Ideally, GM crops and the conventional counterparts should share an identical genetic background so as to address metabolic differences specifically to the transgene action. In strictly self-fecundating crops such as lettuce the homozygosis grade is very high and all marketed varieties are (virtually) highly inbred [15]. In this context, the performance of a given (stable homozygous) GM cultivar is compared to the same non GM cultivar that provided material for the transformation. Moreover, a set of controls are useful to assess whether the metabolic changes are a consequence of the transformation process (e.g., events of T-DNA insertion and position) or the (desired/unexpected) transgene functioning (see section 3.1)

A tight control of environmental conditions (light intensity, photoperiod, temperature, humidity, etc.) and conditions of cultivation (soil composition, watering, fertilization, etc.) can minimize the exogenous effects on plant metabolite variation. Towards that aim, the glasshouse growing system offers a good solution, since it allows similar, if not identical, conditions for the production of modified and the wild type crops (e.g., lettuce) [8]. The metabolic changes due to variation in environment or the cultivation method can be monitored by NMR to reveal the corresponding metabolic patterns and the interactions of environmental and genetic factors at the level of metabolome [16].

Sampling procedures should consider diurnal cycles of metabolism, particularly in situations where these may alter the metabolite content. A limited number of cases where NMR was used for the metabolic study of plant development [8,9,17] showed that each developmental program of plant organs (leaves, fruits, etc.) is reflected in unique metabolic changes, processes that are developmentally regulated. Consequently, the GM and control plant tissue to be analyzed must be sampled from the plants at the same or comparable developmental stage. A sufficient number of samples for every combination of factors (genetic, environmental, etc.) should be analyzed to obtain statistically significant results. It is understood that the larger the variability of metabolic content, the higher is the number of samples required for the analysis. The number of GM and control samples should be comparable to avoid bias in the statistical treatment of data. Proper storage of plant tissues is required to prevent the metabolite decomposition; low temperature storage (−20°C or even lower) and lyophilization can minimize storage-based decomposition.

Simple methods for extraction of water-soluble and lipo-soluble metabolites need to be applied [6,18]. The use of deuterated water for a direct extraction of metabolites from lyophilized tissues is a simple method to avoid any intermediate steps (for example, extraction by H₂O followed by freeze-drying). The major part of water-soluble metabolites has acid-base properties (organic acids, amino acids, phenols, amines, etc.). The simultaneous presence of protonated and deprotonated forms of metabolites and the labile equilibrium between them makes an NMR spectrum pH-dependent. An increment or decrement of pH value as low as 0.1 unit can give rise to an upfield or downfield shift of NMR signals of 0.02 ppm or even higher which corresponds to 12 Hz at 600 MHz (5-10 times the line width). The strong pH-dependence of NMR signals makes the comparison of aqueous extracts with different pHs difficult and precludes the application of automatic or semi-automatic routines for intensity reading. This problem can be overcome using buffered solutions. The concentration of the buffer should be sufficient to assure the minimum pH variation whenever a new kind of plant sample is to be analyzed. For example, a 400 mM phosphate buffer was sufficient for buffering the aqueous solution of metabolites from 25 mg of lyophilized tomato powder [5,9]. On a different note, a relevant loss of sensitivity was observed when a high buffer concentration was used, particularly with cryogenically cooled probes [19]. The phosphate buffer is a suitable one for NMR applications because it has only exchangeable protons that do not produce additional ¹H signals. For the same reason, many organic acids and bases can not be used for buffering except the completely deuterated ones.

According to our data, the NMR signals of several organic acids (citric, malic) capable of chelating metals are broadened probably due to paramagnetic cations (Fe³⁺, Mn²⁺) present in plant tissue extracts. These cations can be chelated by EDTA. In fact, EDTA addition into neutral solutions gives rise to a sharpening of the ¹H-NMR signals for citric and malic acids.

The stability of metabolites during extraction procedures and NMR measurement should also be taken into account. For example, the endogenous enzymatic activity may induce the hydrolysis of sucrose to fructose and glucose.

As mentioned above, the advantage of NMR application in metabolite analysis is the ability to identify a wide range of compounds directly from mixtures of metabolites. The important characteristic in this context is the dynamic range of an NMR spectrometer. The larger the dynamic range, the higher the concentration range and the number of metabolites that can be simultaneously analyzed in an experiment. In the case of modern high-field NMR spectrometers, the dynamic range is higher than 10⁵ [20]. Another important advantage of the NMR spectroscopy relative to other techniques is the ability to determine the concentration of any metabolite using the same internal standard. Concerning the NMR measurements, several important factors are: sufficient time for relaxation (that can be assured by using longer relaxation delay and/or lower pulse duration), signal to noise ratio sufficient to detect minor components, and proper solvent signal suppression to assure the optimal dynamic range.

As an example of NMR metabolic profiling of GM foods, a study of transgenic lettuce (Lactuca sativa cv "Luxor") over-expressing the KNAT1 gene from Arabidopsis is reported. ¹H-NMR spectra of water soluble extracts of control and GM lettuce leaves are shown in Figure 1a and Figure 1b, respectively. The ¹H spectral assignment was obtained using 2D experiments and as previously reported [6,8]. These spectra were conservative, as previously observed for both conventional and transgenic "Cortina" [8]. No novel or unusual signals were observed when the spectra of aqueous extracts from PetE:KNAT1 leaves and conventional "Luxor" counterparts were compared. The position of peaks in ppm was identical in both genotypes, nevertheless intensity differences in some peaks were clearly apparent (Figure 1).

The intensity of major water soluble metabolites seen in the ¹H-NMR spectra was quantified for both GM and control leaves from “Luxor” plants at two growth stages. The tree cluster analysis was used to classify samples of transgenic (PetE:KNAT1) and conventional lettuce without any a priori hypothesis (see Figure 2).

By making a cut-off in the dendrogram at the highest level, all the samples separated into two clusters: the first one includes all PetE:KNAT1 and control samples at the 4^th leaf stage, while the second one includes GM and wild type samples at the head stage. A clear effect of the development is present. If the cut-off is made at a lower level, other sub-clusters are observable in which samples are grouped according to the genotype. Thus, the effect of two factors can be observed: the natural leaf development and the genotype. The leaf development program appears to have the most significant impact on the metabolome.

The natural grouping of lettuce samples can be also explored using principal component analysis (PCA). The PCA map performed on 49 samples is shown in Figure 3; the combined PC1 and PC2 explain 56.1% of variance. The samples along PC1 are grouped according to the developmental stage: the head stage lies in the right area, the 4^th leaf stage falls in the left area (with one exception for one PetE:KNAT1 head stage sample). The control samples show the best separation of two developmental stages. In the case of GM samples, the separation dependence on the growth stage is less evident, suggesting that the growth can be affected by other factors, including the transgene. Finally, PC2 is the most effective in discriminating conventional lettuce from the GM lettuce: by tracing an horizontal line it is possible to separate wild type from the PetE:KNAT1 genotypes.

The influence of each of the two factors, i.e. leaf development and genotype, inclusive of the transgene effect, on a single metabolite can be evaluated using a simple 2² factorial design scheme [21] and the histograms of the mean values shown in Figure 4, Figure 5 and Figure 6. In the factorial-design scheme every factor has two levels: the genetic background factor has Control (C) or PetE:KNAT1 (K) genotype levels, whereas the leaf development factor has 4^th leaf (4L) or head stage (H) levels. Four factor combinations, i.e., C-4L, C-H, K-4L and K-H exist. For each combination a suitable number of samples (9-15 replications) was analyzed.

The effect of each factor, i.e., the genotype and the leaf development, on the level of each metabolite can be calculated separately in cases where no interaction between the factors is obvious. The statistically significant increase or decrease of a particular metabolite level due to a specific factor is denoted by + or -, respectively (Table 1). For instance, the significant decrease of Ile content in GM with respect to control samples is denoted by “-“ in Table 1. The occurrence of significant interaction between the two factors can also be determined in the factorial scheme. The interaction, reported in Table 1 with an asterisk, means that the genotype and leaf development can influence each other. For example, as in the case of malic acid, the leaf development modulates or alters the genotype effect and viceversa, the transgene expression modulates the leaf maturation.

Organic acids. The trend of organic acid content during the growth (4^th leaf stage- head stage) was substantially variable in both GM and C genotypes (Figure 4). The decrease of most organic acids was observed in PetE:KNAT1 leaves with respect to controls, with the exception of citric acid that maintained a higher level in the GM samples during the growth.

The data in Table 1 are indicative of whether the content of a specific metabolite is a result of the transgene interacting with plant developmental phase. For instance, as previously highlighted with regard to malic acid (Table 1, asterisks; Figure 4, third panel), its content decreased two-fold during the development of control plants, whereas a 1.4-fold decrease was observed in the case of GM line. The statistical analysis indicates that, in the case of malic acid, the effects of genetic background and growth stage are not independent from each other (with 95% confidence). Analyses performed on other independent Pet:KNAT1 Luxor lines (not shown) suggest that the developmental effects are modulated by that of the transgene rather than by the transformation process. Finally, the independent effect of both factors are clearly seen for tartaric acid (the last panel, Figure 4).

Amino acids (AA). An increase in the AA content was observed during the growth (4^th leaf - head stage) of both transgenic and control plants (Figure 5), with the exception of glutamic acid (Glu). The molecular abundances for most of the AA were significantly different in GM relative to control leaves. Moreover, a marked interaction between genotype and development was revealed for Val, Thr, Ala and Asn (asterisked in Table 1). In other words, the level of these AA increased during the leaf development, but the increase was significantly higher in conventional than in PetE:KNAT1 lettuce.

Mono- and disaccharides. As for β-glucose and fructose, the variation trends at distinct growth stages of conventional genotype were opposite to those of GM line (Figure 6), while the trend for sucrose was less variable (Figure 6, third panel). The level of both mono-saccharides was influenced by the interaction between the transgene and development factors (Table 1). Interestingly, such interaction suggests that the transgene strongly contrasts (rather than moderate) the developmental factor leading to a drop in β-glucose and fructose. This observation suggests that the simple sugars are major, if not direct, targets of KNAT1 control.

Other compounds. With regard to the caffeoyl-tartaric acid, PetE:KNAT1 leaves showed a significantly lower content than controls. Caffeic acid and its derivatives are precursors in the biosynthesis of lignin units, hence the diminishment of caffeoyl-tartaric acid suggests that lignin content may also be decreased. This is consistent with the low lignin content of plants overexpressing KNAT1 [22], which is known to target lignin genes in model species.

Sister plants derived from a single mother (Lactuca sativa cultivar “Luxor”, compact butter head type, Nunhems, http://www.nunhems.com, out of production) provided leaves for Agrobacteriumtumefaciens mediated genetic transformation, which was performed with the empty vector (pVDH282:00) and target construct (pVDH282-PetE:KNAT1). Primarily transformed lines (T₀) precociously manifested the phenotype of lobed leaves, caused by KNAT1 ectopic expression in simple leafed species [13]. Independent transgenic lines were selected for a single T-DNA locus by southern blot analysis and selfed in controlled conditions to reach T-DNA homozygosis. “Luxor” has a low natural out-crossing rate (0.5%) in glass house high density layout cultivation [14]. Transgenic lines with empty constructs did not differ from the wild type at the phenotypical level, whilst several independent PetE:KNAT1 lines shared similar altered traits. In this work, plants of the T₆ progeny (homozygous for one T-DNA) derived from controlled self-pollination of the PetE:KNAT1-177 line (mild phenotype) were compared to non transformed Luxor individuals descending from the original mother plant used for transformation. The metabolic analyses were also performed on two other independent GM lines and no significant differences were observed for the target compounds (unpublished data) under the conditions reported here, suggesting that metabolic changes reflected transgene expression rather than transformation processes.

Seeds of both PetE:KNAT1 (homozygous for one T-DNA copy, T₆ generation) and conventional lettuce cv "Luxor" (short day nunhems etc.) were sown on April 16^th. The growth rate of the PetE:KNAT1 plants is slower than that of the conventional type [23]. The first sampling was performed when the 4^th leaf was completely expanded, which occurred from the 3^rd to 4^th week after sowing (WAS) for controls and from the 5^th to 6^th WAS for transgenics. The leaflets (n = 4) of three plants formed one pool. Fifteen pools and 11 pools of control and transgenic plants were analyzed, respectively. The second sampling was performed at head stage, which occurred from 11 and 12 WAS for controls and GM lines, respectively. The leaves (n = 6) borne in the middle of rosette of one distinct plant formed one pool. Fourteen and nine pools of control and transgenic lines were collected. The samples were lyophilized according to the procedure reported previously [8] and stored at -20 °C. Seeds of both PetE:KNAT1 and conventional lettuce cv "Luxor" were germinated in trays (ca. 3 dm³ per well) containing soil, peat and sand into 1:1:1 ratio. At the 8^th leaf, plants were put into 10³ cm³ pots and then into 20³ cm³ from ca. 25 leaves onwards in peat and soil (1:3) plus 2 g of fertilizer containing 12% N (5% NH₄⁺ and 7% NO₃^-) per kilogram of substrate. Plants were grown in the greenhouse under a natural light/dark (16/8 h) cycle at 100 μmol m^-2 s^-1 flux of photosynthetically active radiation (PAR), at 25 °C.

Water-soluble extracts were prepared as reported previously [6]: 1 mL of a D₂O phosphate buffer, having 400 mM salt concentration, pD value of 6.5 and containing 0.3 mM of 3-(trimethyl-silyl)propionic-2,2,3,3-d₄ acid sodium salt (TSPA) - 0.05 mM EDTA, was mixed with 21.0 ± 0.2 mg of powdered vegetable tissues. The mixture was centrifuged at 10,000 rpm (5,600 g) for 7 min and the supernatant obtained was filtered and transferred into a standard 5 mm NMR tube.

NMR spectra of the extracts were recorded at 300 K on a Bruker AVANCE AQS600 spectrometer operating at the proton frequency of 600.13 MHz and equipped with a Bruker multinuclear z-gradient inverse probehead capable of producing gradients in the z-direction with a strength of 55.4 G/cm. Proton spectra were referenced to the signals of TSPA methyl group at δ = 0.00 ppm in D₂O phosphate buffer. The ¹H spectra of the aqueous extracts were acquired by co-adding 512 transients with a recycle delay of 2.5 s and 32K data points (acquisition time 40 min). The residual HDO signal was suppressed using a presaturation during the relaxation delay with a long single soft pulse. To avoid possible saturation effects, the experiment was carried out by using a 45° flip angle pulse of 8.0 μs.

2D NMR experiments, namely ¹H-¹H TOCSY and ¹H-¹³C HSQC, were performed using the same experimental conditions as previously reported [6].

The intensity, i.e. the peak height, of 22 ¹H resonances due to assigned water-soluble metabolites were measured with respect to the intensity of TSPA signal (0.3 mM) used as internal standard and normalized to 1000. The twenty two measured resonances are due to Ile (1.02 ppm), Val (1.05 ppm), lactic acid (1.32 ppm), Thr (1.34 ppm), Ala (1.49 ppm), acetic acid (1.92 ppm), Glu (2.07 ppm), GABA (2.30 ppm), malic acid (2.38 ppm), succinic acid (2.41 ppm), Gln (2.48 ppm), citric acid (2.53 ppm), Asp (2.80 ppm), Asn (2.91 ppm), choline (3.21 ppm), β-glucose (3.24 ppm), myo-inositol (3.29 ppm), fructose (4.02 ppm), tartaric acid (4.34), caffeoyl-tartaric acid (5.31 ppm), sucrose (5.42 ppm), and fumaric acid (6.52 ppm)

The statistical treatment of the NMR data was performed using the STATISTICA package for Windows (version 5.1, 1997). Before performing the statistical analysis, all the 22 selected variables were mean-centred and each variable was divided by its standard deviation (autoscaling). Principal component analysis (PCA) was performed using 22 variables. In the tree clustering analysis, the Euclidean distance and the complete linkage method were used to measure the distance between samples in the 22-dimensional space and between clusters, respectively. The effects and interactions reported in the Table 1 were statistically significant within the 95% confidence interval.