- Top of page
- Materials and Methods
- Results and Discussion
- Supporting Information
Substantial progress has been made over recent years in the parallel analysis of transcripts, gene products and metabolites, with a major impact on plant science (Yuan et al., 2008; Saito & Matsuda, 2010; Kaufmann et al., 2011). At present, the most well-developed platform targets mRNA, whereas those aimed at the analysis of proteins and metabolites remain more challenging, largely because the chemical composition of these analytes is much more heterogeneous. Nonetheless, these two classes of molecules are important, as they carry information which cannot be addressed by transcript analysis – for example, many proteins are post-translationally modified and, for the cell, this represents a critical component of its control machinery. A particular priority area is to develop ways to improve the spatial resolution of measurements, as higher organisms, including plants, comprise an array of functionally specialized tissues within every organ (in plants, the root, leaf or seed). Furthermore, the molecular content of many tissues responds to both development and cues from the external environment.
However, methods for untargeted comprehensive analyses of specific tissues are lacking. The deployment of pressure probe glass capillaries or laser capture microdissection has been pioneered for the isolation of specific tissues (Amantonico et al., 2010), and tissue-specific transcriptome profiling can be facilitated by either the incorporation of tagged ribosomal proteins (Zanetti et al., 2005; Mustroph et al., 2009) or the labelling of particular protoplast subpopulations to allow their sorting (Dinneny et al., 2008; Dinneny, 2010; Tsukagoshi et al., 2010). The latter approach may also be feasible for proteomic or metabolomic analyses, although it is unclear to what extent the protoplast preparation process per se may induce cellular reactions likely to alter protein and metabolite composition.
Innovative instrumentation and sample preparation protocols have facilitated matrix-assisted laser desorption/ionization mass spectrometric imaging (MALDI-MSI) of large and varied molecules at both the whole-tissue and single-cell level (Caprioli et al., 1997; Stoeckli et al., 2001; Zimmerman et al., 2008). The versatility of the approach has been well demonstrated in the clinical field, in which, for example, it has been used to define certain complex molecular interactions occurring in various carcinomas (Chaurand et al., 2004; Schwartz et al., 2005; Cornett et al., 2006). Several excellent reviews describing the applications of MALDI-MSI have been published in recent years (Rohner et al., 2005; Burnum et al., 2008; Goodwin et al., 2008; van Hove et al., 2010; Svatoš, 2010; Zaima et al., 2010b), but its potential in plant biology has yet to be widely explored, even though the spatial distribution of specific molecules is a critical component of compartmentalized cellular processes. Among the few plant-based examples to date, as reviewed in Kaspar et al. (2011), are, for example, a study of the distribution of glucosinolates following insect feeding on Arabidopsis thaliana leaves (Shroff et al., 2008) and a description of epicuticular lipid metabolites on the surface of A. thaliana flower and root (Jun et al., 2010). Only rare use has been made of MALDI-MSI to analyse sectioned plant material, but it has been effective, for example, in demonstrating the uneven distribution of specific amino acids, sugars and phosphorylated metabolites within the wheat grain (Burrell et al., 2007), and in being able to distinguish between various organ and tissue types present in the rice grain (Zaima et al., 2010a). The MALDI-MSI approach has been shown to be as sensitive as the well-established liquid chromatography-mass spectrometry (LC-MS) method for the detection of wheat stem oligosaccharides (Robinson et al., 2006), and has been successfully deployed to describe the distribution of both γ-aminobutyric acid in the seed of eggplant (Goto-Inoue et al., 2010) and pesticides in soybean (Mullen et al., 2005) and sunflower (Anderson et al., 2009). No comparable analysis of polypeptide distribution has been published to date, but the potential of the technology has been demonstrated by the identification of a 12-residue oligopeptide in the stem of A. thaliana (Kondo et al., 2006).
The quality of MALDI-MSI is strongly influenced by the complexity of the target, especially where there are structural differences between distinct tissue types in the sample. Careful sample treatment is essential to ensure signal quality and to avoid lateral displacement of the analytes (Schwartz & Caprioli, 2010), as artefacts can arise at any stage between sample collection and MSI analysis (Goodwin et al., 2008). Here, we present a viable strategy for the MALDI-MSI analysis of metabolites in cryodissected barley grains and tobacco roots. Although much has been learned regarding grain development in barley using laser capture microdissection (Thiel et al., 2008), in general, the sample size is too small for certain analytical procedures. Instead, we have focused on an LC-MS-based label-free quantitative approach to underpin a proteomic description of the content of the internal structures of the developing grains (Kaspar et al., 2010). Furthermore, we aim to investigate the metabolite composition in several tissue types. MSI measurements have reproducibly displayed either tissue-specific or co-localized metabolite expression patterns. As a contrast, we have also studied the root, an organ heavily involved in plant–environment interactions (e.g. nutrient transport and sensing). In addition, roots differ totally in tissue composition from seed material with regard to, for example, the intercellular spaces of cortex tissue, high water content and lignin comprising central tissues.
Results and Discussion
- Top of page
- Materials and Methods
- Results and Discussion
- Supporting Information
Both the barley grain and the tobacco root comprise a mixture of tissue types (e.g. parenchymal, collenchymal and vascular tissues), so that sample preparation had to preserve, as closely as possible, the three-dimensional structure of the tissues. During the early stage of development, the barley grain contains a high proportion of water, making sample preparation fairly straightforward; however, more mature grains contain much less free water, so that their sectioning requires particular attention. Specific optimization targets were tissue section preparation, choice of matrix and matrix application, and MALDI-MSI measurement parameters and data analysis.
Tissue section preparation
Thin tissue slices were cut with a conventional cryotome, but the preparation of sections cannot follow the standard histological procedures, for example, slice preparation for immunostaining or dye staining, because embedding in polymeric compounds, such as synthetic resin (HM20), or the use of OCT medium would result in ionization suppression of the analytes during MS measurements. OCT is a polymeric alcohol composed of polyvinyl alcohol and polyethylene glycol (PEG), and is widely used for the embedding of samples before cryosectioning processes (Turbett & Sellner, 1997). Therefore, samples were fixed in the desired orientation with a small droplet of fixation medium on the cutting block of the instrument. For fixation, we tested two compounds: OCT and water (ice). A series of cross-sections was made from both the main and lateral roots of tobacco, and from barley grains harvested at various DAP (Fig. 1a,b). Because it was impossible to avoid contamination with OCT in barley grains younger than 6 DAP and in longitudinal tobacco root sections, water (ice) was used as the fixation medium. However, the use of water resulted in a considerable degree of diffusion of analyte during thaw mounting, visible as a smear around the section (Fig. 1c,d). There are also other plant organs, such as leaves, small seeds and root tips, which have proven to be difficult to section. The film sectioning method (Kawamoto, 2003) generates significantly less distortion and dislocation than occurs in the absence of any supporting material, and has been successfully used for the analysis of rice grains (Zaima et al., 2010a). However, this approach proved to be impractical for the very small samples used.
Figure 1. Tissue sectioning. (a) Representative optimal cutting temperature (OCT) medium-fixed 30-μm longitudinal and cross-sections cut from various regions of a 7-d after pollination (DAP) developing barley (Hordeum vulgare) grain. (b) Water-fixed 55-μm longitudinal section of a lateral root, and cross-sections from a main (55 μm) and a lateral (40 μm) tobacco (Nicotiana tabacum) root. (c) 30-μm cross-section from the scar region of a 3-DAP developing barley grain (water-embedded tissue). (d) 55-μm cross-section from a lateral tobacco root (water-embedded tissue). Delocalized metabolites are visible as a smear surrounding the sections after thaw mounting on indium tin oxide-coated slides. Bars, 1 mm.
Download figure to PowerPoint
The optimal thickness range for tobacco root sections was 35–55 μm. Fig. 1(b) shows a 55-μm longitudinal section of a lateral root, and cross-sections from a main (55 μm) and lateral (40 μm) root, from 8-wk-old tobacco plants. These sections are thicker than those commonly used for MSI experiments to avoid disruption caused by the large intercellular spaces in the parenchymal cortex. The influence of root section thickness on MALDI-MSI output is depicted in Fig. 2(a), which shows images of four molecular adduct ions from 35-, 45- and 55-μm cross-sections of a tobacco lateral root. A slice thickness above 45 μm led to a notable deterioration in measurement sensitivity, particularly when analysing distribution patterns in the central root. The loss of MS signal in this root region probably reflects matrix absorption effects, as the signal intensities for root cortex were rather independent of sample thickness.
Figure 2. The influence of section thickness. Ion images obtained from matrix-assisted laser desorption/ionization mass spectrometric imaging (MALDI-MSI) of cross-sections of three thicknesses of a tobacco (Nicotiana tabacum) lateral root (a) and a 9-d after pollination (DAP) developing barley (Hordeum vulgare) grain (b). Measurements were conducted at a resolution of 30 μm (tobacco root) and 25 μm (barley grain). Some signal intensities proved to be dependent on the section thickness.
Download figure to PowerPoint
For young (1–7 DAP) barley grains, 20-μm slices were effective, but, as the grain developed, the accumulating starch (associated with a loss of water content) forced the section thickness to be increased to 30–40 μm (Fig. 1a). The effect of section thickness on the resulting MALDI-MSI is shown in Fig. 2(b), which presents ion images of four molecular masses from a 7-DAP grain. The signal intensities from 20- and 30-μm sections were consistent, but the detection sensitivity was reduced in 40-μm sections, because of the higher matrix absorption by the tissue. Therefore, the section preparation of barley grains was set to 30 μm. Unlike in the root, there were less tissue type-dependent matrix effects observed for barley grains and the less fragile structure of the grain allowed very thin sections to be used. Sample preparation is particularly demanding when an array of highly differentiated tissues is present. The use of nonembedded frozen sections requires a high degree of technical proficiency. Conventionally, the majority of sections used for MSI are 10–20 μm thick (Goodwin et al., 2008; van Hove et al., 2010), a thickness which provides a reasonable compromise between MSI performance and practicality (Zaima et al., 2010b), particularly when a large number of samples are to be analysed. However, the optimal slice thickness is very much dependent on the kind of tissue under investigation.
Choice of matrix and matrix application
Depending on the alterations in ionizing capacity of the different substrate groups to be analysed, various matrix compounds have been proposed (Svatoš, 2010; Kaspar et al., 2011). We chose to test HCCA, DHB, DMAN and DTCB. DMAN, which belongs to the so-called ‘proton sponges’, is a new matrix introduced by Shroff & Svatoš (2009). According to its strong basicity, this matrix is useful for the analysis of anions. Of the tested matrices, DHB gave the most reproducible results, given the choice of application strategy, matrix-derived background (Fig. S1) and MALDI-MSI reproducibility (Fig. 5). One of the most challenging aspects of MSI relates to the reproducibility in the application of the matrix compound (Gustafsson et al., 2008). The most widely used deposition techniques are spraying with a simple airbrush and the use of a dedicated instrument to obtain vibrational vaporization. Here, the latter approach was adopted, based on an instrument able to apply a series of matrix layers with an incubation and drying period between each deposition step. Reproducible layers were obtained for HCCA, DCTB and DHB, but the DMAN matrix could not be deposited by vibrational vaporization, as its chemistry hindered adequate crystal formation. ImagePrep spray plates can normally be used for several cycles of matrix deposition, but we noted that, after a number of cycles, larger droplets started to form, and these led to a loss in the achievable spatial resolution. Fine layers of matrix crystals could still be obtained, but this required manual intervention during the spraying process. Thus, for optimal reproducibility, fresh spray plates should be used. An alternative approach is to use either microchip spotting technology (< 20-μm spatial resolution) or sublimation (1-μm resolution), compared with the 10–100-μm resolution achieved by spray application (Kaletaşet al., 2009; Agar et al., 2010).
A critical consideration in the choice of matrix is the frequency of matrix-derived MALDI-MSI signals caused by matrix fragmentation and cluster formation. To distinguish matrix- and sample-derived m/z values, every MSI experiment includes the measurement of a blank region with matrix only. During data analysis, sum mass spectra of both measurement regions are then overlaid and compared. The interference of DHB with the signals generated from a barley grain sample is illustrated in Fig. 3, which represents a sum mass spectrum from one MSI experiment in which the matrix-derived and barley grain signals in two specific mass regions are shown in detail. Comparison of the mass signals generated from the barley grain section and blank region revealed m/z 758.6 and m/z 851.5 as being potential grain tissue-related signals, but also showed interference with DHB-related signals for m/z 758.6. Another example of matrix interference is presented in Fig. S2, which shows the mass spectra and corresponding ion images for two particular m/z values. These peaks can only be assigned to the tissue by their distinct distribution pattern in the sample or by comparison of the isotope pattern. Many of the signals in the low mass region (m/z < 400) could be associated with DHB fragmentation and cluster formation, and so these needed to be subtracted from MSI sample sets as part of the data evaluation process. The mass spectra obtained from HCCA, DHB, DMAN and DTCB were compared in both positive and negative ionization mode. In the positive ionization mode, DHB and DMAN produced the fewest peaks, whereas the HCCA spectrum was well populated at m/z < 400 and, similarly for DCTB, in the range m/z 400–800. In the negative ionization mode, few peaks were generated by DHB, and even fewer by DMAN. However, more complex spectra were observed for HCCA and DCTB (Fig. S1). Other ionization-enhancing reagents, such as silver or gold nanoparticles (Hua et al., 2007; Wu et al., 2009), colloids and graphite have been introduced recently (Svatoš, 2010); these appear to have advantages in the context of lipid detection, and also tend to produce fewer and less intense signals, thereby improving the spatial resolution. However, the use of these matrices requires the adaptation of application strategies and the consideration of target compound classes.
Figure 3. The assessment of the background signal generated by the matrix. A representative sum spectrum from one matrix-assisted laser desorption/ionization mass spectrometric imaging (MALDI-MSI) run of a 7-d after pollination (DAP) developing barley (Hordeum vulgare) grain, together with the detail of the mass spectrum for the matrix alone and for the barley grain alone at m/z 759 and m/z 852. au, arbitrary units.
Download figure to PowerPoint
MALDI-MSI parameters and data analysis
The size of the barley grain longitudinal sections increases from a length of c. 2 mm and a diameter of c. 1 mm at 2 DAP to a length of > 1 cm and a diameter of c. 3 mm by 14 DAP. Thus, depending on the sample size, we must adapt the measurement resolutions to extract biologically relevant information, but also to limit the size of the dataset. Metabolite distributions in barley grains appear at the tissue level rather than at the single-cell level, and cellularization levels vary strongly from very young grains (1–6 DAP) to grains in the seed filling or mature stage (7–20 DAP). Some tissues require a higher resolution measurement than others – for example, the aleurone layer is only two to three cell layers thick (Garrett et al., 2011). As a result, the resolution adopted for 1–6-DAP grains was 10–20 μm and, for 7–20-DAP grains, 20–35 μm (for details, see Table S1). It should be noted that most of the current literature describing MALDI-MSI analyses of plant material describes spatial resolutions in the range 100–200 μm (see review by Kaspar et al., 2011). The exceptions are an investigation of eggplant seed with a resolution of 25 μm (Goto-Inoue et al., 2010), and a study of the A. thaliana flower, stem and root with measurement resolutions of 12, 25 and 50 μm (Jun et al., 2010). The highest spatial resolution reported so far for plant material (A. thaliana and Hypericum species), 10 μm, was achieved by matrix-free UV-laser ionization mass spectrometry for the analysis of UV-absorbing secondary metabolite distributions (Hoelscher et al., 2009).
The reproducibility of MALDI-MSI results is reflected by the same distribution pattern of m/z values through independent experiments. In our experiments, MALDI-MSI-derived metabolite distributions were verified through both biological replication and by comparing the outcomes from longitudinal and cross-sections of different seed parts. At least three replicates were conducted to derive spatial and temporal distribution patterns. Thus, it is possible to describe the specific arrangement of molecules both histologically and within a developmental context. In addition, MALDI-MSI spectra of extracts from manually dissected barley grain tissues revealed m/z values with the same distribution pattern as detected by MALDI MSI experiments. Fig. 4 presents a set of ion images of longitudinal and cross-sections of a 7-DAP barley grain. The multi-ion image on the left is from a longitudinal section, the corresponding images from cross-sections are in the centre and the various spectra are on the right. The abundance of the m/z 773.4 product was specific to the scar region of the developing grain (Fig. 4, top), whereas the region surrounding the developing endosperm (in both the longitudinal and cross-sections) featured the m/z 593.3 product (Fig. 4, centre), and the m/z 816.5 analyte was largely localized within the region of the developing embryo (Fig. 4, bottom). Fig. 5 presents ion images for two products of different m/z showing their tissue-specific localization (593.3 and 520.4) across three replicates (top part of the figure); the bottom panel illustrates examples for two co-localized molecular masses (m/z 104.1 and 820.5) across three replicates. The data demonstrate the feasibility of generating reproducible, high-resolution MALDI-MSI profiles from thin barley grain sections, once sample preparation, matrix application and MS measurement have all been optimized.
Figure 4. Reproducibility of matrix-assisted laser desorption/ionization mass spectrometric imaging (MALDI-MSI) across tissues, showing the spectra obtained from 30-μm sections of a 7-d after pollination (DAP) developing barley (Hordeum vulgare) grain. The multi-ion image shows molecular masses m/z 773.4 (blue), m/z 593.3 (red) and m/z 816.5 (green) from a longitudinal section (left), the corresponding selected-ion images (centre) and the respective mass spectra with molecular masses indicated by a dotted line (right). The resolution was 25 μm for the longitudinal section, the cross-section of the scar and the central region, and 20 μm for the embryo cross-section.
Download figure to PowerPoint
Figure 5. Reproducibility of matrix-assisted laser desorption/ionization mass spectrometric imaging (MALDI-MSI) across biological individuals. Ion images from cross-sections of the scar region of three 7-d after pollination (DAP) developing barley (Hordeum vulgare) grains. Measurements were conducted at a resolution of 20 μm.
Download figure to PowerPoint
Once the MS data collection is completed, processing and visualization methods for the evaluation of intensity distributions and statistical comparison are applied. FlexImaging software was used for the evaluation of a sum spectrum of the complete dataset, and to visualize multi-ion and selected-ion images within tissue sections. Baseline subtraction was included in the data acquisition, but additional subtraction of the matrix-derived signal was not possible. As a result, the tissue-derived sum spectra had to be manually compared with the spectra derived from the matrix alone (an example is given in Fig. 3). A number of MALDI imaging software packages (both open-source and commercial) offer data processing algorithms and statistical analysis (Kaspar et al., 2011), but software-driven subtraction of the matrix background in the low m/z range is not yet available. Furthermore, current software packages designed to statistically analyse MALDI-MSI datasets are unable to handle large datasets, for which prior data reduction is required. Depending on the number of single spectra per MALDI-MSI dataset, this type of data reduction implies a loss of resolution. The number of spectra routinely acquired in our experiments varied from 2500 to 18 000 per tissue section (Table S1). This scale of data is beyond the analytical capacity of current software, underlining the need for development in this direction, as exemplified by BioMap analysis software (http://www.maldi-msi.org/).
Great effort is needed for the identification of compounds detected with MALDI-MSI. To reveal almost exact m/z values, we carried out manual measurements on the tissue surface. Several online databases (http://metlin.scripps.edu/), KNApSack (http://kanaya.naist.jp/KNApSAcK/) and Lipidomics Gateway (http://www.lipidmaps.org/), as well as literature data, were used to classify products on the basis of their measured m/z. Additional information was obtained from MS/MS measurements. Diagnostic fragment ions allowed an assignment of precursor masses as molecular species of PC and LPC. These phospholipid classes are expected to produce fragment ions of m/z 104 and 184 and a neutral loss of m/z 59 (Xu et al., 2009; Garrett et al., 2011; Murphy & Gaskell, 2011). The fragmentation is based on CID, a technique designed to control the fragmentation of selected precursor ions. For example, CID of molecular ion m/z 520.4 from barley grain tissue generated fragment ions of m/z 104 and 184. A comparison with the MS/MS spectrum of m/z 520.4 from LPC standard solution showed the same fragment ions (Fig. 6). This protonated ion was identified as an 18 : 2 LPC species with linoleic acid as unsaturated fatty acid residue.
Figure 6. Comparison of collision-induced dissociation (CID) spectra of molecular adduct ion m/z 520.4 from l-α-lysophosphatidylcholine (LPC) standard solution (1% in chloroform) and barley (Hordeum vulgare) grain section (pericarp region). au, arbitrary units.
Download figure to PowerPoint
However, as the mass accuracy of MALDI TOF/TOF instrumentation is insufficient at present to determine the molecular formulae of selected compounds, a targeted analysis based on gas chromatography-mass spectrometry (GC-MS) and/or LC-MS is still required to achieve identification. Improvements in analytical software are still required to enhance the mass accuracy of the sum spectra and to avoid peak shifts during the simultaneous acquisition of different regions of interest. In our protocol, the instrument was calibrated in the positive ionization mode with PEG200 + 600 before the start of an automatic acquisition, but a calibration in between individual measurements is currently not supported by the vendor’s software. The development of software solutions for external calibration during automatic MALDI-MSI runs remains a strong priority.
The data presented here have demonstrated that MALDI-MSI measurements can reliably illuminate the spatial distribution of small molecules in barley grain and tobacco root sections, once sample preparation protocols have been optimized. The metabolite and protein composition of an organ (and of the individual tissues within) depends on the organ’s function. Sampling strategies producing a range of spatial resolutions have been trialled in plants, including manual and laser capture-assisted dissection in conjunction with micropipetting approaches (Table 1). More recently, protoplast labelling and ribosome tagging have been proposed as a means of obtaining a nontargeted analysis of the content of specific tissues (Dinneny et al., 2008; Mustroph et al., 2009). An alternative platform, involving liquid extraction surface analysis (LESA) coupled to MS detection, has been suggested for the direct analysis of tissues. MALDI-MSI has the particular advantage that the distribution of compounds can be assessed from direct measurements of the sample at high spatial resolution. An appropriate choice of experimental protocol, particularly for sample preparation, is critical. However, once established, MALDI-MSI can provide highly selective, rapid and parallel acquisition of many compounds. Thus, MALDI-MSI should find applications in various fields of plant research, including development, host–pathogen interactions, abiotic responses, nutrition and symbiotic interactions, and varietal-specific gene expression. The relevance of development lies in the changes affecting tissues during the process, as these affect the distribution of cellular components. In the interaction between the host and its pathogens, the host’s production of antimicrobials is a key part of host defence. Similarly, the abiotic response, including nutrient stress (e.g. inadequate nitrogen or excess trace elements), involves the induction of a range of molecules in a tissue-specific manner. Finally, the establishment of the effect of single genes is often attempted by making comparisons between isogenic lines, or between mutants (or transformants) and the wild-type, so that the capacity to identify primary or secondary gene products is becoming increasingly important. As observed previously for the barley grain (Bollenbeck et al., 2009; Kaspar et al., 2011), the identification of metabolites or proteins and the confirmation of tentatively assigned m/z signals constitute major challenges for MSI. We suggest that improvements in both sample preparation strategy and the analytical platform (both for spectrum acquisition and post-acquisition analysis) will enhance the relevance of MALDI-MSI technology in plant research. The implementation of tandem mass spectrometry will encourage MALDI-MSI applications in plant research by enabling the identification of metabolites and, via on-tissue digestion, N-terminal peptide derivatization and CID tandem MS, by facilitating the identification of polypeptides. Much of our current effort is focused on data integration. The most widely applied approach is tissue-specific characterization, which requires the integration of the data into existing histological models (Bollenbeck et al., 2009) or into metabolic networks in the case of transcript and metabolite profiling (Thiel et al., 2009). An understanding of the molecular events unfolding during development will be much aided by a knowledge of the localization and abundance of key molecules present in specific tissues. Despite the growing impact of molecular imaging, MALDI-MSI technology is likely to complement, rather than to replace, targeted profiling approaches typified by forward functional genomics.
Table 1. An overview of sampling strategies used to obtain spatially resolved profiling in plant materials
|Sampling technique||Resolution||Analysis targets||Analysis approach||Selected references|
|Manual dissection||Organs, tissues||RNA, metabolites, proteins||Targeted/untargeted||Amantonico et al. (2010)|
|LCM1||Tissues, cells||RNA, metabolites, proteins||Targeted/untargeted||Thiel et al. (2008); Kaspar et al. (2010)|
|Micropipetting||Cells||RNA, metabolites, proteins||Targeted/untargeted||Amantonico et al. (2010)|
|Protoplast sorting||Tissues, cells||RNA, metabolites, proteins||Targeted/untargeted||Dinneny et al. (2008); Tsukagoshi et al. (2010)|
|Tagged ribosomal proteins||Tissues, cells||RNA||Untargeted||Zanetti et al. (2005); Mustroph et al. (2009)|
|Cryosectioning2||Tissues, cells||Metabolites, proteins||Targeted/untargeted||Burrell et al. (2007); Zaima et al. (2010a)|
|Direct surface analysis3||Surfaces, tissues||Metabolites, proteins||Untargeted||Shroff et al. (2008); Jun et al. (2010)|