond to an immediately identifiable chemical compound. As long as these known unknowns can be reproducibly detected, they can serve many useful purposes in diagnostics or systems biology. Once their importance as a diagnostic marker, for example, is validated, more sophisticated analytical technologies can be applied to unravel the exact chemical structure of the compounds of interest.

A number of studies have successfully used metabolic fingerprinting or profiling strategies to discriminate or associate different gene functions, determine system-specific metabolic responses, or simply associate quality features of a sample (Messerli et al., 2007; Beckmann et al., 2008; Cuadros-Inostroza et al., 2010; Scherling et al., 2010; Lisec et al., 2011; Smith and Bluhm, 2011). Still, in the long run, the ultimate goal of every non-targeted study should be not only to make use of the whole metabolic signature for sample discrimination, or marker detection, but also to identify those discriminative compounds.

Compound Annotation: Making Use of High-Resolution, Accurate Mass, Isotope Labeling and Compound Fragmentation for the Annotation of Metabolites

As discussed above, the most straightforward way to identify a compound by MS-based methods is to match a signal of a measured analyte to a pure reference compound. Unfortunately, in many cases this is not possible due to the unavailability of several chemical standards potentially present in plants (Saito and Matsuda, 2010; Lei et al., 2011). As a consequence the structural elucidation of an unknown analyte, not matching any of the commercially available reference compounds, represents one of the major bottlenecks in metabolomics studies (Neumann and Bocker, 2010). A true de novo determination of the final structure of a small molecule, including its stereochemistry, might be possible with higher-dimension NMR methods; this as a rule requires extensive purification and enrichment of the compound under study and thus cannot be achieved for every measured metabolite within complex mixtures (Kim et al., 2011).

As a consequence of this identification problem a hierarchical annotation system was developed for MS-based metabolomics studies, where different levels of compound identification are described (Sumner et al., 2007a). In Figure 2 a number of commonly used tools or strategies for MS-based compound annotation are summarized. But still it must be understood that none of these MS-based methods alone can fully solve the structure of an unknown compound. However, the combination of several of these methods can serve to predict a handful of possible structures, which can be tested with commercial or newly synthesized standards (Neumann and Bocker, 2010).

Figure 2.

 Overview of commonly used mass spectrometry-based metabolite annotation strategies.
The represented compound structure in the center of the figure, with the elemental composition C12H22NO10S3 is 4-methylsulphinylbutyl glucosinolate (glucoraphanin). The five surrounding graphical illustrations represent physicochemical properties of the measured compound, which next to the measured mass, can be exploited for compound annotation or at least for the rejection of unlikely annotation. The combinatorial use of several of the displayed analyses might therefore lead at least to a partial elucidation of the structure.
RIA, relative isotopic abundance; fwhm, full width half maximum.

One of the most significant improvements in compound identification in MS-based metabolomics analysis came with advancements in mass accuracy (measured in parts per million, p.p.m.) and the resolving power or resolution (mm) of the instruments (Schaub et al., 2005; Makarov et al., 2009; Andrews et al., 2011; Pelander et al., 2011). These improvements, in combination with more sophisticated software packages, facilitate a more efficient use of the measured mass spectrometric and chromatographic features (Bocker et al., 2009; Wolf et al., 2010; Boswell et al., 2011a,b; Rasche et al., 2011; Rojas-Cherto et al., 2011).

The ever increasing resolving power and the improved mass accuracy, which can reach a resolution of more than 1 000 000 (at full width half-maximum, fwhm) and sub-p.p.m. mass accuracies in commercially available Fourier transform ion cyclotron (FT-ICR) mass spectrometers (Marshall and Hendrickson, 2008) significantly boosted metabolomics studies. These two features first of all help to separate overlapping mass signals (avoiding confounding signal interference), but also, if isotopic fine structures are to be determined, help to predict the elemental composition of measured metabolic signals with a mass up to 350 Da (Koch et al., 2007). Of course it has to be kept in mind that resolution and accurate mass alone will not be sufficient to predict the elemental composition of a whole metabolome.

Kind and Fiehn have shown that a mass spectrometer with 3 p.p.m. and 2% relative isotope abundance accuracy can predict and annotate elemental compositions more efficiently than a FT-ICR with a mass accuracy of better than 0.1 p.p.m. (Kind and Fiehn, 2006; Koch et al., 2007). Additionally, it should be kept in mind that most of the ultra-high resolution (resolution >350 000) instruments have rather low scanning rates (depending on a measured resolution in the range of seconds), and so are not compatible with the fast chromatography of the <2 μm particle-size ultra-high performance chromatography (UPLC) systems, which achieve chromatographic peak widths of 5 sec and less (Plumb et al., 2004). As a consequence these slow-scanning high-mass-accuracy instruments can only use their full resolution in combination with either direct-infusion-based measurements (Aharoni et al., 2002; Giavalisco et al., 2008), where scanning rates are not limiting, or in combination with sample fractionation strategies, where peaks of interest are collected during LC-MS measurements (Li et al., 2007).

In recent years quadrupole TOF instruments have improved not only their resolution (up to 50 000 fwhm) but also their mass accuracy and their scanning rates (Andrews et al., 2011; Pelander et al., 2011). These instruments are now, with low p.p.m. mass errors, extremely accurate (Stroh et al., 2007). Due to the use of TOF detectors, these instruments have always been fast, but now they reach extremely fast scanning rates with up to 100 mass scans per second (Andrews et al., 2011); this enables several different scan modes, including several tandem MS (MS/MS) scans (Matsuda et al., 2009, 2010). In addition, the new TOF detector provides an excellent dynamic range and relative isotopic abundance (RIA) accuracy of 3% and better (Bocker et al., 2009; Andrews et al., 2011; Pelander et al., 2011). Isotopic abundance distribution is a valuable compound-specific feature, which in combination with a high RIA accuracy can be used to predict or discriminate elemental compositions. Within the last years a number of software packages have therefore implemented RIA (Bocker et al., 2009; Pluskal et al., 2010).

Unfortunately, the RIA accuracy in mass spectrometers of the Orbitrap type is, with a median error of approximately 20%, not as good (Xu et al., 2010) as the one obtained from the TOF-type instruments. However, these slow-scanning ion trap-like systems still have a distinct advantage, since they are capable of performing high-order tandem MS (MSn) experiments. Meaning that these instruments are not only capable of producing primary fragments, but also secondary, tertiary and so on (higher-order MSMS). These generated higher-order fragmentation trees can subsequently be used for efficient elucidation of elemental composition and partial structure (Rojas-Cherto et al., 2011).

Another updated approach to the improved annotation of elemental compositions of metabolites from LC-MS studies was published recently (Boswell et al., 2011a,b). This approach, which is theoretically independent of the type of MS used and also of the LC system (even though high-mass-accuracy systems will benefit more from the approach), aims to help validate the compound annotation by calculating a predicted retention time for each measured compound in a gradient LC run. This calculation is based on the isocratic log κ to volume of organic modifier (Ø) relationship. To get started the system needs to be tuned by a number of internal reference compounds. Subsequent prediction of the potential annotated known structures can reach an accuracy of better than 2% (Boswell et al., 2011a,b). Still, the general applicability of this tool to annotate unknown compounds in a real metabolic study still awaits its validation.

An additional, MS-platform independent approach for the accurate annotation of elemental compositions is based on the use of isotopically labeled compounds. Here one can either make use of single labeled compounds and follow their intercellular metabolism (Feldberg et al., 2009) or label whole metabolomes with isotopically labeled carbon, nitrogen or sulfur (Giavalisco et al., 2008, 2009, 2011). These labeled metabolites, which show elemental composition-specific mass shifts, can be used to significantly improve annotation of elemental composition and reduce false positive rates, but also to allow a clear distinction between biological and non-biological contaminating compounds (Giavalisco et al., 2008, 2009, 2011).

Increasing the Resolution by Increasing the Separation

Eukaryotic organisms are often multicellular and highly compartmentalized. Many metabolic pathways involve reactions in more than one tissue type or subcellular compartment requiring numerous transport processes across different membrane systems. Probably the most prominent example of a well-described multi-compartment pathway in plants is the photorespiration pathway, which includes chemical reaction in chloroplasts, peroxisomes, and mitochondria (Bauwe, 2010; Maurino and Peterhansel, 2010). In addition, many other metabolic pathways might involve metabolic shuttling since the required metabolic enzymes are localized as isoenzymes in more than one subcellular compartment (Kruger and von Schaewen, 2003; Sweetlove et al., 2010). Within those pathways enzymatic activities and therefore metabolic fluxes are regulated by substrate concentration, the presence of different co-factors and intermediates, as well as allosteric activation or feedback inhibition of key enzymes (Stitt et al., 1983; Gunasekaran et al., 2004; Kruger et al., 2007; Takahashi et al., 2011). Therefore, detailed information about tissue-specific or subcellular compartmentalization of metabolites is fundamental for the understanding of metabolic networks and the regulation of metabolic fluxes. A major hurdle for the reliable determination of the spatial distribution of metabolites is their fast turnover and rapid transport between the different compartments (Flugge et al., 1980; Stitt et al., 1982; Ratcliffe and Shachar-Hill, 2005; Furumoto et al., 2011). Although analysis of a wide range of primary and secondary metabolites by GC- and LC-MS is frequently done to discover metabolic phenotypes (Saito and Matsuda, 2010), little information is available about the spatial distribution of these metabolites within different cell types. Methods have been developed to study the spatial distribution of metabolites at the tissue level, like matrix-associated laser desorption ionization mass spectrometry imaging (MALDI-MSI), desorption electrospray ionization mass spectrometry imaging (DESI-MSI), laser capture microdissection (LCM), or cell sap sampling (CSS) and on a subcellular level like fluorescence resonance energy transfer (FRET)-based nanosensor technology, NMR spectroscopy, protoplast fractionation (PF), or non-aqueous fractionation (NAF).

Getting an Overview: ‘Metabolite Analysis at the tissue and Cellular Level’

Increasing the spatial resolution is an important challenge in the field of metabolite analysis, especially in multicellular organisms. In recent years several different approaches have been undertaken to study metabolism/metabolites at a tissue and cellular level. Imaging technologies like MALDI-MSI or DESI-MSI have been successful applied to study the spatial distribution of a broad range of molecules directly from tissue or tissue sections (Kaspar et al., 2011; Muller et al., 2011). The MALDI ionization method produces intact ions due to the use of a pulsed laser beam in combination with an energy-absorbing matrix which co-crystallizes with the analyte (Figure 3D). The absorption of the laser energy leads to explosive desorption of the matrix and the analyte into the gas phase. Matrix-associated laser desorption ionization, like ESI, is one of the so-called soft ionization techniques (Zaikin and Halket, 2006). The use of different matrices allows the efficient ionization and subsequent analysis of molecules between a few hundred Da and 100 kDa (Chughtai and Heeren, 2010; Muller et al., 2011). In plants, MALDI-MSI was successfully applied to study the spatial distribution of molecules such as sugars, lipids, secondary metabolites like glucosinolates, flavonoids and alkanes in leaf tissue and tissue sections of different plants (Cha et al., 2008, 2009; Shroff et al., 2008; Jun et al., 2010; Vrkoslav et al., 2010). For example, the analysis of glucosinolate distribution within Arabidopsis thaliana using MALDI-MSI revealed a non-uniform distribution throughout the leaf tissue (Vrkoslav et al., 2010). The accumulation of glucosinolates close to the leaf margin and the middle vein strongly influences the feeding preference of Helicoverpa armigera larvae and might play a key role as a first barrier against chewing herbivores (Vrkoslav et al., 2010). The MALDI-MSI approach has classically been limited by its requirement for a high vacuum for the ionization of the analyzed molecules, making this technology incompatible with in vivo analysis. However, more recently Coello et al. (2010) have described the use of a near-infrared laser imaging MS approach obviating the need for high vacuum. Even though the presented approach demonstrates the potential of this technique for in vivo imaging studies, it still lacks precision and sensitivity to fully analyze cell-specific or even subcellular metabolic distributions.

Figure 3.

 Methods for studying metabolism with varying degrees of spatial resolution.
Cell fractionation (A) is the classical method for studying metabolism on a subcellular level. Two different methods for subcellular fractionation have been used. Protoplast fractionation (PF) is based on the fractionation of protoplast by centrifugation through a nylon net (1). Cellular content is released (2) and chloroplasts are separated by centrifugation through a silicon oil layer (3). Metabolism within isolated chloroplast is inhibited by perchloric acid (4). For non-aqueous fractionation (NAF) freeze-dried tissue homogenate is fractionated on a non-aqueous density gradient. Using NAF, at least three compartments, namely chloroplasts (a), the cytosol (b), and the vacuole (c), can be unambiguously delineated. To study metabolism on a single-cell level, microcapillary approaches (B) have been used. Sap from a single cell is sampled and analyzed by metabolite profiling. Fluorescence resonance energy transfer (FRET)-based metabolite nanosensors (C) are a promising new technology to study spatial and time-resolved changes in metabolite concentrations in vivo and with a high degree of resolution. Analysis of metabolite composition on a whole tissue level is realized by desorption electrospray ionization mass spectrometry imaging (DESI-MSI) and matrix-associated laser desorption ionization mass spectrometry imaging (MALDI-MSI) (D) by generating spatial-resolved images. Nuclear magnetic resonance (NMR) technology (E) facilitates the analysis of metabolites in non-invasive ways at the cellular and subcellular level. Microdissection (F) allows isolation of specific cell types from a tissue section. Within the isolated cells a broad range of molecules can be analyzed.

The development of atmospheric pressure ionization methods like the DESI represents an additional method with potential for imaging MS in a real in vivo imaging experiment (Costa and Cooks, 2008). For DESI, charged solvent droplets are produced in an electric field and accelerated by a nitrogen stream (Figure 3D). The impact of the droplet on the sample surface results first in a liquid film which leads to the extraction of metabolites. Further droplet collision leads to the generation of analyte-containing microdroplets which became conventionally ionized by ESI. Using DESI-MSI, several different metabolite classes such as sugars, amino acids, fatty acids, lipids and alkaloids could be detected in leaf samples of different plant species, indicating the sensitivity of this method (Jackson et al., 2009; Liu et al., 2011; Muller et al., 2011). The DESI-MSI and MALDI-MSI techniques are so far not applicable to single-cell analysis as the spot size of the laser beam and the nitrogen stream has to be smaller than a single cell. Recently, a reduction in the laser spot size for MALDI-MSI from 150 μm to below 20 μm for analysis at the single cell level has been achieved. However, thus far this reduced laser spot size technique has not yet been applied to plant tissues (Holle et al., 2006; Grassl et al., 2011).

Another approach involving sampling of specific tissue or cell types is LCM (Figure 3F). This approach has already been used in plant systems to analyze the metabolite composition in certain cell types (Kerk et al., 2003; Nakazono et al., 2003; Schad et al., 2005). In general, histological sections are prepared and fixed. The cells of interest have to be identified by either histological staining, reporter gene expression, or morphological properties. Therefore they are selected under a microscope, isolated by a laser beam and collected in a tube by a laser pressure pulse (Nelson et al., 2006). The collected cells can be used for different analytical methods, such as transcript, protein, or metabolite analysis (Westphal et al., 2002; Wittliff and Erlander, 2002; Kerk et al., 2003; Nakazono et al., 2003; Schad et al., 2005). However, one disadvantage of this method is the sample preparation, because histological fixation is required which might affect molecule composition within the selected sections (Nelson et al., 2006). Furthermore, and probably even more important, LCM techniques are very time intensive with respect to the preparation of enough cells for later analysis and are thus definitely not applicable in a high-throughput mode.

Beside LCM, other methods have been used to study metabolite composition at the tissue-specific level. In some special cases such as trichome cels a cellular preparation using classical fractionation techniques is possible and can be used in order to determine the metabolic composition of these special cell types (McDowell et al., 2011; Weinhold and Baldwin, 2011).

In cases where these fractionation methods do not work, microcapillary methods can be used to collected cell sap (Figure 3B), mainly from the vacuole of epidermis cells. Cell sap sampling in combination with capillary electrophoresis (CE) or GC-MS was used to detect a number of different metabolites like organic acids, amino acids, sugars, inorganic anions, and secondary metabolites (Lochmann et al., 1998, 2001; Ebert et al., 2010).

Due to technical limitations the CSS methods are generally not suitable for the analysis of subcellular compartments like the chloroplasts or mitochondria. Cell sap sampling has only been applied in a very special case for analysis of a subcellular compartment, i.e. the vacuole of a single internodal cell of the giant alga Chara australis (Oikawa et al., 2011). After the turgor of the 20-cm Chara australis cell is lost, the sap of its large vacuole can be easily harvested and either directly analyzed or further metabolically fractionated (Oikawa et al., 2011).

Zooming in: large scale analysis at the Subcellular Level

As mentioned above, next to mass spectrometry, NMR spectroscopy is the second main method for metabolite profiling in plants (Figure 3E). Nuclear magnetic resonance is based on the detection of isotopes with non-zero nuclear magnetic moments like, for example, 1H, 13C, 14N, 15N, or 31P (Ratcliffe and Shachar-Hill, 2005; Kim et al., 2011). As those nuclei naturally occur – or can be incorporated into plants – NMR allows us to determine the in vivo metabolite composition in cell suspension cultures, tissues, and even whole plants (Gout et al., 1993, 2000; Scheenen et al., 2000; Libourel et al., 2006; Kim et al., 2011). For the analysis of metabolites on a subcellular level, a compartment-specific NMR signal for the metabolite of interest is required. This can be achieved because the chemical shift of many metabolites depends on the pH (Bligny and Douce, 2001). Using 13C- and 31P-NMR, the compartmentalization of carbohydrates and amino acids within Acer pseudoplatanus cells was analyzed (Aubert et al., 1998; Gout et al., 2011). This study showed how sugar starvation in A. pseudoplatanus cells leads to a very rapid (<2.5 min) response characterized by a dramatic change in phosphorylated intermediates. Irrespective of this successful example, the use of NMR in a non-invasive way still has several limitations because it is restricted to metabolites which exhibit strong NMR signals and more importantly to metabolites which display a chemical shift in dependence on the subcellular compartment. This is true for phosphorylated compounds due to the pH dependence of the chemical shift of the phosphate atom; however, it is not generally true. In addition, due to the complexity of the analyte composition of a living cell, in vivo NMR spectra often have to deal with poor signal to noise ratios and resolution, which limits spectral interpretation (Ratcliffe and Shachar-Hill, 2001; Kim et al., 2011).

A more accurate and sensitive technique to study metabolite compartmentalization is the use of FRET-based nanosensors (Figure 3C). These genetically encoded molecular nanosensors, with their prototype being developed for detecting calcium at nanomolar concentrations (Persechini et al., 1997), represent a new and promising technology to follow the spatial and temporal metabolic changes on the cellular and even subcellular level in living cells (Fehr et al., 2005; Lalonde et al., 2005). However, they do suffer from the severe limitation that each metabolite requires its own sensor protein for detection, thus making it essentially not applicable for true metabolomics.

Historically, the metabolite content in purified organelles has been studied following fractionation of leaf cell protoplasts (Figure 3A; Robinson and Walker, 1980). Protoplast fractionation in combination with enzymatic determination of metabolites has been widely used to quantify a subset of metabolites like adenylates, phosphorylated sugars and Calvin cycle intermediates in different compartments (Robinson and Walker, 1980; Wirtz et al., 1980; Lilley et al., 1982; Stitt et al., 1982; Gardeström, 1993). Besides the isolation of chloroplast and mitochondria, protoplast fractionation has also been used for the isolation of intact vacuoles (Martinoia et al., 1991; Ramos et al., 2011; Tohge et al., 2011).

Although isolation of organelles after protoplast fractionation represents an excellent method for purifying the organelles, this approach must be taken with great caution because the procedure as a rule takes several hours and includes several centrifugation steps, thus presenting a major disturbance of the physiology and thus the biochemistry of the system. In consequence changes in the composition and compartmentalization of metabolites due to the fractionation procedure are very likely, leading to a partly artificial representation of the subcellular concentrations of metabolites (Robinson and Walker, 1980).

In conclusion many of the techniques discussed above allow the reliable and partly in vivo analysis of a small subset of metabolites. However, with the goal of comprehensive metabolite coverage including cellular and subcellular distributions that faithfully reflect in vivo conditions, most of the above-described methods suffer from one shortcoming or another.

Non-Aqueous Fractionation (NAF): A Method Allowing the Application of Metabolomics Technologies to Subcellular Compartments Determined by Continuous Distributions

Non-aqueous fractionation (NAF) is the most widely used method for studying metabolite pool sizes at a subcellular level in plants (Gerhardt and Heldt, 1984; Dancer et al., 1990; Riens et al., 1991; Winter et al., 1992; Farre et al., 2001; Fettke et al., 2005; Krueger et al., 2011). Instead of purifying intact organelles, the NAF method is based on the enrichment of compartments within a continuous non-aqueous density gradient (Figure 4). The main advantage with respect to analyzing metabolism is the fast quenching of metabolic processes to avoid reallocation and conversion of metabolites. As NAF results in a continuous compartment distribution computer-based evaluation of the obtained data is important for obtaining reliable results (Krueger et al., 2011; Klie et al., 2011). If the volumes for the subcellular compartments, the total tissue concentration, and the subcellular distribution of a certain metabolite are known, the exact subcellular concentration of this metabolite can be calculated (Winter et al., 1992). Information about the subcellular localization of different metabolites obtained from NAF studies represents an important step forward for the understanding of plant metabolism. For example, by using NAF technology the prevalent cytosolic localization of pyrophosphate was discovered, whereas it could be shown that the pyrophosphatase is highly active in chloroplasts (Weiner et al., 1987). Non-aqueous fractionation in combination with field flow fractionation (FFF) was applied to analyze the subcellular localization of a recently identified water-soluble heteroglucan (Fettke et al., 2005). In recent years NAF technology has been linked with true metabolomics studies allowing the subcellular localization of a large number of metabolites to be analyzed in parallel (Farre et al., 2001; Benkeblia et al., 2007; Krueger et al., 2011). The evaluation of such data sets using biostatistical and bioinformatic tools suggests the presence of previously unidentified plant compartments within the NAF gradient and shows that the resolution of this method is yet not fully explored (Krueger et al., 2011). Although NAF does not allow purification of intact organelles, and the evaluation of the obtained data is rather complex, NAF is still the method of choice and its full potential has not yet been reached. For example, metabolic flux analysis often ignores metabolite compartmentalization, leading to misinterpretation of the obtained results (Kruger et al., 2007). As NAF allows direct quenching of metabolism by snap-freezing in liquid nitrogen, the combination of NAF with metabolic flux analysis using 13C labeled CO2 is a very attractive approach for the future. However, the complexity of the obtained data from such an approach might require new methods for the computation, interpretation, and visualization of the data as evaluated by Klie and co-workers (Klie et al., 2011).

Figure 4.

 The entire non-aqueous fractionation (NAF) procedure can be divided into experimental- and computational-driven analyses.
The experimental part starts with the fast quenching of metabolism by snap-freezing of plant material in liquid nitrogen. The frozen leaf material is homogenized and freeze-dried in a pre-cooled lyophilizer. The dried powder is resuspended in a non-aqueous mixture of tetrachloroethylene (TCE)/heptane and sonicated on ice. The obtained suspension is poured through a nylon sieve and loaded on a linear density gradient. After 50-min centrifugation at 5000 g and 4°C equilibrium distribution of the fractionated particles is reached. Fractions are taken from the top of the gradient and separated into 10 1-ml aliquots. For the analysis of marker activity and metabolite profiling, aliquots of each fraction are dried and then extracted in the respective buffer or solvent by strong vortexing or shaking in a pre-cooled Retsch mill. The computational phase of the work includes the validation, classification, visualization, and interpretation of the obtained data (cf. Klie et al., 2011).


Metabolic analyses have become an integral part of plant science within recent years. Since the developed technologies have become more and more sophisticated, and often also specific for particular questions, the gap between the width (number of compounds analyzed) and depth (quality of annotated compounds) within metabolomics studies seems to be expanding. As a consequence the key challenges in metabolomics, as in all ‘omics’ technologies, will be to establish and combine methods and strategies which allow us to increase the number of measured and annotated compounds – this means qualitatively, quantitatively and spatially resolved without losing sample throughput and quality controls. In particular, quality control and standardization of the applied methods are necessary to allow cross-laboratory comparisons.

Although in principle many technical solutions are already available for these challenges none of them is yet fully applicable to high-throughput approaches which will become an important part of plant science in the future.


Änne Eckhardt and Gudrun Wolter are kindly acknowledged for their patient help and support in establishing our metabolomics facility. Further we would like to kindly thank Professor Leslie Sieburth for thoroughly proof reading and commenting on the manuscript. Finally we would also acknowledge the Max Planck Society and the University of Cologne for their financial support.