Skip to main content

Thank you for visiting You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Structured plant metabolomics for the simultaneous exploration of multiple factors


Multiple factors act simultaneously on plants to establish complex interaction networks involving nutrients, elicitors and metabolites. Metabolomics offers a better understanding of complex biological systems, but evaluating the simultaneous impact of different parameters on metabolic pathways that have many components is a challenging task. We therefore developed a novel approach that combines experimental design, untargeted metabolic profiling based on multiple chromatography systems and ionization modes, and multiblock data analysis, facilitating the systematic analysis of metabolic changes in plants caused by different factors acting at the same time. Using this method, target geraniol compounds produced in transgenic tobacco cell cultures were grouped into clusters based on their response to different factors. We hypothesized that our novel approach may provide more robust data for process optimization in plant cell cultures producing any target secondary metabolite, based on the simultaneous exploration of multiple factors rather than varying one factor each time. The suitability of our approach was verified by confirming several previously reported examples of elicitor–metabolite crosstalk. However, unravelling all factor–metabolite networks remains challenging because it requires the identification of all biochemically significant metabolites in the metabolomics dataset.


Secondary metabolites play an important role in the adaptation of plants to environmental stress1. Plants react to exogenous factors such as nutrients, hormones and light through signalling pathways that induce downstream stress responses including the modulation of gene expression and the regulation of a broad range of biochemical processes, resulting in the remodelling of metabolism2. Key signalling molecules include Ca2+, nitrates, phosphates, 2,4-dichlorophenoxyacetic acid (2,4-D), naphthalene acetic acid (NAA), indole acetic acid (IAA), 6-benzylaminapurine (BAP), kinetin, abscisic acid (ABA), jasmonates, salicylic acid, gibberellic acid (GA3), ethylene, polyamines, cyclic nucleotides (cAMP and cGMP) and diacylglycerol2,3,4. The accumulation of metabolites in stressed plants may also have economic significance1 because signalling components or elicitors can be used to trigger the production of secondary metabolites in whole plants or plant cell and tissue cultures5.

Design of experiments (DOE) approaches are used to study the influence of multiple factors simultaneously, allowing the influence of each factor to be determined regardless of other parameters while maintaining independence between the assessment of different effects. This contrasts with the classic ‘one factor at a time’ approach, which is laborious, time consuming and lacks the ability to provide a global picture of molecular events6. Factorial designs have recently flourished in plant biology, where they have been used to optimize cultivation parameters for cell and tissue cultures7,8 and to increase the yield of metabolites9,10 or recombinant proteins11 by medium optimization. However, most of these applications of DOE featured a small number of response variables when describing the corresponding system or process. A much more comprehensive multivariate strategy is needed to identify multiple inducible biomarkers in the plant metabolome following the application of diverse elicitors, so the combination of DOE and metabolomics is an attractive approach for the systematic evaluation of changes in plant secondary metabolism12.

Metabolomics generates large, multi-dimensional datasets using automated analytical procedures such as gas chromatography or high-pressure liquid chromatography coupled to mass spectrometry (GC-MS and HPLC-MS). It is therefore necessary to reduce the dimensionality of the data using multivariate statistical methods. The complexity of data mining is enhanced further when the data originate from several sources (e.g. complementary chromatography systems or ionization modes) and data fusion strategies are therefore required. An additional difficulty is encountered when multiple input factors are varied simultaneously, because different sources of variation are mixed. The importance of multiple simultaneous metabolic effects has been underestimated in the past and here we addressed this challenge by combining several orthogonal techniques: reversed-phase ultra-high-pressure liquid chromatography (RP-UHPLC) with positive and negative electrospray ionization (ESI) modes, and hydrophilic interaction liquid chromatography (HILIC), both coupled to time of flight mass spectrometry (TOF-MS) to achieve greater coverage of the metabolome.

Several strategies have been developed for the simultaneous analysis of multiple datasets. The proposed data modelling approach is an extension of the multiple kernel learning method to orthogonal partial least squares discriminant analysis (OPLS-DA), i.e. consensus OPLS-DA, which combines data blocks using the weighted sum of X·XT product association matrices corresponding to their linear kernel13. The OPLS-DA framework is advantageous for data interpretation because relevant metabolic variations are associated with predictive components, whereas unrelated variation is summarized in so-called orthogonal components14. In consensus OPLD-DA, the block weighting is based on modified RV-coefficients so that the Y response orientates the consensus kernel towards improved predictability. Cross-validation is carried out to assess the optimal model size and avoids overfitting, using DQ2 (an adaptation of the conventional Q2 value) for discriminant analysis15.

To our knowledge, this is the first systematic investigation of metabolic remodelling in plants following simultaneous multi-factorial treatment. This novel combination of metabolomics and experimental design, associated with the simultaneous analysis of multiblock omics data, is a powerful approach that allows us to unravel the metabolic responses in transgenic tobacco cells at a global level when diverse input factors such as macronutrients, plant growth regulators and light are varied simultaneously. Furthermore, this high-throughput screening system can be used for process optimization with metabolically engineered cell lines. Herein we hypothesize that product optimization using the simultaneous exploration of multiple factors may achieve more accurate and reproducible results than the assessment of one factor at a time.


UHPLC-QTOF-MS fingerprinting

The acquisition of high-quality metabolomics data is an essential aspect of metabolic profiling because it facilitates the identification step. The UHPLC-QTOF-MS gradient conditions we applied allowed us to monitor more than 3500 features in the m/z range 100–1200 based on RP-UHPLC (ESI+ and ESI ion detection modes) and HILIC (ESI+ ion detection mode). Pareto scaling was applied as data normalization. The multiblock data fusion strategy we used allows the integration of datasets originating from different ionization (ESI+/ESI) and separation (RP/HILIC) modes, yielding 1500, 1366 and 1368 variables for the RP/ESI+, RP/ESI and HILIC/ESI+ blocks, respectively. Some of the signals were present in all blocks, whereas others were found in only one or two data blocks and thus provided complementary information13. The target metabolites in methanol extracts, obtained from tobacco suspensions expressing Valeriana officinalis geraniol synthase (VoGES) gene, were identified based on mass fragmentation and the comparison of characteristic m/z species with internal or publicly accessible natural product databases.

Consensus OPLS-DA data modelling and metabolite identification

A supervised data mining approach (consensus OPLS-DA) was applied to determine the distinct metabolic changes caused by the simultaneous modulation of diverse input factors, and leave-one-out cross-validation (LOOCV) was carried out to evaluate the appropriate number of orthogonal components based on the DQ2 parameter. A series of 1000 permutation tests was performed for each model by randomizing the original Y class response in order to assess the statistical validity of the models13.

Significant models were obtained for most of the factors with a prediction accuracy (PA) of 87–100% and a statistical significance of p < 0.01. Lower prediction accuracy was observed for GA3 (PA = 62.5%, p = 0.013) and ethephon (PA = 60.4%, p = 0.015), whereas KH2PO4 (PA = 57.3%, p = 0.225) and MgSO4 (PA = 47.9%, p = 0.62) were found to be statistically non-significant (Table 1). Random permutations of the design matrix simulate data under the null hypothesis, i.e. no effect of the experimental factor under study. Because similar or higher prediction accuracies could be achieved from randomly permutated designs, we decided to omit KH2PO4 and MgSO4 from subsequent experiments aiming to identify metabolites and to determine their biological importance.

Table 1 Performance indices of the consensus OPLS-DA models evaluated for each of the screened factors.

Next we used S-plots to highlight relevant metabolites based on their contributions to the model, reflecting both their amplitude of variation and their reliability14. This visualization method helps to identify biochemically significant metabolites based on their position in the S-plots. The ideal biomarker has a high covariance magnitude and high correlation reliability. This low risk of spurious correlations corresponds to the upper right quadrant (area 1) or lower left quadrant (area 4) of each S-plot. The biomarkers located in area 1 are associated with metabolites that become more abundant when the factor strength is enhanced or less abundant when the factor strength is reduced, i.e. an upregulation effect (metabolic levels(+)/factor(+) or metabolic levels(−)/factor(−)). Similarly, the biomarkers in area 4 are linked to metabolites that become less abundant when the factor strength is enhanced or more abundant when the factor strength is reduced, i.e. a downregulation effect (metabolic levels(−)/factor(+) or metabolic levels(+)/factor(−)). Because all three data blocks (RPLC NEG, RPLC POS and HILIC) are integrated in a single model for each factor, ions detected by each analytical protocol can be associated in a combined S-plot.

High-resolution QTOF-MS/MS profiling allowed 45 compounds represented by some of the most significant features with respect to the factors we investigated (i.e. they were located in areas 1 and 4 of the S-plots) to be tentatively identified, at least to the level of the compound class (Table 2). However, the complete identification of all other relevant metabolites remains challenging and these preliminary results may serve as a starting point for the further targeted isolation and purification of the metabolites of interest.

Table 2 List of identified metabolites.

Simultaneous treatments reveal nutrient–metabolite and elicitor–metabolite networks

Our primary goal was to identify metabolites that respond to each factor treatment when multiple factors are modulated simultaneously. Using this approach, we avoided the restriction of experimental results caused by the artificially narrow experimental conditions of the one factor at a time approach. In the following sections, bold numbers in parentheses represent the compounds from the list of identified metabolites (Table 2).


KNO3 treatment (Fig. 1) increased the abundance of several ions representing glutamine (8) and scopolin (9), but also single ions representing a 2,4-D metabolite (14) (tR = 8.85 min; m/z 559.2755) and hexosyl geranidiol (5) (tR = 5.96 min; m/z 355.1749). In contrast, KNO3 caused the extracellular loss of potassium ions, detected as K+ (KHCOO)n adducts (44), and inhibited the formation of nicotine (45), hydroxysphingosine (10) and C16:3 monoacylglycerol (29). Single ions representing the following compounds were also less abundant following treatment with KNO3: trihydroxy-C18:2 acids (37, 38) (tR = 11.36/11.18 min; m/z 327.2170/327.2171), a dihydrozeatin (DHZ) adduct (22) (tR = 3.26 min; m/z 354.1794), a cyclanilide metabolite (33) (tR = 9.26 min; m/z 620.0896) and BAP glycoside (26) (tR = 4.08 min; m/z 388.1647).

Figure 1

KNO3 S-plot showing the distribution of metabolites in transgenic tobacco cell cultures exposed to different combinations of environmental factors, revealing metabolites in areas 1 (upper right) and 4 (lower left) that are the most sensitive to changes in KNO3 levels.

Numbers refer to compounds listed in Table 2.

NH4NO3 treatment (Fig. S1) induced the synthesis of the monoterpenoid derivatives malonyl-hexosyl-geraniol (4), pentosyl-hexosyl-geraniol (2) and malonyl-hexosyl-geranidiol (7) as well as scopolin (9). Two single ion species representing BAP glycosides were also more abundant following treatment with NH4NO3, i.e. BAP glucoside (26) (tR = 3.94 min; m/z 388.1634) and BAP riboside (27) (tR = 6.59 min; m/z 358.1532), as well as a salicylic acid glucoside (30) (tR = 3.31 min; m/z 323.0759). In contrast, auxin metabolites representing NAA (18) and 2,4-D (14) as well as single ions representing conjugated dihydroxy-C18:2 fatty acid (43) and hexosyl-hexosyl-geraniol (6) were less abundant following treatment with NH4NO3.

The formation of three metabolites was strongly induced by calcium treatment: hydroxysphingosine (10), scopolin (9) and hexosyl-hexosyl-geraniol (6). Adenosine (11) synthesis was also stimulated by calcium and was represented by a single ion (tR = 2.65 min; m/z 268.1082) (Fig. S2). Several metabolites became less abundant following calcium treatment, namely C16:3 monoacylglycerol (29), geraniol derivatives produced by the heterologous geraniol synthase (1, 2, 4 and 7), and a fatty acid with conjugated triene (40) (tR = 12.15 min; m/z 325.2013).


Numerous metabolites were influenced by IAA (Fig. S3). We observed an increase in the abundance of indole-3-carboxylic acid glucoside (17) and K+ (44). The monoterpenoid derivatives pentosyl-hexosyl-geraniol (2) and malonyl-hexosyl-geraniol (4), a DHZ glycoside (21) (tR = 6.56 min; m/z 384.1906) and its adduct (22), C16:3 monoacylglycerol (29) and nicotine (45) (tR = 3.8096 min; m/z 163.1254) were also more abundant following treatment with IAA. The formation of other metabolites was inhibited, particularly adenosine (11), scopolin (9), BAP glucoside (26), and the geraniol derivatives pentosyl-hexosyl-geranidiol (1) (tR = 5.76 min; m/z 487.2159), hexosyl-geranidiol (5) and malonyl-hexosyl-geranidiol (6) (tR = 8.96 min; m/z 501.2312).

Indole-3-butyric acid (IBA) treatment (Fig. S4) induced the synthesis of IBA glycoside (20), geraniol glycosides (2, 4) and C16:3 monoacylglycerol (29) but caused the loss of other metabolites including geraniol derivatives (1, 5, 6).

NAA treatment (Fig. S5) mainly increased the abundance of its own derivatives (18, 19). The following metabolites became less abundant: several single ions characteristic of fatty acids with conjugated triene (4042), trihydroxy-C18:2 acid (37), indole-3-carboxylic acid glucoside (17) (tR = 4.68 min; m/z 346.0923), pentosyl-hexosyl-geraniol (2), malonyl-hexosyl-geraniol (4) and nicotine (45).

The formation of several compounds was induced by 2,4-D treatment (Fig. S6): 2,4-D metabolites (1416), adenosine (11) and the geraniol derivatives (1, 5, 6). Pentosyl-hexosyl-geraniol (2), malonyl-hexosyl-geraniol (4) and C16:3 monoacylglycerol (29) became less abundant.

Our observations show that naturally occurring (endogenous) auxins such as IAA and IBA upregulate (2 and 4) and downregulate (1, 5 and 6) the synthesis of geraniol glycosides, whereas the synthetic auxins (NAA and 2,4-D) have exactly the opposite effects on these metabolites.


The presence of kinetin (Fig. S7) strongly induced the synthesis of kinetin glucoside (24) and kinetin riboside (25), and to a lesser extent the geraniol derivatives pentosyl-hexosyl-geraniol (2), malonyl-hexosyl-geraniol (4) and malonyl-hexosyl-geranidiol (7) as well as trihydroxy-C18:2 acid (37). DHZ treatment (Fig. S8) induced the formation of three derivatives: DHZ glycoside (21), DHZ adduct (22) and DHZ metabolite (23). BAP treatment (Fig. S9) resulted in the appearance of three derivatives: BAP glucoside (26), BAP riboside (27) and BAP ribotide (28).

Other plant growth regulators

Methyljasmonate (MeJa) treatment (Fig. S10) only induced the formation of C16:3 monoacylglycerol (29) to a significant extent, but we also observed smaller increases in the abundance of a cyclanilide metabolite (33), hydroxysphingosine (10), nicotine (45), unidentified fatty acids with conjugated triene (41, 42), trihydroxy-C18:2 acids (37, 38), adenosine (11) and a 2,4-D metabolite (14). The only heterologous monoterpenoids induced by MeJa were pentosyl-hexosyl-geraniol (2) and malonyl-hexosyl-geranidiol (7). The formation of scopolin (9) was strongly inhibited, whereas the synthesis of hexosyl-hexosyl-geraniol (6), BAP riboside (27) and abscisic acid glycoside (35) was more weakly downregulated.

Salicylic acid treatment (Fig. S11) mainly induced the formation of its polar derivatives salicylic acid glucoside (30) and salicylic acid dihexosyl-glucoside (31), but also C16:3 monoacylglycerol (29), trihydroxy-C18:2 acid (38) and the unidentified fatty acids with conjugated triene (4143). The monoterpenoids pentosyl-hexosyl-geraniol (2) and malonyl-hexosyl-geraniol (4) were also more abundant. Salicylic acid treatment inhibited the formation of a cyclanilide metabolite (32), a 2,4-D metabolite (14), scopoletin (12), a scopoletin derivative (13), hexosyl-hexosyl-geraniol (6) and BAP riboside (27).

GA3 treatment (Fig. S12) induced the formation of C16:3 monoacylglycerol (29), pentosyl-hexosyl-geranidiol (1), trihydroxy-C18:2 acid (37), BAP glucoside (26), adenosine (11), malonyl-hexosyl-geranidiol (7), hexosyl-geranidiol (5), NAA metabolites (18, 19) and the unidentified fatty acid with conjugated triene (42). It inhibited the formation of scopolin (9), a 2,4-D metabolite (14), hexosyl-geraniol (3), hexosyl-hexosyl-geraniol (6), a DHZ adduct (22) and a cyclanilide metabolite (32).

Ethephon treatment (Fig. S13) induced the formation of C16:3 monoacylglycerol (29), pentosyl-hexosyl-geraniol (2), BAP glucoside (26), trihydroxy-C18:2 acid (38), scopolin (9) and malonyl-hexosyl-geraniol (4), but inhibited the formation of a DHZ adduct (22), adenosine (11), hydroxysphingosine (10), hexosyl-geranidiol (5), hexosyl-hexosyl-geraniol (6) and a cyclanilide metabolite (32).

Cyclanilide treatment (Fig. S14) induced the formation of its own metabolites (3234) as well as one NAA metabolite (18), hexosyl-hexosyl-geraniol (6), unidentified fatty acids with conjugated triene (4042), and trihydroxy-C18:2 acids (37, 38). The formation of numerous compounds was downregulated, including pentosyl-hexosyl-geraniol (2), malonyl-hexosyl-geraniol (4), C16:3 monoacylglycerol (29), pentosyl-hexosyl-geranidiol (1), conjugated dihydroxy-C18:2 fatty acid (43), scopolin (9), scopoletin (12) and malonyl-hexosyl-geranidiol (7).

ABA treatment (Fig. S15) resulted in the accumulation of ABA glycoside (35), an ABA metabolite (36) and its glycoside (39), trihydroxy-C18:2 acids (37, 38), and to a lesser degree the unidentified fatty acids with conjugated triene (4042), pentosyl-hexosyl-geranidiol (1) and malonyl-hexosyl-geranidiol (7). The abundance of DHZ (22), a 2,4-D metabolite (14), a cyclanilide metabolite (32), kinetin riboside (25), scopolin (9), and DHZ glycoside (21) declined in response to ABA.


Light was the only physical factor we included in our experimental design, and it was associated with an increase in the levels of adenosine (11), two unidentified fatty acids with conjugated triene (41, 42), trihydroxy-C18:2 acids (37, 38), malonyl-hexosyl-geranidiol (7), and to a lesser extent glutamine (8), a 2,4-D metabolite (14) and C16:3 monoacylglycerol (29). Light inhibited the production of alkaloids, i.e. scopoletin (12), a scopoletin derivative (13) and scopolin (9), and the monoterpenoids pentosyl-hexosyl-geraniol (2), hexosyl-geraniol (3), and malonyl-hexosyl-geraniol (4). A DHZ adduct (22) and hydroxysphingosine (10) were also less abundant under strong illumination (Fig. S16).


Cluster analysis provided a global overview of regulation events that follow changes in the experimental factors. This approach allows the grouping of metabolites with similar ion features using a dendrogram, and a heat map summarizes the contribution of each of the factors in the context of each identified metabolite and thus enables the visualization of upregulation and downregulation in response to different treatments (Fig. 2).

Figure 2

Cluster analysis and heat map.

Both geranidiol glycosides (1 and 5) are located in a small cluster, whereas the third geranidiol derivative (7), an esterified monoglycoside, is found in a neighbouring cluster (Fig. 2). The geraniol metabolites (1 and 5) are upregulated by 2,4-D, DHZ, GA3 and ABA and downregulated by NH4NO3, CaCl2, IAA, IBA, BAP, salicylic acid and ethephon, whereas the upregulation of compound (7) is related to NH4NO3, kinetin, MeJa, GA3, ABA, and light, and its downregulation is associated with CaCl2, NAA, BAP and cyclanilide. The geraniol glycosides (2, 3 and 4) are located in a common larger cluster. These three metabolites are upregulated by IAA, IBA, kinetin and ethephon, but downregulated by CaCl2, 2,4-D, NAA and light. Compounds (2 and 4) are distinguished from hexosyl-geraniol (3) mainly by the action of NH4NO3, i.e. (2 and 4) are upregulated whereas (3) is downregulated by NH4NO3. The heat map also shows the upregulation of hexosyl-hexosyl-geraniol (6) by CaCl2, 2,4-D, BAP and cyclanilide, and its downregulation by NH4NO3, IAA, IBA, DHZ, MeJa, salicylic acid, GA3, ethephon and ABA.


The combination of fractional factorial design and consensus OPLS-DA methods allowed us to systematically explore the effect of multiple factors on plant metabolism, using transgenic tobacco cell cultures as a model system. This simultaneous application of treatment stress assesses all experimental factors under diverse conditions that could occur in nature. We tentatively identified 45 constituents in areas 1 and 4 of the S-plots following the fractionation and analysis of plant cell extracts by UHPLC-QTOF-MS, corresponding to metabolites whose abundance changed substantially in response to the experimental factors. These metabolites represented multiple classes of natural products: monoterpenoids, nitrogen-containing compounds, coumarins, fatty acids and their esters, and derivatives of phytohormones and plant growth regulators used as additives in the experiments.

Geraniol and its glycosides do not occur naturally in tobacco plants and their presence in our samples reflects the activity of the stably integrated VoGES gene16. However, the glycosylation profile of geraniol produced by our cell suspension cultures differed to that observed in whole plants. The cells produced seven distinct geraniol glycosides whereas 19 variants were produced by transgenic tobacco plants and Nicotiana benthamiana leaves used for transient expression. The acetylated glycosides produced at later stages of plant development were not monitored in our plant cell cultures. The cell suspension cultures produced geraniol monoglycosides and diglycosides, whereas the whole plants also produced geraniol glycosides with three or more sugar adducts. However, our cell suspension cultures accumulated geranidiol derivatives that are not produced in agroinfiltrated or transgenic plants, which instead produce geranic acid glycosides16.

Nitrogenous compounds are found in tobacco cells because they have multiple core metabolic functions and are also precursors in the biosynthesis of tobacco alkaloids17. Phytoalexins defend plants against biotic and abiotic stress18. The coumarin scopoletin (12) is one of the phytoalexins produced in tobacco19,20,21. C16:3 monoacylglycerol (29) is a glyceride which can be formed by the esterification of glycerol with one fatty acid or by enzymatic hydrolysis of a fatty acid from diacylglycerol by the action of diacylglycerol lipase. Diacylglycerol acts as a signalling molecule during plant development and in response to stress tolerance, nutrient deficiency and other environmental stimuli22,23,24.

Our metabolomic analysis showed that plant cells react strongly to phytohormones and plant growth regulators. Plants limit the impact of harmful xenobiotic compounds by hydroxylation, glutathione conjugation, glycosylation, malonylation and sulfonylation25,26. Most of the phytohormone and growth regulator derivatives we identified were polar glucosylated products, which are generally more soluble than the parent molecule thus facilitating elimination. We also detected malonylated geraniol glycosides, confirming that malonylation is one of the key mechanisms used by tobacco cells to metabolize xenobiotic compounds27.

The production of glutamine was strongly upregulated by KNO3 treatment in our experiment. Nitrate is assimilated by plant cells from nitrite and ammonium, and is then converted into the amino acid glutamine28. The protein kinase CIPK23 is involved in both nitrate and potassium signalling29. CIPK23 phosphorylates nitrate transporter NPF6.3 after interacting with the calcineurin-B-like protein CBL9, and reduces nitrate uptake capacity in the presence of high external NO3 concentrations, whereas the CBL1-CIPK23 and CBL9-CIPK23 complexes activate the K+ channel AKT130. A monoacylglycerol derivative (29) was also less abundant following KNO3 treatment. This may be a hydrolysis product of diacylglycerol which activates protein kinase C and induces nitrate reductase gene expression31.

The NO3/NH4+ ratio in the culture medium influences the activity of auxins and cytokinins32. We observed an increase in the abundance of ions representing 2,4-D and the loss of ions representing DHZ metabolites and BAP glycoside in the KNO3 S-plot, but an increase in the abundance of BAP glycosides and the loss of ions representing 2,4-D and NAA metabolites in the NH4NO3 S-plot. This indicates that the balance between NO3 and NH4+ ions may affect phytohormone sensitivity.

NH4NO3 and CaCl2 had the most significant impact on the biosynthesis of geraniol glycosides among the inorganic factors we tested. NH4NO3 induced the formation of geraniol glycosides (2, 4, 7) but inhibited the formation of hexosyl-hexosyl-geraniol (6), whereas calcium showed the opposite behaviour. Higher concentrations of useable nitrogen also enhanced the accumulation of linalool and citronellol by Saccharomyces cerevisiae33. Geraniol blocks calcium and potassium channels in mammalian cells34 and similar cyclic nucleotide-gated ion channels are found in plants35.

Ca2+ induced the hydroxylation of sphingosine and adenosine agreeing with previous observations that sphingosine-1-phosphate increases cytosolic free Ca2+36 and cyclic adenosine monophosphate regulates calcium channels in the plasma membrane of Arabidopsis thaliana leaf guard and mesophyll cells37.

Scopolin and its 7-O-glucoside are key components of the abiotic stress response18 and the abundance of both compounds increased following treatment with all three statistically significant nutrients in our study. The accumulation of scopoletin in tobacco cells and its conversion to a glucoside is also induced by 2,4-D21. Ions representing scopolin were also detected following the treatment of our cells with 2,4-D. Scopoletin synthesis was strongly inhibited by MeJa concurring with data showing that scopoletin biosynthesis induced by Alternaria alternata is strongly dependent on jasmonic acid but not ABA, although MeJA does not induce scopoletin production in the absence of A. alternata38. In our cells, the formation of scopoletin was also inhibited by GA3.

We detected nicotine produced in trace amounts by our transgenic tobacco cell cultures, which comprise green (photosynthesizing) cells derived from the aerial parts of the plant, although de novo nicotine synthesis takes place mainly in the roots39. MeJa induced nicotine production in our tobacco cells in a similar manner as previously shown for N. attenuata39. The downregulation of nicotine production we observed following the treatment with KNO3 and NAA agrees with previous reports for cultured tobacco callus, and the effect of K+ is probably mediated by NAA40. We also observed that nicotine biosynthesis was induced by IAA but moderately suppressed by 2,4-D41. The combined effect of MeJa, auxins and K+ on the regulation of nicotine synthesis suggests that multiple factors contribute to the same process.

The biosynthesis of monoterpenoid glycosides appeared to be influenced by auxins and cytokinins, perhaps reflecting the antagonistic crosstalk between these two phytohormone classes42. Our experimental results support the idea that phytohormones function in a complex network involving the different hormonal pathways but that there is also elaborate crosstalk with nutrients and elicitors. The auxin-sensitive signalling protein SHY2 is regulated by the cytokinin-induced protein ARR1 (Arabidopsis response regulator), which in turn is repressed by gibberellin thus connecting three hormones in one network42. This may explain why our GA3 S-plot contained ion traits that were also affected by auxins and cytokinins. We also observed evidence for ethylene/cytokinin and cytokinin/ABA crosstalk42.

Light has a potent effect on monoterpenoid metabolism by modulating the expression of monoterpenoid synthase genes, controlling precursor synthesis, and affecting constitutive promoter activity43,44. Geraniol glycosides were also influenced by light in our cell cultures. Light induced the formation of malonyl-hexosyl-geranidiol (7) but suppressed the formation of pentosyl-hexosyl-geraniol (2), hexosyl-geraniol (3) and malonyl-hexosyl-geraniol (4).

We have developed a systematic approach, which implements an experimental design strategy in the context of metabolomics to account for the diverse factors applied simultaneously to plant cells. This is a valuable method for the investigation of complex environmental stress and its impact on plant metabolism by optimizing the number of experiments needed to assess the factors. Our approach significantly reduces the time and effort required for testing by using consensus OPLS-DA models to evaluate and interpret metabolic changes caused by the simultaneous application of diverse ecological factors. This systematic workflow may facilitate the discovery and characterization of factor–nutrient–elicitor networks and appropriate biomarkers. Finally, we conclude that this novel approach should be able to streamline process optimization for the reproducible production of any secondary metabolite in plant cell cultures by the simultaneous exploration of multiple factors rather than the assessment of one factor at a time.

Materials and Methods

Plant cell cultures, treatments and harvesting

We used tobacco (N. tabacum cv. Samsun NN) transgenic cell suspension cultures, expressing stably V. officinalis geraniol synthase. The cell cultures were initiated and maintained as previously described8. Two levels (low/high) were used for each of the factors selected for analysis. The low level of each macronutrient in the plant cell culture medium was based on classical Murashige and Skoog (MS) medium11,45 whereas the high level was based on our recent medium optimization study, although the concentrations of NH4NO3 were reversed11,45. The specific low/high concentrations were prepared as follows: KNO3 (18.79/70.16 mM), CaCl2 (2.99/10.84 mM), KH2PO4 (1.25/2.72 mM), MgSO4 (1.5/3.0 mM) and NH4NO3 (4.24/20.61 mM). For the auxins, cytokinins and plant regulators (IAA, IBA, NAA, 2,4-D, kinetin, DHZ, BAP, MeJa, salicylic acid, GA3, ethephon, cyclanilide and ABA) the low and high levels were set to 0 and 10 μM, respectively. Finally, the low and high levels of light were set to 11.50 and 35.62 μmol/cm2/s, respectively8. The factor levels are summarized in Table S1.

The following cultivation conditions were used: flask volume 50 ml (Erlenmeyer glass flasks), filling volume 25 ml, inoculum size 1.4 g fresh weight (FW), triacetyl-β-cyclodextrin concentration 2 mM, and shaking frequency 180 rpm8. The plant cell suspension cultures were grown for 9 days before the cells were harvested, then filtered twice under vacuum and frozen at −20 °C. The cultures were elicited with phytohormones and plant growth regulators 6 days post-inoculation. The experiment was conducted at 26 °C with a 16-h photoperiod in an ISF1-X shaker (Kühner AG, Birsfelden, Switzerland).

Sample preparation

The samples were extracted as previously described46. Briefly, ~100-mg aliquots of plant material were mixed with methanol (1:3 w/v) in a shaker mill (TissueLyser, Retsch, Haan, Germany) and pulverized with a steel ball at 25 beats per second for 1 min. Homogenized samples were sonicated (15 min), centrifuged (1750 × g, 10 min, 25 °C) and supernatants were passed through a 0.2-μm membrane filter.


The samples prepared above were analysed on an Acquity UPLC system coupled to a QTOF Premier MS detector (Waters, Milford, MA, USA). For RP-UHPLC, a Waters Acquity BEH C18 column (2.1 × 150 mm, 1.7 μm) was used with water containing 0.1% (v/v) formic acid (A) and acetonitrile containing 0.1% (v/v) formic acid (B) as eluents applied as the following gradient: 0 min, 5% B; 23.5 min, 80% B; 24 min, 96% B; 26 min, 96% B. The flow rate was 0.4 ml/min with the column temperature set to 45 °C. HILIC separation was carried out using a Waters Acquity HILIC column (2.1 × 150 mm, 1.7 μm) with 33 mM aqueous ammonium formate, pH 4.5 (A) and acetonitrile (B) as eluents applied as the following gradient: 0 min, 4% A; 3 min, 4% A; 17.5 min, 32% A; 18 min, 55% A; 20 min, 55% A. The flow rate was 0.5 ml/min with the column temperature set to 50 °C.

A pool was created by adding equal volumes from all samples to serve as a QC injection. Nine QC injections in total were distributed at regular intervals in the analytical batch. An acceptable variation was achieved for all peaks, including those with the highest intensity (coefficient of variation less than 40%).

Positive and negative ESI was applied in separate analytical runs (positive only for the HILIC method) with a desolvation gas flow of 780 l/h at 400 °C, a capillary voltage of 4.5 kV and a cone voltage of 45 V. Mass spectra were acquired over the m/z range 100–1200 in “W mode” using leucine enkephalin as a lock mass standard.

Raw data processing

Mass/retention time markers were extracted from the raw UHPLC-ESI-MS data using MarkerLynx XS v4.1 (Waters). The following method parameters were set: retention time window 1.6–24.6 min, m/z range 100–1200, XIC window 0.02 Da, peak smoothing activated, marker intensity threshold 30, mass window 0.04 Da, retention time window 0.2 min, noise elimination level 6.0, and deisotope data activated. All mass (m/z)/retention-time features related to the auxins, cytokinins and plant growth regulators provided as supplements in the plant cell culture media, as well as features derived from impurities in the LC eluents, were removed from the raw metabolomics datasets by removing all features detected in blank runs (solvent injection) or in analytical runs of the pure additives before data evaluation.

Experimental design

The experimental design was based on an orthogonal array with 96 runs created with the free open-source R package DoE.base47 as described in the supplementary material. Given the size of this experiment, tests for effects of 2-level factors at significance level 5% can detect effects of size “one standard deviation” with about 99% power, effects of size “half a standard deviation” with about 68% power, and effects of size “0.75 standard deviations” with about 95% power. The design was optimized to screen 14 factors by keeping the confounding of low-order effects minimal: all main effects are orthogonal to each other (orthogonal array), the design was based on an array with the lowest possible number of squared canonical correlations from three-factor sets equal to 148 and the factors were accommodated on columns of the base array such that confounding between main effects and two-factor interactions, and subsequently among two-factor interactions, was minimized8.

This fractional factorial design, with a randomized run order, was used to screen 12 two-level factors, one three-level factor and one four-level factor: we thus screened for the effects of light and 18 diverse substances representing macronutrients, auxins, cytokinins and elicitors. For the two-level factors, we investigated the presence of the high levels of NH4NO3, KNO3, CaCl2, KH2PO4, MgSO4, MeJa, salicylic acid, GA3, ethephon, cyclanilide, ABA and light. For auxins (four-level factor), exactly one of IAA, IBA, NAA or 2,4-D was present, whereas for cytokinins (three-level factor), exactly one of kinetin, DHZ or BAP was present. The experimental design with 96 runs is summarized in coded values in Table S2. The remaining potentially relevant confounding between main effects and two-factor interactions in terms of triples of factor comparisons are summarized in Fig. S17. For each such triple, the comparison between the levels of each factor in the triple might be affected by an interaction between the other two factors (e.g. an interaction between KNO3 and NH4NO3 might affect the assessment of the BAP vs. DH-z comparison for cytokinins). Sceptics might argue that this possibility for confounding is a reason to refrain from using an experimental design approach in favour of only changing one factor at a time (OFAT approach). However, if two-factor interactions are indeed relevant – as would be necessary for the experimental design approach to suffer from misleading conclusions in terms of factor level comparisons – the conclusions from an OFAT approach are also limited in the same manner and would be valid only for the exact settings at which the other factors have been fixed. Furthermore, to achieve a reasonable amount of replication, the OFAT approach would need a much larger number of experimental runs – e.g. 24 runs for the reference level combination (the number obtained for the four-level factor in the 96-run experiment) might be combined with 24 runs each for the other level of the 12 two-level factors (12*24), and 24 runs each for the other levels of the three-level and the four-level factors (5*24), resulting in a total of 18*24 = 432 runs instead of the 96 runs in our experiment.

Data processing and analysis

For each experimental factor, a consensus OPLS-DA model was built to relate the experimental metabolomics data (X) to a class matrix consisting of zeros and ones, filled according to the levels of each factor (Y). The columns of the experimental design were therefore used individually as a response matrix in the context of supervised analysis. For auxins and cytokinins, a response vector was generated individually for each hormone and filled with zero when the corresponding hormone was absent, whereas a value of one indicated its presence. The consensus OPLS algorithm implements data fusion based on multiple kernel learning. The joint analysis of multiple data tables is achieved by the combination of association matrices computed for each block. Therefore, requirements in terms of memory resources and computation time are minimized without information loss even if the experimental data include a large number of signals. A block-scaling step ensures fairness between blocks by offering equal starting chances to contribute to the model. RV coefficients are then computed to build a consensus matrix and orientate the model towards better prediction performance. A common subspace is built using a kernel version of the OPLS algorithm and the optimal number of orthogonal components is estimated by cross-validation. Because systematic variations are summarized using Y-predictive and Y-orthogonal components (OPLS framework), the interpretation of the multiblock model is straightforward. Like classical multivariate methods, a consensus score plot allows the distribution of the observations to be evaluated. Because linearity is maintained, variable loadings can easily be computed for biomarker discovery. The weight of each block in the projection also allows the role of each data source to be evaluated13.

The OPLS model can be summarized as follows:

where X contains the normalized metabolomic data from data block i (i  [RPPOS, RPNEG, HILICPOS]), Y represents a 0/1 indicator variable for experimental condition j (j  [NH4NO3, KNO3, CaCl2, KH2PO4, MgSO4, IAA, IBA, NAA, 2,4-D, kinetin, DHZ, BAP, MeJa, salicylic acid, GA3, ethephon, cyclanilide, ABA, light]), tp is the Y-predictive score matrix, pp is the Y-predictive loading matrix for X, to is the Y-orthogonal score matrix, po is the Y-orthogonal loading matrix for X, qp is the Y-predictive loading matrix for Y, and E and F are the residual matrices for X and Y, respectively. Note that the four indicator variables for the cytokinins sum to a column of “+1” entries, as do the three indicator variables for the auxins. Consensus OPLS-DA modelling was carried out under the MATLAB® v8 environment (The MathWorks, Natick, USA) with combinations of toolboxes and in-house functions, including the publicly available RV-coefficients MATLAB m-file49 and the KOPLS open source package50.

Cluster analysis

Subsets of metabolites sharing similar patterns were investigated using cluster analysis. For that purpose, the contribution (loading) of each ion feature associated with an identified metabolite was collected across all significant consensus OPLS-DA models and displayed in a dendrogram and a heat map. This strategy highlights upregulation and downregulation. Cluster analysis was carried out with the Bioinformatics Toolbox v4.2 under the MATLAB® v8 environment (The MathWorks) using Euclidean distances and the Ward aggregation method.

Factors with more than two levels

Auxins and cytokinins were two factors in our design associated with three and four levels, respectively. Exactly one auxin and one cytokinin were included in each run of the experimental design. Consequently, the four Y(j) indicator columns corresponding to auxins and the three Y(j) indicator columns corresponding to cytokinins are linearly dependent, as stated above. Our analysis included all indicator columns, because each is treated separately. This implies that downregulation or upregulation must be interpreted within the linearly-dependent groups, e.g. if the three auxins IAA, IBA and 2,4-D are identified as downregulators, the fourth (NAA) must be an upregulator (relative to the other auxins). This behaviour is clearly shown in the heat map and also implies analogous dependencies among the S-plots of the auxins and cytokinins, respectively.

Additional Information

How to cite this article: Vasilev, N. et al. Structured plant metabolomics for the simultaneous exploration of multiple factors. Sci. Rep. 6, 37390; doi: 10.1038/srep37390 (2016).

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.


  1. Ramakrishna, A. & Ravishankar, G. A. Influence of abiotic stress signals on secondary metabolites in plants. Plant Signal Behav 6, 1720–1731 (2011).

    CAS  PubMed  PubMed Central  Google Scholar 

  2. Tuteja, N. & Sopory, S. K. Chemical signaling under abiotic stress environment in plants. Plant Signal Behav 3, 525–536 (2008).

    PubMed  PubMed Central  Google Scholar 

  3. Karuppusamy, S. A review on trends in production of secondary metabolites from higher plants by in vitro tissue, organ and cell cultures. J Med Plants Res 3, 1222–1239 (2009).

    CAS  Google Scholar 

  4. Ramachandra Rao, S. & Ravishankar, G. A. Plant cell cultures: Chemical factories of secondary metabolites. Biotech Adv 20, 101–153 (2002).

    CAS  Google Scholar 

  5. Zhao, J., Davis, L. C. & Verpoorte, R. Elicitor signal transduction leading to production of plant secondary metabolites. Biotech Adv 23, 283–333 (2005).

    CAS  Google Scholar 

  6. Vaidya, B. K., Mutalik, S. R., Joshi, R. M., Nene, S. N. & Kulkarni, B. D. Enhanced production of amidase from Rhodococcus erythropolis MTCC 1526 by medium optimisation using a statistical experimental design. J Ind Microbiol Biotechnol 36, 671–678 (2009).

    CAS  PubMed  Google Scholar 

  7. Nas, M. N., Eskridge, K. M. & Read, P. E. Experimental designs suitable for testing many factors with limited number of explants in tissue culture. Plant Cell Tiss Organ Cult 81, 213–220 (2005).

    Google Scholar 

  8. Vasilev, N., Schmitz, C., Grömping, U., Fischer, R. & Schillberg, S. Assessment of cultivation factors that affect biomass and geraniol production in transgenic tobacco cell suspension cultures. PLoS One 9, e104620 (2014).

    ADS  PubMed  PubMed Central  Google Scholar 

  9. Prakash, G. & Srivastava, A. K. Statistical elicitor optimization studies for the enhancement of azadirachtin production in bioreactor Azadirachta indica cell cultivation. Biochem Eng J 40, 218–226 (2008).

    CAS  Google Scholar 

  10. Srivastava, S. & Srivastava, A. K. Statistical medium optimization for enhanced azadirachtin production in hairy root culture of Azadirachta indica. In Vitro Cell Dev Biol. Plant 48, 73–84 (2012).

    CAS  Google Scholar 

  11. Vasilev, N. et al. Optimization of BY-2 cell suspension culture medium for the production of a human antibody using a combination of fractional factorial designs and the response surface method. Plant Biotechnol J 11, 867–874 (2013).

    CAS  PubMed  Google Scholar 

  12. Marti, G. et al. Metabolomics reveals herbivore-induced metabolites of resistance and susceptibility in maize leaves and roots. Plant Cell Environ 36, 621–639 (2013).

    CAS  PubMed  Google Scholar 

  13. Boccard, J. & Rutledge, D. N. A consensus orthogonal partial least squares discriminant analysis (OPLS-DA) strategy for multiblock Omics data fusion. Anal Chim Acta 769, 30–39 (2013).

    CAS  PubMed  Google Scholar 

  14. Wiklund, S. et al. Visualization of GC/TOF-MS-based metabolomics data for identification of biochemically interesting compounds using OPLS class models. Anal Chem 80, 115–122 (2008).

    CAS  PubMed  Google Scholar 

  15. Westerhuis, J. A., van Velzen, E. J. J., Hoefsloot, H. C. J. & Smilde, A. K. Discriminant Q(2) (DQ(2)) for improved discrimination in PLSDA models. Metabolomics 4, 293–296 (2008).

    CAS  Google Scholar 

  16. Dong, L. et al. Characterization of two geraniol synthases from Valeriana officinalis and Lippia dulcis: Similar activity but difference in subcellular localization. Metab Eng 20, 198–211 (2013).

    PubMed  Google Scholar 

  17. Ghosh, B. Polyamines and plant alkaloids. Indian J Exp Biol 38, 1086–1091 (2000).

    CAS  PubMed  Google Scholar 

  18. Jeandet, P., Clement, C., Courot, E. & Cordelier, S. Modulation of phytoalexin biosynthesis in engineered plants for disease resistance. Int J Mol Sci 14, 14136–14170 (2013).

    PubMed  PubMed Central  Google Scholar 

  19. Grosskinsky, D. K. et al. Cytokinins mediate resistance against Pseudomonas syringae in tobacco through increased antimicrobial phytoalexin synthesis independent of salicylic acid signaling. Plant Physiol 157, 815–830 (2011).

    CAS  PubMed  PubMed Central  Google Scholar 

  20. Taguchi, G., Yazawa, T., Hayashida, N. & Okazaki, M. Molecular cloning and heterologous expression of novel glucosyltransferases from tobacco cultured cells that have broad substrate specificity and are induced by salicylic acid and auxin. Eur J Biochem 268, 4086–4094 (2001).

    CAS  PubMed  Google Scholar 

  21. Taguchi, G., Yoshizawa, K., Kodaira, R., Hayashida, N. & Okazaki, M. Plant hormone regulation on scopoletin metabolism from culture medium into tobacco cells. Plant Sci 160, 905–911 (2001).

    CAS  PubMed  Google Scholar 

  22. Dong, W., Lv, H., Xia, G. & Wang, M. Does diacylglycerol serve as a signaling molecule in plants? Plant Signal Behav 7, 472–475 (2012).

    CAS  PubMed  PubMed Central  Google Scholar 

  23. Okazaki, Y. & Saito, K. Roles of lipids as signaling molecules and mitigators during stress response in plants. Plant J 79, 584–596 (2014).

    CAS  PubMed  Google Scholar 

  24. Singh, A., Bhatnagar, N., Pandey, A. & Pandey, G. K. Plant phospholipase C family: Regulation and functional role in lipid signaling. Cell Calcium 58, 139–146 (2015).

    CAS  PubMed  Google Scholar 

  25. Kreuz, K., Tommasini, R. & Martinoia, E. Old enzymes for a new job (herbicide detoxification in plants). Plant Physiol 111, 349–353 (1996).

    CAS  PubMed  PubMed Central  Google Scholar 

  26. Sandermann, H. Jr. Higher plant metabolism of xenobiotics: the ‘green liver’ concept. Pharmacogenetics 4, 225–241 (1994).

    CAS  PubMed  Google Scholar 

  27. Taguchi, G. et al. Malonylation is a key reaction in the metabolism of xenobiotic phenolic glucosides in Arabidopsis and tobacco. Plant J 63, 1031–1041 (2010).

    CAS  PubMed  Google Scholar 

  28. Betti, M. et al. Glutamine synthetase in legumes: recent advances in enzyme structure and functional genomics. Int J Mol Sci 13, 7994–8024 (2012).

    CAS  PubMed  PubMed Central  Google Scholar 

  29. Tsay, Y. F., Ho, C. H., Chen, H. Y. & Lin, S. H. Integration of nitrogen and potassium signaling. Annu Rev Plant Biol 62, 207–226 (2011).

    CAS  PubMed  Google Scholar 

  30. Leran, S. et al. Nitrate sensing and uptake in Arabidopsis are enhanced by ABI2, a phosphatase inactivated by the stress hormone abscisic acid. Sci Signal 8, ra43 (2015).

    PubMed  Google Scholar 

  31. Chandok, M. R. & Sopory, S. K. ZmcPKC70, a protein kinase C-type enzyme from maize. Biochemical characterization, regulation by phorbol 12-myristate 13-acetate and its possible involvement in nitrate reductase gene expression. J Biol Chem 273, 19235–19242 (1998).

    CAS  PubMed  Google Scholar 

  32. George, E. F., Hall, M. A. & De Klerk, G.-J. Plant propagation by tissue culture. The background Vol. 1 (Springer, Dordrecht, The Netherlands, 2008).

  33. Carrau, F. M. et al. De novo synthesis of monoterpenes by Saccharomyces cerevisiae wine yeasts. FEMS Microbiol Lett 243, 107–115 (2005).

    CAS  PubMed  Google Scholar 

  34. de Menezes-Filho, J. E. et al. Geraniol blocks calcium and potassium channels in the mammalian myocardium: useful effects to treat arrhythmias. Basic Clin Pharmacol Toxicol 115, 534–544 (2014).

    PubMed  Google Scholar 

  35. Kaplan, B., Sherman, T. & Fromm, H. Cyclic nucleotide-gated channels in plants. FEBS Lett 581, 2237–2246 (2007).

    CAS  PubMed  Google Scholar 

  36. Wu, J. et al. Long-chain base phosphates modulate pollen tube growth via channel-mediated influx of calcium. Plant J 79, 507–516 (2014).

    CAS  PubMed  Google Scholar 

  37. Lemtiri-Chlieh, F. & Berkowitz, G. A. Cyclic adenosine monophosphate regulates calcium channels in the plasma membrane of Arabidopsis leaf guard and mesophyll cells. J Biol Chem 279, 35306–35312 (2004).

    CAS  PubMed  Google Scholar 

  38. Sun, H. et al. Scopoletin is a phytoalexin against Alternaria alternata in wild tobacco dependent on jasmonate signalling. J Exp Bot 65, 4305–4315 (2014).

    CAS  PubMed  PubMed Central  Google Scholar 

  39. Baldwin, I. T. Methyl jasmonate-induced nicotine production in Nicotiana attenuata: Inducing defenses in the field without wounding. Entomol Exp Appl 80, 213–220 (1996).

    CAS  Google Scholar 

  40. Piñol, M. T., Palazón, J., Altabella, T., Cusido, R. & Serrano, M. Effect of auxin on alkaloids, K+ and free amino acid content in cultured tobacco callus. Physiol Plantarum 65, 299–304 (1985).

    Google Scholar 

  41. Furuya, T., Kojima, H. & Syono, K. Regulation of nicotine biosynthesis by auxins in tobacco callus tissues. Phytochemistry 10, 1529–1532 (1971).

    CAS  Google Scholar 

  42. El-Showk, S., Ruonala, R. & Helariutta, Y. Crossing paths: cytokinin signalling and crosstalk. Development 140, 1373–1383 (2013).

    CAS  PubMed  Google Scholar 

  43. Lu, S. et al. Cloning and functional characterization of a beta-pinene synthase from Artemisia annua that shows a circadian pattern of expression. Plant Physiol 130, 477–486 (2002).

    CAS  PubMed  PubMed Central  Google Scholar 

  44. Schnurr, J. A. & Guerra, D. J. The CaMV-35S promoter is sensitive to shortened photoperiod in transgenic tobacco. Plant Cell Rep 19, 279–282 (2000).

    CAS  PubMed  Google Scholar 

  45. Murashige, T. & Skoog, F. A revised medium for rapid growth and bio assays with tobacco tissue cultures. Physiol Plantarum 15, 473–497 (1962).

    CAS  Google Scholar 

  46. De Vos, R. C. H. et al. Untargeted large-scale plant metabolomics using liquid chromatography coupled to mass spectrometry. Nat Protoc 2, 778–791 (2007).

    CAS  PubMed  Google Scholar 

  47. Grömping, U. DoE.base: Full factorials, orthogonal arrays and base utilities for DoE packages. Package version 0.27. Available: Accessed 18 Aug 2015.

  48. Grömping, U. & Xu, H. Q. Generalized resolution for orthogonal arrays. Ann Statist 42, 918–939 (2014).

    MathSciNet  MATH  Google Scholar 

  49. Smilde, A. K., Kiers, H. A. L., Bijlsma, S., Rubingh, C. M. & van Erk, M. J. Matrix correlations for high-dimensional data: the modified RV-coefficient. Bioinformatics 25, 401–405 (2009).

    CAS  PubMed  Google Scholar 

  50. Bylesjö, M., Rantalainen, M., Nicholson, J. K., Holmes, E. & Trygg, J. K-OPLS package: kernel-based orthogonal projections to latent structures for prediction and interpretation in feature space. BMC Bioinformatics 9, 106 (2008).

    PubMed  PubMed Central  Google Scholar 

Download references


This research was funded by the European Union Seventh Framework Programme SmartCell (Rational design of plant systems for sustainable generation of value-added industrial products, Grant agreement No. 222716). We acknowledge technical assistance by C. Schmitz and editorial assistance by R. M. Twyman.

Author information




N.V. and S.S. conceived this project; N.V. and U.G. designed the experiments; N.V. and G.L. performed the experiments; J.B., N.V., U.G. and S.R. analysed data; N.V., J.B., U.G. and G.L. wrote the paper with input from S.S., S.G., S.R. and R.F. All the authors discussed the results and commented on the manuscript.

Ethics declarations

Competing interests

The authors declare no competing financial interests.

Electronic supplementary material

Rights and permissions

This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Vasilev, N., Boccard, J., Lang, G. et al. Structured plant metabolomics for the simultaneous exploration of multiple factors. Sci Rep 6, 37390 (2016).

Download citation

Further reading


By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.


Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing