Simultaneous analysis of metabolites in potato tuber by gas chromatography–mass spectrometry

Authors


For correspondence (fax +49 331567 8201; e-mail roessner@mpimp-golm.mpg.de).

Summary

A new method is presented in which gas chromatography coupled to mass spectrometry (GC–MS) allows the quantitative and qualitative detection of more than 150 compounds within a potato tuber, in a highly sensitive and specific manner. In contrast to other methods developed for metabolite analysis in plant systems, this method represents an unbiased and open approach that allows the detection of unexpected changes in metabolite levels. Although the method represents a compromise for a wide range of metabolites in terms of extraction, chemical modification and GC–MS analysis, for 25 metabolites analysed in detail the recoveries were found to be within the generally accepted range of 70–140%. Further, the reproducibility of the method was high: the error occurring in the analysis procedures was found to be less than 6% for 30 out of 33 compounds tested. Biological variability exceeded the systematic error of the analysis by a factor of up to 10. The method is also suited for upscaling, potentially allowing the simultaneous analysis of a large number of samples. As a first example this method has been applied to soil- and in vitro-grown tubers. Due to the simultaneous analysis of a wide range of metabolites it was immediately apparent that these systems differ significantly in their metabolism. Furthermore, the parallel insight into many pathways allows some conclusions to be drawn about the underlying physiological differences between both tuber systems. As a second example, transgenic lines modified in sucrose catabolism or starch synthesis were analysed. This example illustrates the power of an unbiased approach to detecting unexpected changes in transgenic lines.

Introduction

A central interest of our laboratory relates to the understanding of carbohydrate metabolism using the potato as a model plant. To this end we have created numerous transgenic plants characterized by either antisense inhibition or ectopic overexpression of proteins mediating pathways of carbohydrate metabolism. In order to understand better the complexity of events occurring at the metabolite level, it is also necessary to analyse various stages of development, looking at metabolites in many different pathways.

We therefore decided to develop a method that would allow the rapid, highly sensitive and quantitative simultaneous analysis of various metabolites indicative for carbon and nitrogen metabolism, such as sugars, sugar alcohols, dimeric and trimeric saccharides, amines, amino acids and organic acids. In addition, this method should be open to the range of metabolites analysed, and should be suitable for the development of automatic compound identification and quantification.

Here we present an approach essentially based on gas chromatography linked to mass spectrometry (GC–MS), which goes a long way towards meeting the above criteria. When applied to the polar phase of potato tuber extracts, this method allows the detection of a total of 150 compounds in a single run of 1 h duration. Among these 150 compounds we identified 77 with respect to their chemical nature. The reproducibility of this method was high: the cumulative standard deviation was 6% or less for 30 out of 33 compounds tested with respect to the extraction, chemical modification and GC–MS analysis steps. For a wide range of compounds the biological variability was significantly in excess of the variability introduced by the method.

As a first example we applied this analytical approach to two types of potato tuber: soil-grown, and artificially induced in tissue culture. Whereas earlier data have suggested that these two systems are highly comparable, the broad metabolic profiling method described here reveals major and significant differences between the two systems.

We subsequently analysed a range of transgenic plants displaying a modified sucrose or starch metabolism. In addition to the insights provided by the profiles, the identification of several disaccharides including trehalose, maltose and isomaltose, and sugar alcohols including maltitol, specifically in individual transgenic lines illustrates the power of open metabolic profiling for detecting unexpected events.

Results and Discussion

Rationale for developing an analytical system based on GC–MS

Analysis of metabolites in plants is still often performed using either specific enzyme assays or chromatographic separations such as GC or HPLC, which take retention time/coelution as the main parameter for compound identification. Although powerful, these approaches suffer from at least two drawbacks: they represent non-open, biased approaches (the investigator will only find information about the metabolite that was experimentally targeted by the particular analysis protocol); and they are labour intensive because the detection systems used either provide information for only a single compound per assay (as is typical for enzyme-based metabolite assays) or, in the case of chromatographic separation coupled to non-specific detection, can only be applied to mixtures of low complexity, which in turn often necessitates clean-up steps.

Therefore approaches should be developed that combine high sensitivity and high specificity (such as that achieved by enzyme tests) with the potential to accommodate the analysis of highly complex mixtures of compounds. Mass spectrometry obviously meets these requirements as it combines high specificity, based upon compound specific fragmentation patterns, with high sensitivity. It has also been used extensively in plant sciences, although as a rule it has been applied for identifying a small number of specific compounds or for identification of compounds in extracts displaying a low complexity (e.g. Delarue et al. 1998 ; Kamboj et al. 1999 ; Yamaguchi et al. 1990 ). In order to apply the selectivity of mass spectrometry to complex mixtures, a previous chromatographic separation is required. For this purpose either GC, liquid chromatography (LC) or capillary electrophoresis can be linked to mass spectrometric detection (e.g. Godber & Parsons, 1998; Katona et al. 1999 ; Prinsen et al. 1998 ; Tanaka et al. 1998 ). In order to automatically identify a compound in a complex mixture the retention time is an essential parameter, as the analysis software needs to be able to limit the window within which it searches for a particular mass spectrum. Therefore reliability and reproducibility of retention time under given conditions is of crucial importance in the choice of separation method.

Comparison of the three methods mentioned above revealed that GC separation best fulfilled the criterion of high reproducibility in retention time for a set of given compounds. For this reason we decided to develop a method based GC separation and linked to a quadruple mass detector. However, GC–MS is not sufficient for a comprehensive analysis of plant metabolites as it is limited to those classes of compounds that are or can be made volatile, and thus can pass the GC separation under the conditions applied. There is undoubtedly scope for profiling methods to be developed for the more recent LC–MS technologies.

Extraction of polar metabolites from potato tubers followed by methoximation and silylation results in complex and reproducible GC–MS chromatograms of >150 compounds

Metabolites present in organisms such as higher plants are composed of multiple classes of chemical compounds. In order to be able to identify and quantify as many different metabolites as possible by the most reliable and least labour-intensive method, we tested a variety of extraction and chemical modification methods.

With respect to extraction, the addition of methanol to frozen plant material followed by a short heat treatment was found to give the most satisfying results. The combination of methanol and high (70°C) temperature is known to inactivate a majority of enzymes in several systems ( Bligh & Dyer, 1959; Katona et al. 1999 ), which is a necessary prerequisite to preventing changes in metabolite composition due to enzymatic conversions in the homogenate.

In order to make various classes of compounds volatile and thus accessible for analysis by GC, modification of polar functional groups is necessary. To find the most efficient derivatization reagents capable of working with a wide range of compound classes, we tested several trimethylsilylation reagents: N,O-bis(trimethylsilyl)acetamide (BSA); trimethylsilylimidazole (TMSI); bis(trimethylsilyl)trifluoroacetamide (BSTFA); 1,1,1,3,3,3-hexamethyl- disilazane (HMDS); and N-methyl-N-(trimetylsilyl)trifluoroacetamide (MSTFA) ( Katona et al. 1999 ). Among those reagents, MSTFA gave the best results with the broadest range of chemical compounds and produced the least by-products. Thus all subsequent silylation reactions were carried out using this reagent.

The carbonyl groups in sugars and sugar derivatives are another functional group requiring modification before GC–MS analysis. Methoxymation prevents ring formation by reducing sugars and stabilizes carbonyl moieties in the β-position, and has been described as a suitable approach ( Schweer, 1982). We therefore studied different conditions for the methoximation reaction. Methoximation produces two different stereoisomers that were resolved by our chromatography. It is vital that the methoximation reaction proceeds to completion, otherwise a third peak representing the non-methoxyaminated substance will appear. For this reason, reaction times and temperatures of the methoximation procedure were optimized. The best results were obtained following incubation with methoximation reagent for 90 min at 30°C.

The chemically modified extracts were subsequently subject to GC–MS analysis, leading to peak resolution and signal responses that were acceptable for both qualitative identification and quantification purposes. A typical example of a GC–MS total ion chromatogram from the polar phase of a methanol of greenhouse-grown potato tubers is shown in Fig. 1.

Figure 1.

GC–MS total ion chromatogram of a tuber extract from Solanum tuberosum L. cv. Desirée, processed and analysed as described in Experimental procedures.

(a) Complete chromatogram, 4.0–50.0 min.

(b) Demonstration of sample complexity and analyte range by a representative expansion of the chromatogram in (a) for the region 21.5–25.0 min.

Peak identification: 1, glyceraldehyde MEOX1 TMS; 2, heptanoic acid TMS (time reference); 3, phosphoric acid TMS; 4, nonanoic acid TMS (time reference); 5, unknown substance; 6, malic acid TMS; 7, ribitol TMS (quantitative internal standard); 8, undecanoic acid TMS (time reference); 9, asparagine, N,N,O-TMS; 10, tridecanoic acid TMS (time reference); 11, glucose MEOX1 TMS; 12, citric acid TMS; 13, pentadecanoic acid TMS (time reference); 14, nonadecanoic acid TMS (time reference); 15, sucrose TMS; 16, tricosanoic acid TMS (time reference); 17, heptacosanoic acid TMS (time reference); 18, hentriacontanoic acid TMS (time reference); 19, glutamic acid, N,O,O-TMS; 20, pyroglutamic acid, N,O-TMS; 21, glutamine, N,N,N,O-TMS; 22, phenylalanine, N,O-TMS; 23, glucoheptonic acid TMS; 24, ribonic acid TMS; 25, unknown substance; 26, unknown substance; 27, mannitol TMS; 28, quinic acid TMS. Derivatives are per-trimethylsilylated unless otherwise indicated.

In Fig. 1(a) complex pattern of major, minor and trace peaks can be observed. If the amount of a specific compound exceeds the column capacity or the linear response range of the detector then a subsequent analysis should be performed with a higher sample dilution.

When two compounds coelute, differences in the respective mass spectra (leading to different ions specific for the compounds in question) can be used selectively to quantify both compounds. Sensitivity can be increased further by running the quadrupole mass detector in the single ion-monitoring mode. In this mode only a small mass range is analysed, and many more scans are performed per time unit resulting in an approximately 30-fold increase in sensitivity for a given m/z ratio.

77 compounds of known chemical structure can be identified in the polar fraction of potato tuber extracts

As shown in Fig. 1 the total ion chromatogram displays the peak profile of a complex mixture. In order to identify the chemical nature of as many peaks as possible, two strategies were followed. First, the spectra of all identifiable peaks were compared with commercially available electron impact mass spectrum libraries such as NIST for Masslab (Fisons, Manchester, UK) or WILEY (Palisade Cooperation, New York, USA). Secondly, GC–MS analysis was performed using several hundred standard compounds which we assumed to be present in detectable amounts in plant tissues, thus creating our own reference library containing both the retention index of these compounds (as determined under our conditions) and the corresponding mass spectrum.

When these approaches were applied to extracts from potato tubers of non-genetically modified reference plants and to various transgenic lines, we could identify a total of 77 compounds ( Table 1). However, there were still a large number of peaks which were not found either in any of the commercial libraries or in the several hundred compounds that we tested directly. These peaks will require significant further efforts before a chemical name and structure can be assigned to them. The mass spectra of all standard compounds that were processed can be found at http://www.mpimp-golm.mpg.de/willmitzer/index-e.html.

Table 1.  Metabolites identified from a methanol tissue extract of potato tubers
Amino acidsOrganic acidsSugarsSugar alcoholsAromatic amines
  1. Identification was performed by demonstration of cochromatography of a standard substance and by comparison of the mass spectrum to the standard (see Experimental procedures). The m/z value indicated after each compound is the specific ion used for quantifying each metabolite. GABA, gamma-amino butyric acid.

β-alanine (248)2-aminoadipic acid (260)erythrose (205)erythritol (307)dopamine (426)
alanine (116)ascorbic acid (332)fructose (307)glycerol (205)noradrenaline (355)
arginine (348)benzoic acid (179)fructose-6-phosphate (315)inositol (305)normetanephrine (297)
asparagine (188)citric acid (465)fucose (160)maltitol (361)octopamine (426)
aspartic acid (232)fumaric acid (245)galactose (160)mannitol (319)tyramine (338)
cystathionine (278)glucoheptonic acid (435)glucose (160)xylitol/arabitol (319) 
cysteine (220)gluconic acid (333)glucose-6-phosphate (387)  
GABA (304)glutaric acid (261)isomaltose (480)  
glutamic acid (246)glyceric acid (189)mannose (160)  
glutamine (156)hydroxybenzoic acid (267)maltose (361)  
glycine (174)isocitric acid (273)raffinose (437)  
histidine (154)α-ketoglutaric acid (198)ribose (307)  
homocysteine (234)lactic acid (117)sucrose (451)  
homoglutamine (170)malic acid (245)trehalose (191)  
homoserine (218)oxalic acid (147)xylose/arabinose (307)  
isoleucine (158)oxaloacetic acid (333)   
leucine (158)6-phospho-gluconic acid (387)   
l-hydroxyproline (230) 3-phospho-glycerate (299)   
lysine (317)phosphoric acid (299)   
methionine (176)quinic acid (345)   
ornithine (142)ribonic acid (307)   
phenylalanine (218)shikimic acid (372)   
proline (142)succinic acid (247)   
serine (116)
threonine (291)
tryptophane (202)
tyrosine (218)
valine (144)

As shown in Table 1, we were also able to detect various sugar phosphates. Although surprising at first glance due to the assumed low volatility of these substances, the mass spectrum clearly identified the phosphorylated compounds due to the presence of an M-15 fragment peak at m/z = 706 for both glucose-6-phosphate and fructose-6-phosphate. The appearance of free phosphate-TMS esters (m/z = 314) at the same retention time further indicates that intact sugar phosphates enter the ionization source.

Variability caused by extraction, chemical modification and analysis by GC–MS is small when compared to the biological variability within samples

An essential factor in assessing the quality of analytical processes is the reproducibility of results. In order to test our system with respect to this parameter, we divided the analytical process into three components that contribute to the observed variability. These were variability caused by: (i) the final analysis (chromatography, detection, stability of chemically modified samples); the sample preparation; the biological material.

To gain an insight into the reproducibility of the final analysis under authentic conditions, a single chemically modified sample composed of 23 representative standard substances aliquoted in separate measurement vials was measured every 2 h over a 30 h period (the processing time of a typical series of 30 samples).

In this experiment most compounds yielded stable results. In general a loss of only 2% of the initial signal was observed by the end of the 30 h analysis period (data not shown). The only exceptions were glutamine, tryptophane and glycerol, where losses of 24, 10 and 9%, respectively, were observed. We therefore conclude that under typical conditions the reliability of the method is very high.

Before the samples are injected into the GC–MS they undergo a complex series of extraction and chemical modification steps. To see how these steps influence the variability of the results obtained, a single tuber extract was divided into 20 aliquots, chemically processed and finally analysed by GC–MS. Analysis of the data shown in Table 2 demonstrate that the standard deviation introduced by these steps is, as a rule, below 6% of the mean. This was the case for 29 of the 33 metabolites tested in this experiment.

Table 2.  Reproducibility of derivatization reaction and GC–MS analysis using potato tuber extracts from plants expressing an invertase in the cytosol ( Sonnewald et al. 1997 )
Derivative aResponse ratio bSDSD (%) (n = 20)
  • Tuber extracts from 6 representative developing tubers were pooled and divided into 20 aliquots. Each aliquot was derivatized and analysed as described in Experimental procedures. MEOX, methoxyaminated derivative; TMS, trimethylsilylated derivative; SD, standard deviation.

  • a

    Derivatives were per-trimethylsilylated if not otherwise indicated.

  • b

    Response ratios represent peak area ratios using ribitol as quantitative internal standard.

alanine, N,O-TMS0.720.05928.28
asparagine, N,N,O-TMS2.780.11264.06
aspartic acid, N,O,O-TMS0.870.00961.11
cysteine, N,O,S-TMS0.010.00076.65
glutamic acid, N,O,O-TMS1.070.01941.82
glutamine, N,N,O-TMS1.610.07764.81
glycine, N,N,O-TMS0.170.00613.52
isoleucine, N,O-TMS0.280.01465.18
leucine, N,O-TMS0.150.00936.12
lysine, N,N,N,O-TMS0.030.00092.68
methionine, N,O-TMS0.320.00752.36
phenylalanine, N,O-TMS0.590.00510.86
proline, N,O-TMS0.040.00277.31
pyroglutamic acid, N,O-TMS1.640.03282.01
serine, N,O,O-TMS0.510.02494.88
threonine, N,O,O-TMS0.040.00174.76
tyrosine, N,N,O-TMS0.760.03985.26
valine, N,O-TMS0.730.04105.61
citric acid TMS6.780.38965.74
fumaric acid TMS0.020.00094.89
isocitric acid TMS0.050.00173.39
malic acid TMS0.490.00390.82
quinic acid TMS0.080.00182.36
shikimic acid TMS0.010.00032.37
succinic acid TMS0.060.00173.01
fructose MEOX1 TMS0.010.00033.32
glucose MEOX1 TMS8.360.46055.51
glucose-6-P MEOX1 TMS0.270.01274.68
mannose MEOX TMS0.030.00103.01
sucrose TMS0.050.00142.34
glycerol TMS0.010.001914.10
inositol TMS0.030.00062.04
mannitol TMS0.390.00310.78

As both of these analyses demonstrate the robustness and high reproducibility of the analytical method per se, we decided to test the biological variability of samples. Tuber slices from nine individual wild-type plants grown side-by-side under identical conditions were analysed. Under ideal growth conditions the metabolites display standard deviations below 20% of the mean ( Table 3) with only two exceptions, glucose and fructose, which are well known to demonstrate a higher variability in potato tubers ( Merlo et al. 1993 ; Veramendi et al. 1999a ; Veramendi et al. 1999b ). In most cases biological variability exceeded the experimental error by at least a factor of two (observed range 1.5- to 10-fold).

Table 3.  Variability of individual potato tubers
Derivative aRelative response ratio b
(g−1 FW)
SD a
(n = 9)
% a
  • Samples were obtained from nine wild-type potato tubers of a single harvest. Extraction, derivatization, and analysis were performed as described in Experimental procedures.

  • a

    Derivatives were per-trimethylsilylated if not otherwise indicated.

  • b

    Response ratio was normalized with respect to the fresh weight of the tuber sample.

alanine, N,O-TMS10.182.59525.48
asparagine, N,N,O-TMS112.0426.69323.82
aspartic acid, N,O,O-TMS32.083.0859.62
cysteine, N,O,S-TMS0.390.10225.98
glutamic acid, N,O,O-TMS66.055.5828.45
glutamine, N,N,O-TMS179.9240.42122.47
glycine, N,N,O-TMS4.640.75816.33
isoleucine, N,O-TMS11.031.26211.44
leucine, N,O-TMS2.870.37212.94
lysine, N,N,N,O-TMS8.841.07212.12
methionine, N,O-TMS18.752.92515.60
phenylalanine, N,O-TMS30.465.95319.54
proline, N,O-TMS1.810.25213.95
pyroglutamic acid, N,O-TMS94.1920.27721.53
serine, N,O,O-TMS10.381.64515.84
threonine, N,O,O-TMS2.880.45215.70
tyrosine, N,N,O-TMS59.9312.45720.79
valine, N,O-TMS25.092.52010.04
citric acid TMS553.4256.16510.15
fumaric acid TMS0.200.03115.17
isocitric acid TMS3.480.46413.33
malic acid TMS11.024.18137.94
quinic acid TMS5.160.87316.93
shikimic acid TMS0.420.05513.22
succinic acid TMS0.250.07429.97
fructose MEOX1 TMS6.013.37456.14
glucose MEOX1 TMS93.4853.75457.50
glucose-6-P MEOX1 TMS1.860.23112.40
sucrose TMS168.3634.69920.61
inositol TMS9.440.9139.67
mannitol TMS6.280.94415.05
mannose MEOX TMS0.280.06824.28

We therefore conclude that with the GC–MS technology the variability in results is essentially due to the variability within the biological samples themselves.

GC–MS allows both absolute and relative determination of the compounds detected

In order to extend the evaluation of the suitability of this GC–MS approach for metabolic profiling, we quantified representative metabolites in potato tubers. We established calibration curves for 33 compounds, members of five chemical classes: amino acids, organic acids, sugars, saccharides and sugar alcohols. A linear relationship covering the normal concentration range present in plant tissues was observed (data not shown).

These metabolites were then quantified in developing potato tubers ( Table 4); the resulting absolute levels with respect to fresh weight were found to be in the same range as previously reported by both our and other groups using enzymatically linked photometric assays or HPLC analysis of extracts ( Burrell et al. 1994 ; Geigenberger et al. 1998 ; Sweetlove et al. 1996 ; Trethewey et al. 1998 ). Thus we judged this technology valid. The only major exception observed was with respect to the citric acid content. Citric acid levels were approximately 10 times higher using the GC–MS approach in comparison with previous studies using spectrophotometric-based methods. We were not able to determine the reason for this discrepancy.

Table 4.  Quantitative determination of metabolite concentra- tions in developing potato tubers
MetaboliteConcentration
(μmol g−1 FW)
SE
(n = 6)
%
  1. Single tuber samples of six plants were measured. Developing tubers were harvested during the spring season from 10-week-old greenhouse plants grown in 2 l pots. SE, Standard error.

β-alanine0.150.016.6
alanine1.680.3420.2
ascorbic acid0.530.1426.4
asparagine5.622.0135.7
aspartic acid1.270.118.7
cysteine0.410.012.4
glutamine1.080.2624.1
glycine0.230.0313.0
isoleucine0.940.1617.0
leucine0.430.1227.9
lysine1.060.1615.1
methionine0.650.0812.3
phenylalanine0.590.0915.2
proline0.180.015.5
serine1.340.1511.2
valine2.510.2911.5
fumaric acid0.200.012.8
glyceric acid0.150.013.3
citric acid18.863.2717.3
isocitric acid0.170.0424.1
malic acid5.390.7313.5
oxalic acid1.000.1919.5
quinic acid15.672.3615.1
shikimic acid0.370.025.8
succinic acid0.970.1010.6
fructose0.010.0025.0
galactose0.010.0023.5
glucose23.844.0917.1
glucose-6-phosphate0.210.001.8
mannose0.140.004.7
sucrose25.913.2912.7
inositol0.060.0111.7
mannitol0.060.006.1

In addition to the quantitative measurements, we performed a series of recovery experiments for 25 metabolites drawn as representatives of these five compound classes, and found that the calculated recoveries were within the generally accepted range for analysis work at 70–140% (see Experimental procedures). Taking together the results of the analysis of variability, the data from the recovery experiments, and the comparability of the absolute values determined with previous studies, we conclude that the GC–MS-based approach is valid for the study of metabolism in potato tubers.

In many cases absolute concentrations are not of prime importance. For the analysis of a special environmental or developmental situation, or for the comparison of a specific genotype with standard tissue samples, relative data are sufficient. The current profiling techniques applied in RNA expression analysis provide only relative data. The determination of relative values can be easily achieved by the GC–MS method as well. For this purpose, within each chromatogram the peak areas derived from specific ion traces indicative of each analysed compound are normalized by the peak area derived from an internal standard, such as ribitol, present within the same chromatogram, resulting in response ratios for all compounds analysed (see Experimental procedures). The response ratios are subsequently converted to relative response ratios through division by the fresh weight of each sample, thus achieving further normalization. These relative response ratios can be directly compared between different samples without knowledge of absolute compound concentrations. The specific ion masses used for the respective analyte for quantification are given in Table 1.

Determination of both relative values and absolute concentrations have advantages and disadvantages. However, for comparative purposes between two sample types we routinely describe changes by calculating the quotients of the various relative response ratios for each compound.

Automization of the chromatogram analysis

It is obvious from the preceding descriptions that evaluation by eye of each single chromatogram would be an impossible task and would represent a major stumbling block for the efficient use of this technology. We therefore decided to use a chromatography data analysis algorithm that allows multiparallel and automatic identification and quantification of compounds present in a GC–MS chromatogram. This basic algorithm, which is part of the masslab software distributed by ThermoQuest (Manchester, UK), allows the use of two strategies for identification of a compound and quantification of a set of given compounds. For identification, the first parameter used by the software is the retention time of the compound relative to a standard compound. Initially we used just one compound for determination of relative retention times; however, this proved insufficient as non-linear shifts of retention times occur in the elution profiles during the lifetime of a GC column. We therefore decided to spike every sample before injection with several compounds that were absent from the fraction under analysis. For the work presented here we used fatty acids ( Fig. 1); the use of fatty acids has been described as an alternative approach to using n-alkanes for determining the Kovats index ( Castello, 1999; Gonzalez & Nardillo, 1999). Subsequently the relative retention time of each compound is interpolated with respect to its two nearest neighbours and taken as the first identification criterion. As a second subsequent criterion we used the full electrone impact ionization (EI) mass spectrum for the compounds in question. A match factor was automatically calculated which provides an indication of the reliability of assignment. In other words, a confidentiality index is provided which alerts the investigator when manual inspection of the compound assignment is necessary.

For quantification purposes a single specific ion known to be selective and sensitive for the compound in question was taken, and the response ratio calculated as described above. For the analysis presented here ribitol was used as the quantitative internal standard (see Experimental procedures).

Applying this method to potato tuber extracts, we can currently identify and automatically quantify 60 of the 77 known compounds with a very high fidelity, i.e. less than 0.1% manual inspection. The remaining 18 compounds can be quantified, but this process requires at least 5% reassessments of the original chromatogram by a trained user. It is to be expected that with the development of improved software and GC–MS hardware the number of compounds requiring supplementary manual assessment will decrease.

Metabolic profiling using GC–MS reveals major differences between soil- and in vitro-grown tubers

When analysing the growth and development of plant systems, in vitro culture is an interesting technology as it allows the manipulation of environmental conditions in a defined manner. With respect to potato tubers, numerous reports including work from our own group have described the suitability of in vitro-grown tubers for the analysis of gene expression or changes in metabolite composition. Morphological, molecular and biochemical data appear to be in agreement with the assumption that in vitro-grown tubers represent a faithful phenocopy of soil-grown potato tubers ( Debon et al. 1998 ; Desire et al. 1995 ; Veramendi et al. 1999a ; Visser et al. 1994 ). In view of the compelling evidence for the equivalence of these systems, we decided to apply the metabolic profiling approach in order to assess a broader range of compounds using this unbiased approach.

Data describing the ratios of many metabolites in both systems are summarized in Fig. 2, which shows that in vitro- and soil-grown potato tubers are not as similar as previously thought. Major differences were found in the group of amino acids, most notably in glutamic acid and other amino acids derived from α-ketoglutaric acid such as glutamine, proline and arginine. In addition, large increases were observed in the levels of amino acids derived from oxaloacetic acid, such as asparagine and lysine, within in vitro-grown tubers. However, tyrosine, glycine, alanine, β-alanine and phenylalanine decreased in in vitro-grown when compared to soil-grown tubers. In general, microtubers were found to contain a much higher amount of amino acids in comparison to soil-grown tubers. A high anaplerotic demand for carbon may explain the decreases observed in citric acid-cycle intermediates such as citric acid, malic acid, succinic acid and isocitric acid. Alternatively, many of these changes could be a consequence of the higher amount of nitrate supplied to the microtubers. As the in vitro-grown tubers are probably not limited in carbohydrate supply due to the large amount of sugars present in the tuber-inducing medium, the rate of nitrate assimilation may be markedly increased in comparison with soil-grown tubers ( Morcuende et al. 1998 ).

Figure 2.

Comparison of metabolite levels in developing soil-grown tubers and in vitro tubers.

The quotient of mean relative response ratio from in vitro tuber samples (n = 6) and mean relative response ratio of soil-grown tuber samples (n = 6) are plotted using a logarithmic scale. Values <1 represent a decrease in metabolite levels in the in vitro tuber compared to soil-grown tuber; values >1 represent an increase. The significance of the changes was evaluated by a t-test; black bars indicate statistically significant differences between the systems (P < 0.05), white bars show non-significant differences, grey bars indicate metabolites which were detected in the in vitro tubers but were below the detection limit for the soil-grown tuber samples. In these cases the numerical value of the detection limit of the respective compound within the soil-grown tuber samples was used to calculate an estimated, representative quotient.

Another important difference between both types of tubers is the higher amount of compounds indicative of osmotic stress found in the in vitro tubers ( Hare et al. 1998 ; Hare et al. 1999 ; Karakas et al. 1997 ). For example, glycerol, mannitol, inositol, and proline are significantly increased. Finally, some of the changes seen in the amino acid pools, such as the increase in asparagine, glutamine, and glutamic acid, have been observed in water-stressed leaves of tomato ( Bauer et al. 1997 ), sugar beet ( Gzik, 1996) and mulberry ( Ramanjulu & Sudhakar, 1997). The reason for the possible water stress could be the high amount of sucrose present in the tuber-inducing medium.

In conclusion, the application of a broad profiling method to two systems previously thought to be very comparable reveals that caution must be exercised when interpreting biological events by the analysis of a restricted number of parameters.

Metabolic profiling of several transgenic lines modified in either sugar or starch metabolism reveals unexpected changes in disaccharides and sugar alcohols

In order to explore further the power of metabolic profiling for detecting unexpected changes, tubers from three transgenic and one wild-type line were subjected to GC–MS analysis, as described above. The transgenic lines analysed were either altered in sucrose metabolism (expression of a yeast invertase in either the cytosol or apoplast ( Sonnewald et al. 1997 ; Trethewey et al. 1998 ), or inhibited in starch biosynthesis following antisense RNA-mediated repression of the activity of ADP-glucose pyrophosphorylase ( Müller-Röber et al. 1992 ).

GC–MS analysis revealed that amongst the metabolites analysed a large fraction (33) were significantly changed in at least one of the transgenic lines. A comparison of the data obtained for the three transgenic lines using GC–MS profiling with the values previously published from HPLC analysis for amino acids and organic acids ( Trethewey et al. 1998 ; Trethewey et al. 1999 ) shows that the profiling data are in broad agreement with the previously determined values ( Fig. 3). Our general experience suggests that the differences observed in the extent of quantitative changes are due to the normal differences between the growing conditions of different batches of tubers.

Figure 3.

Comparison of metabolite levels in wild-type developing potato tubers with those in tubers of transgenic potato plants.

Transgenic plants exhibiting antisense repression of ADP-glucose pyrophosphorylase (AGP93, black bars), or overexpressing a yeast invertase in the apoplast (U-IN1–33, grey bars) or in the cytosol (U-IN2–30, white bars). The quotients of the mean relative response ratio from transgenic tuber samples (n = 6) and from wild-type tuber samples (n = 6) are plotted using a logarithmic scale, as described for Fig. 2. Only changes in metabolite levels that were evaluated to be significantly different from the wild type are shown (P < 0.05). Dotted bars indicate metabolites which were detected in the transgenic tubers but were below the detection limit in wild-type tubers. In these cases the numerical value of the detection limit of the respective compound in the soil-grown tuber samples was used in order to estimate a representative quotient. Metabolites that did not show significant changes between the four genotypes were leucine, isoleucine, beta-alanine, ornithine, valine, asparagine, glycine, glutamine, threonine, glutamic acid and GABA.

All the scientific conclusions previously drawn about the redistribution of metabolism in the transgenic lines can be drawn from this single profiling analysis. In particular, the increase in glycolysis, amino acids and organic acids in the line expressing the yeast invertase in the cytosol can be clearly deducted from the data in Fig. 3. With respect to the transgenic line expressing a yeast invertase in the apoplast (U-IN1), it is apparent from Fig. 3 that this line does not have an elevated respiratory flux.

With respect to the line expressing the yeast invertase in the cytosol, the appearance of 6-phosphogluconic acid is a new result and can be taken as a strong indication of a significantly increased level of the oxidative pentose phosphate pathway, which might be expected given the large increase in glucose-6-phosphate in this line.

A most remarkable and unexpected finding concerns the presence of dramatically increased levels of disaccharides such as maltose, isomaltose and trehalose in the line expressing the cytosolic invertase. This result has been found in three independent lines (data not shown) Specifically, the presence of trehalose is an exciting observation. In addition to confirming and extending previous observations on the capacity of higher plants to synthesize trehalose under specific conditions ( Goddijn & Smeekens, 1998; Müller et al. 1999 ), this finding is of relevance as trehalose has been discussed extensively in the literature as a possible signal for the metabolic status of a given cell. The fact that trehalose appears only in those cells where an increased rate of glycolysis and respiration has been observed makes this a fascinating and stimulating observation. Further comparisons with other lines displaying a change in the rate of glycolysis and respiration must reveal whether or not this exciting but nevertheless simple interpretation is true.

The transgenic lines have been analysed in significant detail in earlier publications by applying conventional techniques such as enzyme assays and HPLC. Nevertheless, the occurrence of trehalose and other disaccharides has not been noted, mainly due to the fact that the methods used previously were not capable of determining these compounds. The unbiased approach for metabolic profiling presented here gives a clear advantage in providing the opportunity to find unexpected events, as exemplified by the identification of trehalose, and thus may provide novel insights into metabolic networks.

Conclusions

Using GC–MS as a system for separation and detection of metabolites, we were able to develop a simple procedure allowing the simultaneous analysis of a large group of compounds representing sugars, sugar alcohols, amino acids, organic acids and some special compounds in plant extracts. This method allows an unbiased study of a range of metabolic pathways with only minimal effort. As this technology utilizes MS, it also combines high sensitivity with high specificity and, as shown here, high reproducibility.

We are presently in the process of applying this technique to a number of other plant species. Depending upon the species and the tissue, between 50 and 400 different compounds can be detected when the analysis is applied to hydrophilic and lipophilic classes of compounds. As this system has the potential to be fully automated, we believe that in the future it will represent a major tool for characterizing the metabolic status of a plant with respect to environmental, developmental or genetic factors.

Experimental procedures

Plant material

Potato plants, Solanum tuberosum L. cv. Desirée, were supplied by Saatzucht Lange (Bad Schwartau, Germany). Plants were grown in a greenhouse at 22°C under a 16 h light/8 h dark regime with supplementary artificial light under a minimum of 250 μmol photons m2 sec−1. In this paper the term ‘developing tubers’ refers to tubers above 10 g that were harvested from 10-week-old plants.

Microtubers were generated in vitro as described by Veramendi et al. (1999a) . Single-node explants were cultured in MS media ( Murashige & Skoog, 1962) containing 6% (w/v) sucrose and 11.6 μm kinetin. Sets of 15 explants were kept in the dark in glass jars containing 50 ml MS medium. Microtubers were harvested after 4 weeks at 20°C.

Chemicals

All chemicals and pure standard substances were purchased from either Sigma–Aldrich Chemie GmbH (Deisenhofen, Germany) or Merck KGaA (Darmstadt, Germany).

Extraction and derivatization of potato tuber metabolites for GC–MS analysis

Tuber slices or whole microtubers were immediately frozen in liquid nitrogen and stored at −70°C until further analysis. A polar metabolite fraction was obtained from either approximately 50 mg microtuber or 100 mg tuber slices by Ultra Thurax T25 (IKA Labortechnik, Staufen, Germany), homogenization in 1400 μl 100% methanol with 50 μl internal standard (2 mg ribitol ml−1 water). The mixture was extracted for 15 min at 70°C. The extract was vigorously mixed with 1 vol water and subsequently centrifuged at 2200 g. Aliquots of the methanol/water supernatant (1.00 or 0.25 ml) were dried in vacuo for 6–16 h. The dried residue was redissolved and derivatized for 90 min at 30°C (in 80 μl of 20 mg ml−1 methoxyamine hydrochloride in pyridine) followed by a 30 min treatment at 37°C (with 80 μl MSTFA). 40 μl of a retention time standard mixture was added prior to trimethylsilylation. This retention time standard mixture contained 3.7% (w) heptanoic acid, 3.7% (w) nonanoic acid, 3.7% (w) undecanoic acid, 3.7% (w) tridecanoic acid, 3.7% (w) pentadecanoic acid, 7.4% (w) nonadecanoic acid, 7.4% (w) tricosanoic acid, 22.2% (w) heptacosanoic acid and 55.5% (w) hentriacontanoic acid dissolved in tetrahydrofuran at 10 mg ml−1 total concentration.

Standard substances for peak identification were dissolved in water at 10 mg ml−1. A 5 μl volume of standard solution was dried in vacuo and derivatized with 50 μl of 20 mg ml−1 methoxyamine hydrochloride in pyridine and 50 μl MSTFA, as described above.

To establish the efficiency of the extraction procedure, the recovery of various standard metabolites was determined by the addition of authentic metabolite standards to the tissue sample at the start of the extraction procedure. Standard substances were added in threefold excess of the determined endogenous concentrations. Estimates of recovery were 138% for alanine, 100% for aspartic acid, 107% for glycine, 107% for isoleucine, 99% for leucine, 92% for lysine, 106% for phenylalanine, 90% for proline, 118% for valine, 127% for serine, 110% for fructose, 78% for fructose-6-phosphate, 101% for galactose, 129% for glucose, 78% for glucose-6-phosphate, 115% for maltose, 100% for sucrose, 101% for inositol, 108% for mannitol, 139% for glyceric acid, 117% for fumaric acid, 98% for isocitric acid, 118% for malic acid, 99% for shikimic acid and 73% for succinic acid.

GC–MS analysis

Sample volumes of 1 μl were injected with a split ratio of 25 : 1 using a hot-needle technique. The GC–MS system consisted of an AS 2000 autosampler, a GC 8000 gas chromatograph and a Voyager quadrupole mass spectrometer (ThermoQuest, Manchester, UK). The mass spectrometer was tuned according to the manufacturer's recommendations using tris-(perfluorobutyl)-amine (CF43). Gas chromatography was performed on a 30 m SPB-50 column with 0.25 mm inner diameter and 0.25 μm film thickness (Supelco, Bellfonte, CA, USA). Injection temperature was 230°C, the interface set to 250°C and the ion source adjusted to 200°C. The carrier gas used was helium set at a constant flow rate of 1 ml min−1. The temperature program was 5 min isothermal heating at 70°C, followed by a 5°C min−1 oven temperature ramp to 310°C and a final 1 min heating at 310°C. The system was then temperature equilibrated for 6 min at 70°C prior to injection of the next sample. Mass spectra were recorded at two scans per sec with an m/z 50–600 scanning range. The chromatograms and mass spectra were evaluated using the masslab program (ThermoQuest, Manchester, UK). A retention time and mass spectral library for automatic peak quantification of metabolite derivatives was implemented within the masslab method format.

Statistical analysis

The t-tests were performed using the algorithm incorporated into Microsoft excel (Microsoft Corporation, Seattle, WA, USA). The word significant is used in the text when the change in question has been confirmed to be statistically significant (P < 0.05) with the t-test.

Acknowledgements

We would like to thank the gardeners for taking excellent care of the greenhouse plants. Our special thanks are assigned to Dr Alisdair Fernie for proofreading the manuscript and his helpful discussions. We would also like to thank Kristina Zubow for her patient and tireless efforts in creating the mass spectral library.

Ancillary