Systematic Structural Characterization of Metabolites in Arabidopsis via Candidate Substrate-Product Pair Networks

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/261255659

Systematic Structural Characterization of Metabolites in Arabidopsis via Candidate Substrate-Product Pair Networks ARTICLE in THE PLANT CELL · MARCH 2014 Impact Factor: 9.34 · DOI: 10.1105/tpc.113.122242 · Source: PubMed

CITATIONS

READS

15

83

9 AUTHORS, INCLUDING: Kris Morreel

Yves Van de Peer

Vlaams Instituut voor Biotechnologie

Ghent University

50 PUBLICATIONS 2,886 CITATIONS

394 PUBLICATIONS 33,374 CITATIONS

SEE PROFILE

SEE PROFILE

Ruben Vanholme

John Ralph

Ghent University

University of Wisconsin–Madison

32 PUBLICATIONS 1,304 CITATIONS

340 PUBLICATIONS 17,775 CITATIONS

SEE PROFILE

SEE PROFILE

Available from: John Ralph Retrieved on: 04 February 2016

The Plant Cell, Vol. 26: 929–945, March 2014, www.plantcell.org ã 2014 American Society of Plant Biologists. All rights reserved.

LARGE-SCALE BIOLOGY ARTICLE

Systematic Structural Characterization of Metabolites in Arabidopsis via Candidate Substrate-Product Pair Networks

CW

Kris Morreel,a,b,1 Yvan Saeys,a,b,c Oana Dima,a,b Fachuang Lu,d Yves Van de Peer,a,b,e Ruben Vanholme,a,b John Ralph,d Bartel Vanholme,a,b and Wout Boerjana,b a Department

of Plant Systems Biology, VIB, 9052 Ghent, Belgium of Plant Biotechnology and Bioinformatics, Ghent University, 9052 Ghent, Belgium, c Department for Inflammation Research Center, VIB, 9052 Ghent, Belgium d Department of Biochemistry and the Department of Energy Great Lakes Bioenergy Research Center, Wisconsin Energy Institute, University of Wisconsin, Madison, Wisconsin 53726 e Genomics Research Institute, University of Pretoria, Pretoria 0028, South Africa b Department

Plant metabolomics is increasingly used for pathway discovery and to elucidate gene function. However, the main bottleneck is the identification of the detected compounds. This is more pronounced for secondary metabolites as many of their pathways are still underexplored. Here, an algorithm is presented in which liquid chromatography–mass spectrometry profiles are searched for pairs of peaks that have mass and retention time differences corresponding with those of substrates and products from well-known enzymatic reactions. Concatenating the latter peak pairs, called candidate substrate-product pairs (CSPP), into a network displays tentative (bio)synthetic routes. Starting from known peaks, propagating the network along these routes allows the characterization of adjacent peaks leading to their structure prediction. As a proof-of-principle, this high-throughput cheminformatics procedure was applied to the Arabidopsis thaliana leaf metabolome where it allowed the characterization of the structures of 60% of the profiled compounds. Moreover, based on searches in the Chemical Abstract Service database, the algorithm led to the characterization of 61 compounds that had never been described in plants before. The CSPP-based annotation was confirmed by independent MSn experiments. In addition to being high throughput, this method allows the annotation of low-abundance compounds that are otherwise not amenable to isolation and purification. This method will greatly advance the value of metabolomics in systems biology. INTRODUCTION Metabolomics is increasingly used as a powerful systems biology tool. The identification of the many metabolites in biological samples, however, remains the main bottleneck in the field. Since 2000, methods have been developed to profile as many metabolites as possible from living tissues (Oliver et al., 1998; Nicholson et al., 1999; Tweeddale et al., 1999). The ongoing attempts to cover the whole metabolome have led to the optimization of separation methods based on gas chromatography– mass spectrometry (Fiehn et al., 2000), liquid chromatography– mass spectrometry (LC-MS) (Tolstikov and Fiehn, 2002; von Roepenack-Lahaye et al., 2004), capillary electrophoresisMS (Soga et al., 2003; Sato et al., 2004), and NMR spectroscopy (Nicholson et al., 1999). Whichever separation technology is used, only a minority of the profiled metabolites can be

1 Address

correspondence to [email protected]. The authors responsible for distribution of materials integral to the findings presented in this article in accordance with the policy described in the Instructions for Authors (www.plantcell.org) are: Kris Morreel ([email protected]) and Wout Boerjan ([email protected]). C Some figures in this article are displayed in color online but in black and white in the print edition. W Online version contains Web-only data. www.plantcell.org/cgi/doi/10.1105/tpc.113.122242

identified (Fernie, 2007), limiting the information that is gained in systems biology experiments. Compounds that remain unknown can be purified for structural elucidation with NMR, but the purification step is tedious, not always successful, and not a reasonable option for low-abundance peaks. The limited identification approaches are especially cumbersome for secondary metabolites that are relatively unknown and outnumber the primary metabolites. This necessitates the development of new methods to characterize the structures of as many compounds as possible that, as a consequence, will yield extra information on the various biochemical pathways operating in the considered tissue. Currently, high-throughput structural annotation of compounds is based on the availability of databases containing chemical formulae and/or mass spectral fragmentation data. When an accurate mass can be obtained, the chemical formula can be computed and databases screened for candidate molecules (Aharoni et al., 2002; Kind and Fiehn, 2006). This approach can lead to tens or hundreds of candidate molecules, but does not guarantee that any of these corresponds with the actual structure. Complementary, mass spectral fragmentation data can be consulted. For this purpose, metabolomics-based mass spectral libraries, such as the Golm Metabolome Database (Kopka et al., 2005), MassBank (Horai et al., 2010), or METLIN (Smith et al., 2005), have been constructed. Nonetheless, the donation of MS fragmentation spectra occurs at a low pace; hence, these libraries currently represent only a few

930

The Plant Cell

thousand compounds, whereas the number of metabolites in, for example, the plant kingdom is estimated to be 200,000 (Dixon and Strack, 2003; Fernie et al., 2004). Alternatively, software is being developed to improve the elucidation of MS fragmentation data (Neumann and Böcker, 2010). These libraries and software packages are promising for structure elucidation and indeed have led to the structural elucidation of 167 metabolites via reversed phase LC-MS analyses of nine Arabidopsis thaliana tissues harvested at multiple developmental stages (Matsuda et al., 2010), for example. Furthermore, based on the similarity of their MS fragmentation spectra, these authors also constructed a network containing 467 metabolites, including 95 structurally assigned compounds. The obtained clusters represent different classes of secondary metabolites, underscoring the assertion that mutually comparing MS fragmentation spectra of peaks offers a promising avenue for high-throughput structural elucidation. Reversed phase LC-MS profiles of plant extracts are rich in diverse classes of secondary metabolites. Because most of the profiled compounds from each of these classes are expected to show mass and retention time differences corresponding with those between substrates and products of well-known enzymatic reactions, we hypothesized that it should be possible to annotate pairs of peaks (often referred to as m/z features in metabolomics literature) that represent candidate substrate-product pairs (CSPPs) for a particular enzymatic conversion. Based on this fundamental idea, we developed an algorithm to search all possible CSPPs, based on a given list of (bio)chemical conversions. Assembling these CSPPs into a network permitted the proposal of structures for unknown peaks whenever they were connected to peaks with known structures. As a proof of concept, we applied this algorithm to the data obtained from reversed phase LC–negative electrospray ionization–MS profiling of the rosette leaf extracts from biological replicates of Arabidopsis Columbia-0 plants. The CSPP approach lead to the structural annotation of 145 of the estimated 229 metabolites belonging to various classes, for example, glucosinolates, flavonoids, benzenoids, phenylpropanoids, (neo)lignans/oligolignols, indolics, and apocarotenoids. Remarkably, based on searches in the CAS database, 61 of these compounds, all of which were quite compellingly structurally elucidated, have not been described before in any plant species. RESULTS CSPP Network Method Overview To elaborate the concept of CSPPs, methanol extracts from rosette leaves of 19 biological replicates of Arabidopsis Col-0 plants were analyzed by ultrahigh performance LC–Fourier transform-ion cyclotron resonance (FT-ICR)–MS. Following chromatogram integration and alignment, 3060 peaks, characterized by a retention time and an accurate m/z value, and corresponding to ;229 profiled compounds (see Methods; Supplemental Figure 1), were obtained. Because these peaks are biochemically related, peak pairs with a mass and retention time difference corresponding exactly to the expected mass and polarity shift from well-known enzymatic conversions in secondary metabolism are expected to be present. To test this assumption, we first compiled an arbitrary list of (bio)chemical conversions, of which some are expected to occur frequently in metabolism (the “true” conversions), whereas

Table 1. (Bio)Chemical Conversions for CSPP Network Generation Nr

Short

Con

m/z Dif

Elua

#CSPP

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38

Box Qui Shi Tar Cul Mal Rhab Col Cat Red Van Syr Phb Caf Dql Hql Cou Sil Iso Val Fer Pen Pcl Pbl Cal Syl Sin Qul Sun Hyd Gun Gly Oxy Ace Hex Mth Met Mox

b-Oxidation Quinate Shikimate Tartarate Coumaryl alcohol Malate Deoxyhexose Coniferyl alcohol Catechol Reduction Vanillate Syringate Hydroxybenzoate Caffeate Dimethoxyquinol Hydroxyquinol Coumarate Sinapyl alcohol Isoprenylation Vanillyl alcohol Ferulate Pentose Protocatechus alcoholc Hydroxybenzyl alcohol Caffeyl alcohol Syringyl alcohol Sinapate Quinol Syringyl Hydration Guaiacyl Glycerol Oxygenation Acetylation Hexose Malate_hexosed Methylation Methoxylatione

26.016 174.053 156.042 132.006 116.063 116.011 146.058 162.068 136.016 2.016 150.032 180.042 120.021 162.032 152.047 108.021 146.037 192.079 68.063 136.052 176.047 132.042 122.037 106.042 148.052 166.063 206.058 92.026 226.084 18.011 196.074 74.037 15.995 42.011 162.053 46.042 14.016 30.011

1 1 1 1 2 1 1 2 2 1 2 2 2 2 2 2 2 2 2 2 2 1 2 2 2 2 2 2 2 1 2 3 1 3 1 3 2 3

97 102 103 108 142 144 144 150 151 152 154 156 160 163 165 169 169 171 173 173 174 182 183 185 194 213 217 225 245 254 290 292 318 328 341 346 493 532

P.C.

y

y

y

y y y y y y y y y y

Nr, number; Short, shorthand naming; Con, conversion; m/z Dif, m/z difference; Elu, elution behavior of product peak versus substrate peak; #CSPP, number of CSPPs obtained for the conversion; P.C., prominent conversion (based on peak pair generation); y, yes. a Elution behavior: 1, product elutes earlier; 2, product elutes later; 3, not known. b Shorthand name is based on rhamnose. When the shorthand name is underlined or italics, > or #225 CSPPs were obtained for the conversion, respectively. Conversions written in italics are expected to occur to a lesser extent in Arabidopsis secondary metabolism. Conversions written in bold do not represent a true (bio)chemical conversion but are associated with a structural moiety that is often observed among the profiled metabolites. c This conversion can also be the addition of a methoxyquinol. d Hydroxycinnamoylmalate and hydroxycinnamoylhexose can be transesterified. e Methoxylation is often observed in phenylpropanoid metabolism, yet occurs by a separate oxygenation and methylation enzymatic reaction.

Candidate Substrate-Product Pair Networks

others occur rarely or not at all (the “false” conversions, see below; Table 1). Next, to obtain the CSPPs for each conversion, an algorithm was developed that used the following procedure: For each “substrate” peak, the list of 3060 peaks was queried to find “product” peaks for which the m/z value was equal to the m/z value of the “substrate” peak incremented by a mass equal to the mass change expected from the conversion. If such a peak pair was found, a CSPP was declared when the retention time of the “product” peak was smaller (conversions for which the product is more polar than the substrate) or larger (conversions for which the product is more apolar than the substrate) than that of the “substrate” peak (Figure 1; Supplemental Figure 2; see Methods for a detailed explanation of the CSPP algorithm). CSPPs were then concatenated into a network in which nodes and edges represent peaks and CSPP conversions. Statistical analysis of this CSPP network provided insight into its inherent metabolic network properties (see below; Figure 2). Subsequently, the validity of the “true” (bio)chemical conversions was assessed by comparison of the number of CSPPs presented in Table 1 with the number of pairs of chromatogram peaks obtained when the mass difference was systematically varied without taking the retention time into account (Table 1, Figure 3; Supplemental Figure 3). The latter method allows the relevant conversions for inclusion into the CSPP algorithm to be deduced from the data at hand. Finally, CSPP-based structural elucidation was performed via network propagation starting from known nodes (Figure 4). Whenever possible, structural elucidation of adjacent unknown nodes was aided by considering the Pearson correlation between the levels of both peaks across biological replicates and by MS fragmentation (MS2) spectral similarity matching (Figure 1). Information on all structurally characterized peaks is summarized in Supplemental Data

931

Set 1 and Supplemental Figure 4. An overview of the profiled pathways is shown in Figure 5. Proposed structures were verified via MSn ion trees, i.e., by a nested fragmentation approach in which MS2 first product ions are further fragmented into MS3 second product ions.

CSPP Network Statistical Analysis The likelihood with which a CSPP reflects a metabolic conversion can be derived from the number of CSPPs that are obtained with a random selection of both well-known “true” (bio)chemical and “false” erroneous conversions (Table 1). These conversions had to be chosen a priori to allow a valid statistical analysis and to determine the number of false positive CSPPs (see below). Among the “true” biochemical conversions were enzymatic reactions that prevail in secondary metabolism such as methylation and oxygenation, and, because many phenolics were expected (D’Auria and Gershenzon, 2005), phenolic derivatizations, for example, the condensation of organic acids or saccharides with phenolics and chemical conversions arising from radical coupling of monolignols leading to (neo)lignans. For the “false” conversions, the masses of various aromatic moieties that are characteristic for the structures of flavonoids and (neo)lignans, e.g., quinol, but that do not arise directly from a chemical or enzymatic addition or condensation reaction, were chosen. Additionally, “pseudo” erroneous conversions were considered as well, i.e., enzymatic reactions that were not expected to occur frequently in Arabidopsis secondary metabolism, such as isoprenylation. In total, this (bio)chemical conversion list contained 38 reactions and yielded 7958 CSPPs (Table 1; see Methods).

Figure 1. Overview of the CSPP Algorithm. A CSPP is defined based on a particular mass difference and retention time order, yet the CSPP algorithm computes the Pearson product-moment correlation coefficient and the MS2 spectral similarity as well, which can be used as additional filters. [See online article for color version of this figure.]

932

The Plant Cell

Figure 2. CSPP Distribution Properties. (A) Histogram of CSPP number. The curve represents a Gaussian kernel density function (weight = 40). (B) CSPP number and average Pearson correlation coefficient versus the conversion type. (Bio)chemical conversions were ordered with increasing CSPP number (see Table 1, Nr). The dashed lines indicate the knot from which the CSPP number per conversion increases more steeply and all conversions represent well-known enzymatic reactions. A Pearson correlation coefficient was computed for each CSPP between its “substrate” and “product” levels (based on the MS ion current signal) across biological replicates. The correlation coefficients obtained for all CSPPs belonging to the same conversion were then averaged. The positive association between these average Pearson correlations and the number of CSPPs was also evaluated via a Pearson correlation (see r2 and P values inserted in the plot). (C) and (D) Node connectivity distribution of CSPP networks. Log-log plot of network node connectivity distribution involving all reactions (C) and loglog plot of network node connectivity distribution involving only the “high CSPP number” conversions (D). The accuracy of the fit to a power law distribution for the log-log plots in (C) and (D) was computed via the plfit (D statistic) and power.law.fit (-2logLik) functions in R (see Methods). The significance of the better accuracy obtained for the log-log plot in D was tested via bootstrapping (inset; see Methods). -2logLik, 22 times the logarithm of the likelihood; Conf. Lim., one-sided confidence limit; D, Kolmogorov-Smirnov goodness-of-fit statistic; k, node connectivity or the number of edges a node possesses. a and xmin are estimates for the power law function parameters.

CSPP Number Distribution Depending on the type of conversion, between 97 and 532 CSPPs were obtained. By dividing this range into 19 classes (x axis), a histogram of the number of conversions (y axis) in each class was made (Figure 2A). Clearly, the histogram was not normally distributed but showed a bi- or multimodal distribution, demonstrating that the conversions could be partitioned into at least two groups. The largest group was represented by the mode at ;150 to 175 CSPPs and, by considering the underlying normal distribution, the number of CSPPs for the 29 conversions

comprising this normal distribution ranged between ;50 and ;250. Most of these conversions (Table 1, italics) were a priori not expected to occur frequently in secondary metabolism in Arabidopsis leaves (e.g., isoprenylation or quinol addition) (D’Auria and Gershenzon, 2005). Except for glycerol addition (Gly, Table 1), the eight conversions with more than 250 CSPPs are well known in Arabidopsis secondary metabolism, such as methylation (Met, Table 1) or hexosylation (Hex, Table 1). Thus, these data indicate that mass differences for which high numbers of CSPPs are found are more likely to be associated with true biochemical conversions.

Candidate Substrate-Product Pair Networks

933

Figure 3. Peak Pair Generation. The number of chromatogram peak pairs for a particular mass difference, up to precisely three decimals, was computed. The mass differences varied between 0.001 and 250.000 D; thus, 250,000 mass differences were considered. (A) Manhattan plot showing the number of peak pairs (y axis) versus the mass difference (x axis). (B) Manhattan plot with mass differences ranging from 0 to 20 D. (C) and (D) Expansion of Manhattan plot showing the mass difference region for reduction (C) and for hexosylation (D). (E) and (F) Distribution of the number of peak pairs versus retention time difference between both peaks of the peak pair. Plots are given for mass differences corresponding to reductions (E) and hexosylations (F). The curved line in (A) and straight lines in (C) and (D) represent the minimum number of peak pairs necessary to consider the mass difference relevant for inclusion as a CSPP conversion type (see Supplemental Methods for further explanation). [See online article for color version of this figure.]

Correlation of CSPP Candidate Substrate and Product Peak Abundances The partitioning of the conversion types into two groups arises from Figure 2B as well; when ordering all 38 conversion types presented in Table 1 by their CSPP number-based rank (x axis, Nr in Table 1), and plotting this versus the number of CSPPs (y axis), a segmented linear function was obtained with a knot at 200 to 225 CSPPs. Assuming that levels of secondary metabolic pathway intermediates might be mutually more highly correlated than those with the rest of metabolism (see Discussion and Supplemental Methods), higher correlations are expected for CSPPs representing true (bio)chemical conversions than for those associated with false conversions. Therefore, across biological replicates, the Pearson product-moment correlation

coefficients betweenthe MS ion current–based abundances of both “substrate” and “product” peaks for each CSPP were computed. Next, the average was computed of the correlations obtained for all CSPPs within each conversion type. These average correlation coefficients are displayed in Figure 2B. As can be observed, the average correlation coefficient increases from left to right in Figure 2B, i.e., conversions with a higher number of CSPPs represent more highly correlated CSPPs as well. In fact, a significant association (Pearson r2 = 0.57, P = 1.7 3 1024) existed between the average Pearson correlation coefficient for each conversion type and its number of CSPPs. The dichotomy in both plots (A and B) shown in Figure 2 supports the notion that the group of conversions with the higher number of CSPPs (“high CSPP” group) is enriched in CSPPs that have a true biochemical background than the

934

The Plant Cell

Figure 4. CSPP Analysis of Aliphatic Glucosinolates. (A) Pathway overview. Full arrows represent enzymatic reactions, whereas dashed arrows represent multiple enzymatic conversions. Dotted arrows indicate the various compound classes. (B) Nodes and edges represent chromatogram peaks and (bio)chemical conversions (see Table 1 for conversion types). Node labels are based on the XCMS integration and alignment algorithm. Whenever the similarity between the MS2 spectra of candidate substrate and product exceeds 0.8, the edge

Candidate Substrate-Product Pair Networks

group of conversions with lower CSPP numbers (“low CSPP” group). Additionally, these data supported the use of correlations to filter CSPPs representing true (bio)chemical conversions from the total list. CSPP Network Topology Subsequently, metabolic networks were made by concatenating “substrate”-“product” peak pairs, one based on the total number of CSPPs presented in Table 1, and one based on the “high CSPP” group only. In a metabolic network in which the edges and nodes represent enzymatic reactions and metabolites, the number of connections per node, i.e., the node connectivity, follows a scale-free (power law) rather than a random distribution (Jeong et al., 2000). This implies that few nodes are highly connected whereas the majority are scarcely connected. For the CSPP networks, the connections are based partially on CSPPs having a biochemical origin and partially on CSPPs that arise when two peaks have, purely by chance, a mass and retention time difference corresponding with that of a biochemical conversion. The latter can be regarded as “random” CSPPs. Therefore, the node connectivity of CSPP networks is expected to be a mixture distribution, i.e., a composite of a random and a scalefree distribution based on the presence of “random” and “biochemical” CSPPs. The higher the fraction of “biochemical” CSPPs, the more the underlying scale-free distribution will prevail in the mixture distribution profile. This should be the case for the network of the “high CSPP” group (high CSPP network) compared with that of the network containing all conversion types (full network). In agreement, based on the likelihood ratio (-2logLik) and Kolmogorov-Smirnov goodness-of-fit (D value) tests, the topology of the high CSPP network was more accurately modeled by a scale-free distribution (see log-log plots in Figures 2C and 2D) than that of the full network. Based on bootstrapping results, a one-sided 95% confidence interval for the D value was constructed in the case of the high CSPP network. This showed that the probability of obtaining a D value as low as 0.0467 (Figure 2D) when drawing a subnetwork from the full network was 0.8) to restrict the number of edges in the network. For

Candidate Substrate-Product Pair Networks

example, 4,680,270 edges would have been computed from our set of 3060 peaks. Because the CSPP networks are based on a priori defined biochemical conversions, they contain much fewer connections. For example, in this study, 7650 edges emerged from 38 considered conversions. Therefore, a much less stringent correlation threshold can be used. The most appropriate correlation threshold can be estimated based on the data obtained from the MS2 spectral similarity matching (see Methods). Whenever the “substrate” and “product” in a CSPP have similar MS2 spectra, they usually have similar molecular structures (Rasche et al., 2012; Rojas-Cherto et al., 2012), conferring a high probability that the CSPP represents a “true” metabolic reaction. Therefore, the MS2 spectral similarity is a measure for judging the likelihood that a CSPP for which a moderate to high correlation coefficient was obtained represents a “true” metabolic reaction. By considering only those CSPPs that were obtained with the “high CSPP” type conversions and for which a MS2 spectral similarity was computed (151 CSPPs in total), the number of CSPPs in which the MS2 spectra of “substrate” and “product” are similar or nonsimilar can be counted. In Figure 6 (left plot), the probability of a false positive is plotted for various correlation coefficient thresholds (computations are illustrated for r2 > 0.6 and r2 > 0.7 in Figure 6). Above a correlation coefficient threshold of 0.6, 85 CSPPs belonged to the “MS2 spectrally similar group,” whereas only 14 CSPPs belonged to the “MS2 spectrally nonsimilar group.” From this perspective, by selecting only CSPPs from the “high CSPP” group conversions and, furthermore, only those that are associated with moderate to high correlation coefficients, the chance

939

on a false positive drops to 14% [=14/(85+14)*100]. The association of higher correlation coefficients with higher MS2 spectral similarities is further strengthened by regarding the low-correlated CSPPs (r2 < 0.6) in which the absence of a MS2 spectral similarity prevailed (23 and 29 CSPPs belonged to the “MS2 spectrally similar group” and the “MS2 spectrally nonsimilar group,” respectively; Figure 6). This is reflected in the odds ratio of 7.66, which indicates that the chance that a highly correlated CSPP will belong to the “MS2 spectrally similar group” rather than to the “MS2 spectrally nonsimilar group,” is more than 7 times higher than that for a low-correlated CSPP (Figure 6, right plot). It should be stressed that, in the discussion above, MS2 spectral similarities are used to assess the validity of including correlation coefficients as a filter to select CSPPs that are more likely associated with “true” biochemical conversions. Logically, adding the MS2 spectral similarity itself as a second filter will, in combination with the correlation coefficient, diminishes the chance of a false positive even more (Figure 1) (Rasche et al., 2012; Rojas-Cherto et al., 2012). However, calculating the chance of a false positive using the CSPP algorithm with the inclusion of both filters is impossible as it would need the unambiguous structural identification of all characterized molecules. CSPP Networks Do Not Show a Small World Behavior A more in-depth analysis of the number of false positive CSPPs obtained with various correlation thresholds provides information on the magnitude of the correlation coefficients

Figure 6. Effect of the Correlation Coefficient Threshold as a Filter for CSPP Selection. CSPPs obtained from the “high CSPP” group conversions for which a MS2 spectral similarity was computed were selected (151 CSPPs). CSPPs for which the “global common” (see Methods) was at least 2 or having a “global similarity” above 0.8 (see Methods) were classified as “MS2 spectrally similar.” Other CSPPs were “MS2 spectrally nonsimilar.” They were further classified as “high correlation” CSPPs whenever their associated correlation coefficient was higher than the threshold; otherwise, they were annotated as “low-correlation” CSPPs. This cross-tabulation was performed for different correlation coefficient thresholds (cor) and used for computing the chance on a false positive for the “high correlation” CSPPs (false pos), the chance on a true negative for the “low-correlation” CSPPs (true neg), and the odds ratio (see Discussion for explanation).

940

The Plant Cell

associated with secondary biochemical conversions. Lowering the correlation threshold from 0.9 to 0.7 (Figure 6, left plot) increases the number of false positive CSPPs and thus lowers the odds ratio (Figure 6, right plot). Remarkably, when the correlation threshold is set at 0.6 rather than 0.7, a considerable increase in the odds ratio is observed. This odds ratio jump reflects the improved classification of low-correlated (r2 < 0.6) CSPPs as true negative CSPPs (Figure 6, middle plot). Obviously, there is still a large number of moderately correlated (0.6 < r2 < 0.7) CSPPs that belong to the “MS2 spectrally similar” group and thus have a high probability of representing true biochemical reactions. Evaluating correlation thresholds from 0.6 to 0.4 did not substantially change the fraction of true negative CSPPs among the CSPPs with a correlation below the threshold, but the number of false-positive CSPPs increased considerably, leading to a decrease of the odds ratio. This data suggests that most biochemically valuable CSPPs are moderately to highly correlated. A more in-depth classification of the latter CSPPs based on the extent that they reflected a biochemical conversion is given as Supplemental Methods. Correlations among metabolite abundances arise when environmental changes lead to metabolite abundance fluctuations that in their turn affect the complex regulation of metabolism. However, initial metabolome experiments (Roessner et al., 2001) as well as simulation studies (Steuer et al., 2003; Müller-Linow et al., 2007) have shown that correlations do not necessarily reflect the pathway architecture. More specifically, profiling studies of mainly primary metabolites (Roessner et al., 2001) have shown that most metabolite pairs have low correlation coefficients and only a few metabolite pairs are highly correlated. Therefore, the moderate to high correlation coefficients observed in the CSPP network could be associated with the biochemical nature of the profiled compounds that were all secondary metabolites. In the early plant metabolomics literature, highly positive correlations were observed between the abundances of metabolites that are in chemical equilibrium (Roessner et al., 2001). However, such an explanation does not hold for the highly correlated CSPPs, as all “high CSPP” group conversions represent irreversible reactions. Alternatively, highly correlated CSPPs could arise when the abundances of the candidate substrate and product are controlled by the same enzymatic reaction(s) (Camacho et al., 2005). In the latter case, if control is shared by a few enzymatic reactions, the coresponse of the candidate substrate and product abundances to an altered reaction rate should lie in the same direction for each of the controlling metabolic steps (Supplemental Figure 9, left plot) (Camacho et al., 2005). If the direction of the coresponse to at least one of the controlling reactions differs from those of the remaining controlling reactions, a moderate instead of a high correlation might still be observed (Supplemental Figure 9, right plot) (Camacho et al., 2005). Another difference between the CSPP networks in this study and primary metabolic networks is the absence of negative correlations. Negative correlations emerge when the levels of two metabolites are controlled by mass conservation (Camacho et al., 2005), for example, when two metabolites are part of a moietyconserved cycle or when they belong to different branches that compete for the same precursor. The predominance of moderate

to high positive correlations together with the absence of negative correlations in our CSPP networks suggests that branches, cycles, and amphibolic reactions are less frequent in secondary than in primary metabolic networks. This lack of interconnectivity in the CSPP networks compared with primary metabolic networks can also be retrieved from the network diameter, i.e., the average shortest path calculated across all pairs of compounds. A network diameter of 24 was obtained for the CSPP network on which the structural characterization was based, i.e., the CSPP network that contained only the “high CSPP” group conversions together with the conversions having a high number of peak pairs. Furthermore, the diameter of the latter CSPP network did not change much when allowing only moderate to highly correlated edges (r2 > 0.6, diameter = 23). These network diameters are far higher than those of (primary) metabolic networks (Jeong et al., 2000) that were between 3 and 4. This suggests that the small world character of primary metabolic networks, i.e., that any pair of metabolites can be linked via relatively short paths, cannot be extrapolated to secondary metabolic networks, although some caution is needed because many secondary pathway intermediates are not presented in the CSPP network. Nevertheless, the system biology information displayed by the CSPP network could not be retrieved from genome-based metabolic models as many enzymes operating in secondary metabolism are still unknown. The newly described CSPP algorithm opens up a major avenue for the structural elucidation of the many unknowns in metabolomic experiments. Via network propagation, the structures of unknowns can be deduced from the structures of known precursors and can subsequently be used to aid in the structural elucidation of other unknowns that are connected in the network. The limited structural knowledge of the differential peaks in comparative profiling studies prohibits a clear understanding of the living system, a restriction that is largely overcome by the proposed CSPP method. By annotating peaks that differ in mass in agreement with a certain enzymatic conversion and taking into account (1) their retention time order, (2) the correlation between their abundances across biological replicates, and (3) their MS2 spectral similarity, 60% of the compounds profiled by reversed phase LC-negative ionization MS of Arabidopsis rosette leaves could be structurally characterized. Moreover, the value of the method extends beyond the plant field and will also propel forward metabolomics in the human/ animal field where the metabolome is heavily influenced by the microbiome and the nutritional composition.

METHODS Growth Conditions and Extraction Arabidopsis thaliana Columbia-0 seeds were randomly sown in trays (19 biological replicates). Following vernalization at 4°C in the dark for 48 h, plants were transferred to short-day-conditioned growth room (8 h light, 22°C, 55% relative humidity, 120 PAR). After 2 months, one rosette leaf from each plant with a length of ;1 cm was randomly harvested and snapfrozen in liquid nitrogen. Following ball-milling with a Retsch mill (25 Hz) for 20 s, the homogenized plant material was extracted with 0.5 mL methanol. Of the supernatants, 0.4 mL was lyophilized and redissolved in

Candidate Substrate-Product Pair Networks

0.8 mL milliQ water/cyclohexane (1/1, v/v). From the water phase, 10 mL was used for phenolic profiling.

Metabolite Profiling Extracts were analyzed with an Accela ultrahigh performance LC system (Thermo Electron) consisting of an Accela autosampler coupled to an Accela pump and further hyphenated to a LTQ FT Ultra mass spectrometer (Thermo Electron) consisting of a linear ion trap mass spectrometer connected with an FT-ICR mass spectrometer. The separation was performed on a reversed phase Acquity UPLC HSS C18 column (150 mm 3 2.1 mm, 1.8 mm; Waters) with aqueous 0.1% acetic acid and acetonitrile/water (99/1, v/v, acidified with 0.1% acetic acid) as solvents A and B. At a flow of 300 mL/min and a column temperature of 60°C, the following gradient was applied: 0 min 1% B, 30 min 60% B, and 35 min 100% B. The autosampler temperature was 5°C. Analytes were negatively ionized with an electrospray ionization source using the following parameter values: spray voltage 3.5 kV, capillary temperature 300°C, sheath gas 40 (arb), and aux gas 20 (arb). Full FT-ICR-MS spectra between 120 and 1400 m/z were recorded (1.2 to 1.7 s/scan) at a resolution of 100,000. In parallel, four datadependent MSn spectra were recorded on the ion trap mass spectrometer using low resolution data obtained during the first 0.1-s period of the previous full FT-ICR-MS scan: A MS2 scan of the most abundant m/z ion of the full FT-ICR-MS scan, followed by two MS3 scans of the most abundant first product ions and a final MS4 scan of the most abundant second product ion obtained from the base peak in the MS2 spectrum. These MSn scans were obtained with 35% collision energy. Using RecalOffline vs. 2.0.2.0614 (Thermo Electron), the full FT-ICR-MS scans were sliced from each chromatogram raw file and subsequently converted to netCDF with Xcalibur version 2.0 SR2 (Thermo Electron). Integration and alignment was performed with the XCMS package (Smith et al., 2006) in R version 2.6.1 using the following functions: xcmsSet (fwhm = 6, max = 300, snthresh = 2, mzdiff = 0.01), group (bw = 10, max = 300, mzwid = 0.01), retcor (method= “loess,” span = 0.2, family = “symmetric,” plot type = “mdevden”). Following retention time correction, a second peak grouping was performed: group (bw = 8, max = 300, mzwid = 0.01). This process resulted in 3060 integrated peaks. It should be stressed that the number of peaks does not reflect the number of compounds. Each compound is represented by multiple peaks: Besides the base peak, peaks representing isotopes and adducts and peaks due to in-source fragmentation of the compound show up in the chromatogram as well. Chemical formulae of compounds of interest were obtained with the Qual Browser in Xcalibur version 2.0 SR2. Instead of the pseudo-molecular ions of the compounds, all peaks were used for the CSPP algorithm. This avoids that the peak grouping algorithm, to define the compounds, introduces flaws into the CSPP network. Flaws can occur because (1) peaks belonging to two highly correlated, coeluting compounds would be grouped together and (2) the selection of the pseudo-molecular ion (i.e., the deprotonated compound) within the peak group can be erroneous.

Peak Grouping Procedure Peaks associated with the same compound have the same retention time and their levels are highly correlated across chromatograms. The number of such peak groups can be regarded as an estimate for the number of profiled compounds; these were searched using a small range for the retention time window (varying from 1 to 6 s with the latter value representing half of the baseline peak width) and the Pearson correlation threshold (varying from 0.70 to 0.95) (Supplemental Figure 1). At a retention time window and a correlation threshold of 1 s and 0.8, an optimum was observed in the surface plot of Supplemental Figure 1D, which projects the co-optimization [(meanP/meanG)*P*G] of the number of peak groups (G = 229) and the number of peaks that could be assigned to

941

a peak group (P = 2400) onto the retention time window and the correlation threshold. The 660 unassigned peaks are low-abundance peaks for which the m/z value was often not accurately determinable. Because G was one order of magnitude smaller than P, a correction factor (meanP/ meanG) was added to the formula. Peak grouping was performed using an in-house-written script in R version 2.13.1. CSPP Network Generation A script was written in Perl to search for peaks that show mass and retention time differences corresponding with those of substrates and products of enzymatic conversions (Supplemental Figure 2). Central in the algorithm is the detection of all candidate product peaks (because multiple isomers might be present) that associate with a particular candidate substrate peak. In the article, candidate product and candidate substrate are annotated as “product” and “substrate.” For each considered conversion, e.g., hexosylation, the “node” list of 3060 peaks, in which each row represented a peak, was ordered with increasing m/z (Supplemental Figure 2). Starting from the “substrate” peak i at m/z = massi, the m/z value of the theoretical hexosylation product was computed by adding 162.053 D (m/z = massi + 162.053). The list was then searched for peaks with the corresponding m/z value. Whenever a peak j was found at m/z = massj (massj = massi + 162.053 6 error with error equal to the chosen m/z window, i.e., 0.008 in this study), its retention time was considered. Given that a reversed phase column is used, the “product” peak is expected to elute before or after the “substrate” peak when the “product” is more polar or apolar than the “substrate,” respectively. In the case of hexosylation, the “product” is expected to elute earlier than its aglycone “substrate.” Thus, when the retention time of peak j was shorter than that of the “substrate” peak i (tRj225 CSPPs; see Table 1 and Results, “real” trait contained 3439 values). From both generated “artificial” traits, values equal to zero were subsequently excluded leading to “artificial” traits containing 8805 and 7188 values for the full and high CSPP networks. Because the latter network contained fewer CSPPs, the corresponding “artificial” trait contained less information and parameter estimation is expected to be less precise as compared with the full network. Modeling of a power-law distribution was performed using the plfit function for which the R source code was downloaded (http:// www.santafe.edu/~aaronc/powerlaws/) (Clauset et al., 2009). In addition, using the xmin value estimated by the plfit procedure, the power.law.fit function in the igraph package was applied. To determine whether the “high CSPP” network fitted a power-law distribution better than the full network, a one-sided 95% confidence interval for the Kolmogorov-Smirnov goodness-of-fit (D) statistic had to be constructed. This confidence interval was obtained via bootstrapping using the sample and plfit functions in R. In bootstrapping, a random sample is drawn from the “artificial” full network trait (8805 values) for which the D statistic is computed. This procedure was repeated 999 times after which the distribution of the D statistic can be plotted and the confidence interval determined. The ratio (=0.43) of CSPPs present in the high CSPP network (=3439 “real” values) versus those present in the full network (=7958 “real” values) determined the size of each bootstrap sample (=8805 * 0.43 = 3786). Samples were generated without replacement. This bootstrap strategy was pursued based on the lower information present in the high CSPP network and because the high CSPP network comprises a subset of the full network.

Peak Pair Generation Mass differences that predominated among all possible peak pairs as well as the corresponding retention time difference distributions were determined and the corresponding plots (Manhattan plots and histograms; Figure 3) constructed with in-house-written R scripts using R version 2.13.1. Mass differences between 0.000 and 250.000 D with a precision of 0.001 D were considered. A threshold on the minimum number of peaks pairs had to be computed to determine which mass differences correspond to “true” (bio)chemical conversions occurring in metabolism. This threshold decision should take three properties of the Manhattan plot into account. (1) The number of peak pairs decreased with incrementing mass differences. (2) Regions are present that are enriched in mass differences having a high number of peak pairs. Most of these mass differences could only be reasoned to occur from the intervention of at least three different conversions. Properties (1) and (2) indicate that the threshold should be based on the local mass difference region. (3) In each nominal mass difference interval, the maximum number of peaks was observed at values

Candidate Substrate-Product Pair Networks

close to unit mass differences. This is the consequence of the isotopic masses of oxygen (15.995 D), carbon (12.000 D), hydrogen (1.007 D), and nitrogen (14.003 D) that are all close to unit mass values. As a consequence, most of the computed number of peak pairs corresponded to erroneous mass differences and should not be taken into account for threshold computation. Thus, for each mass difference, the final threshold was chosen based on the maximum number of peak pairs (max) observed in a one-unit mass difference interval. A lognormal distribution was observed for all so-obtained max values. For the reasoning above, the selection of the more prominent available mass differences corresponding to “true” (bio)chemical conversions was based on the following procedure: For each mass difference, the threshold was based on the logarithm of the maximum number of peak pairs (logmax) that was observed in the range of mass differences up to 1 D beyond the considered mass difference (e.g., when the considered mass difference was 14.000 D, the threshold was based on maximum number of peak pairs that was observed in the mass difference region between 14.000 and 15.000 D). The mean and SD of the logmax value across a mass difference range of 10 D was determined, a 95% confidence interval was computed at each mass difference, and the confidence limit of this logarithmic trait was back-transformed to obtain the threshold. In Figure 3, the smoothed threshold curve was obtained using the smooth.spline function (spar = 0.85) in R.

Synthesis of G(8–O–4)FA Glu Compound S1 (Supplemental Figure 7) was obtained by alkaline hydrolysis from its parent methyl ester that was synthesized according to the method of Helm and Ralph (1992). Acetylation of S1 with acetic anhydride and pyridine was followed by an amidation with Glu hydrochloride in the presence of N,N’-dicyclohexylcarbodiimide and 4-dimethylaminopyridine produced compound S2. Hydrolysis of S2 with 1 M sodium hydroxide in 50% ethanol resulted in the target product, G(8–O–5)FA glutamate. The product contained two isomers, the chemical shifts for Glu moiety were the same and those for the rest of the molecular structure were different although they could not be assigned clearly for each isomer. 1H NMR (acetone-d6), dH, 2.02 to 2.25 (2H, m, G4), 2.47 (2H, m, G3), 3.46 to 3.62 (2H, m, A9), 3.55 to 3.72 (2H, m, A9), 4.35/4.42 (1H, m, A8), 4.67 (1H, m, G3), 4.90/4.91 (1H, d, J = 5.96 Hz, A7), 6.65/6.68 (1H, d, J =15.69 Hz, B8), 6.75/6.77 (1H, d, J = 8.12 Hz, A5), 6.85/6.89 (1H, dd, J = 8.12, 1.80 Hz, A6), 7.03/7.08 (1H, br-s, B6), 7.01/7.16 (1H, d, J = 1.80 Hz, B5), 7.09/7.11 (1H, d, J =1.0 Hz, A2), 7.22/7.17 (1H, d, J = 1.0 Hz, B2), 7.48/7.51 (1H, d, J =15.69 Hz, B7); dc, 27.88 (G4), 30.50 (G3), 52.19 (G2), 54.33 (A8), 56.1056.24 (OMe), 61.86/62.06 (A9), 73.64/73.72 (A7), 85.72/87.35 (A8), 111.28/ 111.42 (A2), 111.64/111.75 (B2), 115.03/115.20 (A5), 117.94/118.33 (B5), 119.97/120.07 (B8), 120.42/120.50 (A6), 129.81/130.00 (B1), 133.68/ 134.05 (A1), 141.51/141.25 (B7), 146.62/146.77 (A4), 147.96/147.08 (A3), 151.14/151.71 (B4), 151.45/151.54 (B3), 150.84 (B4), 166.82 (B9), 173.44 (G1), 174.10 (G5). Supplemental Data The following materials are available in the online version of this article. Supplemental Figure 1. Peak Grouping Surface Plots. Supplemental Figure 2. CSPP Generation Algorithm. Supplemental Figure 3. Retention Time Difference Distributions. Supplemental Figure 4. Annotated Molecular Structures. Supplemental Figure 5. CSPP Subnetworks of Flavonoids and Phenylpropanoids. Supplemental Figure 6. CSPP Subnetwork of Highly Correlated CSPPs.

943

Supplemental Figure 7. Synthesis of G(8–O–4)FA Glu. Supplemental Figure 8. Gas Phase Fragmentation Pathways of Simple Phenolics and Phenylpropanoids, 59-O-b-D-Glucosyl Dihydroascorbigen, and Corchoionoside C Anions. Supplemental Figure 9. Effect of Shared Control on Correlation. Supplemental Table 1. MS2 Spectra of Simple Phenolics. Supplemental Table 2. MS 2 Spectra of Monolignol-Related Compounds. Supplemental Data Set 1. Structurally Annotated Chromatogram Peaks. Supplemental References. Supplemental Methods.

ACKNOWLEDGMENTS This work was supported by the Stanford University Global Climate and Energy Project (“Towards New Degradable Lignin Types,” “Efficient Biomass Conversion: Delineating the Best Lignin Monomer-Substitutes,” and “Lignin management: optimizing yield and composition in lignin-modified plants”), by the Multidisciplinary Research Project “Biotechnology for a sustainable economy” of Ghent University, by the European Community’s Seventh Framework Programme (FP7/2009) under grant agreement 251132 (SUNLIBB), and the “Bijzonder Onderzoeksfonds-ZwareApparatuur” of Ghent University for the FT-ICR-MS (Grant 174PZA05). Y.S. and R.V. are postdoctoral fellows of the Research Foundation-Flanders. We thank Sabine Montaut for sharing the high-resolution tandem mass spectrometry data recorded for 59-O-b-D-glucosyldihydroascorbigen and Annick Bleys for help in preparing the article. AUTHOR CONTRIBUTIONS K.M. designed research. K.M., Y.S., O.D., F.L., and R.V. performed research. K.M., Y.S., B.V., and Y.V.d.P. contributed new analytical tools. K.M., Y.S., O.D., F.L., J.R., and W.B. analyzed data. K.M., Y.S., J.R., and W.B. wrote the article.

Received December 20, 2013; revised February 20, 2014; accepted March 11, 2014; published March 31, 2014.

REFERENCES Aharoni, A., Ric de Vos, C.H., Verhoeven, H.A., Maliepaard, C.A., Kruppa, G., Bino, R., and Goodenowe, D.B. (2002). Nontargeted metabolome analysis by use of Fourier transform ion cyclotron mass spectrometry. OMICS 6: 217–234. Camacho, D., de la Fuente, A., and Mendes, P. (2005). The origin of correlations in metabolomics data. Metabolomics 1: 53–63. Cataldi, T.R.I., Lelario, F., Orlando, D., and Bufo, S.A. (2010). Collision-induced dissociation of the A + 2 isotope ion facilitates glucosinolates structure elucidation by electrospray ionizationtandem mass spectrometry with a linear quadrupole ion trap. Anal. Chem. 82: 5686–5696. Clauset, A., Shalizi, C.R., and Newman, M.E.J. (2009). Power-law distributions in empirical data. SIAM Rev. Soc. Ind. Appl. Math. 51: 661–703. D’Auria, J.C., and Gershenzon, J. (2005). The secondary metabolism of Arabidopsis thaliana: Growing like a weed. Curr. Opin. Plant Biol. 8: 308–316.

944

The Plant Cell

Dixon, R.A., and Strack, D. (2003). Phytochemistry meets genome analysis, and beyond. Phytochemistry 62: 815–816. Fabre, N., Poinsot, V., Debrauwer, L., Vigor, C., Tulliez, J., Fourasté, I., and Moulis, C. (2007). Characterisation of glucosinolates using electrospray ion trap and electrospray quadrupole time-of-flight mass spectrometry. Phytochem. Anal. 18: 306–319. Fabre, N., Rustan, I., de Hoffmann, E., and Quetin-Leclercq, J. (2001). Determination of flavone, flavonol, and flavanone aglycones by negative ion liquid chromatography electrospray ion trap mass spectrometry. J. Am. Soc. Mass Spectrom. 12: 707–715. Fernie, A.R. (2007). The future of metabolic phytochemistry: larger numbers of metabolites, higher resolution, greater understanding. Phytochemistry 68: 2861–2880. Fernie, A.R., Trethewey, R.N., Krotzky, A.J., and Willmitzer, L. (2004). Metabolite profiling: From diagnostics to systems biology. Nat. Rev. Mol. Cell Biol. 5: 763–769. Fiehn, O., Kopka, J., Dörmann, P., Altmann, T., Trethewey, R.N., and Willmitzer, L. (2000). Metabolite profiling for plant functional genomics. Nat. Biotechnol. 18: 1157–1161. Halkier, B.A., and Gershenzon, J. (2006). Biology and biochemistry of glucosinolates. Annu. Rev. Plant Biol. 57: 303–333. Helm, R.F., and Ralph, J. (1992). Lignin-hydroxycinnamyl model compounds related to forage cell wall structure. 1. Ether-linked structures. J. Agric. Food Chem. 40: 2167–2175. Horai, H., et al. (2010). MassBank: A public repository for sharing mass spectral data for life sciences. J. Mass Spectrom. 45: 703–714. Iijima, Y., et al. (2008). Metabolite annotations based on the integration of mass spectral information. Plant J. 54: 949–962. ! eková, S. (2003). Jandera, P., Halama, M., Novotná, K., and Bunc Characterization and comparison of HPLC columns for gradient elution. Chromatographia 57: S153–S161. Jeong, H., Tombor, B., Albert, R., Oltvai, Z.N., and Barabási, A.-L. (2000). The large-scale organization of metabolic networks. Nature 407: 651–654. Justesen, U. (2000). Negative atmospheric pressure chemical ionisation low-energy collision activation mass spectrometry for the characterisation of flavonoids in extracts of fresh herbs. J. Chromatogr. A 902: 369–379. Kind, T., and Fiehn, O. (2006). Metabolomic database annotations via query of elemental compositions: mass accuracy is insufficient even at less than 1 ppm. BMC Bioinformatics 7: 234. Klie, S., Mutwil, M., Persson, S., and Nikoloski, Z. (2012). Inferring gene functions through dissection of relevance networks: interleaving the intra- and inter-species views. Mol. Biosyst. 8: 2233–2241. Kopka, J., et al. (2005). [email protected]: The Golm Metabolome Database. Bioinformatics 21: 1635–1638. Loewenstein, Y., Raimondo, D., Redfern, O.C., Watson, J., Frishman, D., Linial, M., Orengo, C., Thornton, J., and Tramontano, A. (2009). Protein function annotation by homology-based inference. Genome Biol. 10: 207. Matsuda, F., Hirai, M.Y., Sasaki, E., Akiyama, K., YonekuraSakakibara, K., Provart, N.J., Sakurai, T., Shimada, Y., and Saito, K. (2010). AtMetExpress development: a phytochemical atlas of Arabidopsis development. Plant Physiol. 152: 566–578. Montaut, S., and Bleeker, R.S. (2010). Isolation and structure elucidation of 59-O-b-D-glucopyranosyl-dihydroascorbigen from Cardamine diphylla rhizome. Carbohydr. Res. 345: 1968–1970. Morreel, K., Kim, H., Lu, F., Dima, O., Akiyama, T., Vanholme, R., Niculaes, C., Goeminne, G., Inzé, D., Messens, E., Ralph, J., and Boerjan, W. (2010). Mass spectrometry-based fragmentation as an identification tool in lignomics. Anal. Chem. 82: 8095–8105. Morreel, K., Ralph, J., Kim, H., Lu, F., Goeminne, G., Ralph, S., Messens, E., and Boerjan, W. (2004). Profiling of oligolignols

reveals monolignol coupling conditions in lignifying poplar xylem. Plant Physiol. 136: 3537–3549. Müller-Linow, M., Weckwerth, W., and Hütt, M.-T. (2007). Consistency analysis of metabolic correlation networks. BMC Syst. Biol. 1: 44. Neumann, S., and Böcker, S. (2010). Computational mass spectrometry for metabolomics: identification of metabolites and small molecules. Anal. Bioanal. Chem. 398: 2779–2788. Nicholson, J.K., Lindon, J.C., and Holmes, E. (1999). ‘Metabonomics’: understanding the metabolic responses of living systems to pathophysiological stimuli via multivariate statistical analysis of biological NMR spectroscopic data. Xenobiotica 29: 1181–1189. Oliver, S.G., Winson, M.K., Kell, D.B., and Baganz, F. (1998). Systematic functional analysis of the yeast genome. Trends Biotechnol. 16: 373–378. Rasche, F., Scheubert, K., Hufsky, F., Zichner, T., Kai, M., Svatoš, A., and Böcker, S. (2012). Identifying the unknowns by aligning fragmentation trees. Anal. Chem. 84: 3417–3426. Rochfort, S.J., Trenerry, V.C., Imsic, M., Panozzo, J., and Jones, R. (2008). Class targeted metabolomics: ESI ion trap screening methods for glucosinolates based on MSn fragmentation. Phytochemistry 69: 1671–1679. Roessner, U., Luedemann, A., Brust, D., Fiehn, O., Linke, T., Willmitzer, L., and Fernie, A.R. (2001). Metabolic profiling allows comprehensive phenotyping of genetically or environmentally modified plant systems. Plant Cell 13: 11–29. Rojas-Cherto, M., Peironcely, J.E., Kasper, P.T., van der Hooft, J.J.J., de Vos, R.C.H., Vreeken, R., Hankemeier, T., and Reijmers, T. (2012). Metabolite identification using automated comparison of high-resolution multistage mass spectral trees. Anal. Chem. 84: 5524–5534. Sato, S., Soga, T., Nishioka, T., and Tomita, M. (2004). Simultaneous determination of the main metabolites in rice leaves using capillary electrophoresis mass spectrometry and capillary electrophoresis diode array detection. Plant J. 40: 151–163. Smith, C.A., O’Maille, G., Want, E.J., Qin, C., Trauger, S.A., Brandon, T.R., Custodio, D.E., Abagyan, R., and Siuzdak, G. (2005). METLIN: a metabolite mass spectral database. Ther. Drug Monit. 27: 747–751. Smith, C.A., Want, E.J., O’Maille, G., Abagyan, R., and Siuzdak, G. (2006). XCMS: Processing mass spectrometry data for metabolite profiling using nonlinear peak alignment, matching, and identification. Anal. Chem. 78: 779–787. Soga, T., Ohashi, Y., Ueno, Y., Naraoka, H., Tomita, M., and Nishioka, T. (2003). Quantitative metabolome analysis using capillary electrophoresis mass spectrometry. J. Proteome Res. 2: 488–494. Sønderby, I.E., Geu-Flores, F., and Halkier, B.A. (2010). Biosynthesis of glucosinolates: Gene discovery and beyond. Trends Plant Sci. 15: 283–290. Stein, S.E., and Scott, D.R. (1994). Optimization and testing of mass spectral library search algorithms for compound identification. J. Am. Soc. Mass Spectrom. 5: 859–866. Steuer, R., Kurths, J., Fiehn, O., and Weckwerth, W. (2003). Observing and interpreting correlations in metabolomic networks. Bioinformatics 19: 1019–1026. Sumner, L.W., et al. (2007). Proposed minimum reporting standards for chemical analysis Chemical Analysis Working Group (CAWG) Metabolomics Standards Initiative (MSI). Metabolomics 3: 211– 221. Tolstikov, V.V., and Fiehn, O. (2002). Analysis of highly polar compounds of plant origin: combination of hydrophilic interaction chromatography and electrospray ion trap mass spectrometry. Anal. Biochem. 301: 298–307.

Candidate Substrate-Product Pair Networks

Tweeddale, H., Notley-McRobb, L., and Ferenci, T. (1999). Assessing the effect of reactive oxygen species on Escherichia coli using a metabolome approach. Redox Rep. 4: 237–241. von Roepenack-Lahaye, E., Degenkolb, T., Zerjeski, M., Franz, M., Roth, U., Wessjohann, L., Schmidt, J., Scheel, D., and Clemens, S. (2004). Profiling of Arabidopsis secondary metabolites by capillary liquid chromatography coupled to electrospray ionization quadrupole time-of-flight mass spectrometry. Plant Physiol. 134: 548–559.

945

Watson, J.D., Sanderson, S., Ezersky, A., Savchenko, A., Edwards, A., Orengo, C., Joachimiak, A., Laskowski, R.A., and Thornton, J.M. (2007). Towards fully automated structure-based function prediction in structural genomics: A case study. J. Mol. Biol. 367: 1511–1522. Wolf, S., Schmidt, S., Müller-Hannemann, M., and Neumann, S. (2010). In silico fragmentation for computer assisted identification of metabolite mass spectra. BMC Bioinformatics 11: 148.

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

SUPPLEMENTAL FIGURES

Supplemental Figure 1. Peak Grouping Surface Plots. In order to group peaks belonging to the same compound, various settings of the retention time window and the correlation threshold were evaluated on the number of peaks that could be grouped (A), the number of peak groups (B), the average number of assigned peaks per peak group (C) and on an arbitrarily trait representing the co-optimization of both the number of peak groups and the number of assigned peaks to a peak group (D). For the latter trait, to allow that the number of peak groups had a similar impact as the number of assigned peaks, the number of peak groups was multiplied by a correction factor, i.e., the quotient of the average number of assigned peaks and the average number of peak groups. The trait was subsequently obtained by multiplying the corrected number of peak groups with the number of assigned peaks.

1

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Supplemental Figure 2. CSPP Generation Algorithm. Integrated and aligned chromatogram peaks are ordered with increasing   m/z   (“node”   file)   and   used   for   the   “edge”   file   generation   (see   Methods   for   explanation). The CSPP algorithm, based on searching peak pairs of which the m/z values differ by the mass expected for e.g., a hexosylation (162.053 Da) and the elution order is in agreement with that expected upon a reversed phase separation – i.e., the hexosylated product eluting earlier – is mentioned in pseudo-code.

2

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Supplemental Figure 3. Retention Time Difference Distributions. Distribution of the number of peak pairs or CSPPs versus the retention time difference between both peaks of each peak pair or between “substrate”   and   “product”   peak   of   each   CSPP,   respectively.   For   a   particular   mass   difference,   CSPPs   are obtained by truncating the peak pair-based retention time difference distribution. More specifically, when   the   “product”   is   more   polar   or   apolar   than   the   “substrate”,   only   negative   or   positive   retention   time differences are allowed, respectively. In any case, a retention time difference of 0 ± 0.2 min is excluded. SD, standard deviation.

3

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Supplemental Figure 4. Annotated Molecular Structures.

4

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

5

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

6

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

7

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

8

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

9

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Supplemental Figure 4. Annotated Molecular Structures. See the footnote of Supplemental Data Set 1 for structural elucidation details. Identity, linkage and attachment position of sugars should be interpreted with caution as they cannot be determined by MS2.

10

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

11

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Supplemental Figure 5. Partial CSPP Sub-networks of Flavonoids (A) and Phenylpropanoids (B). The sub-network for the flavonoids was elaborated starting with one of the more prominent flavonols, i.e., kaempherol 3-O-glucosyl-7-O-rhamnoside 37,   as   “bait”   (Veit and Pauli, 1999; YonekuraSakakibara et al., 2008) (Supplemental Data Set 1). The phenylpropanoid metabolism sub-network was obtained by using the nodes representing sinapoyl malate (58 and 59), sinapoyl glucose (53 and 55) and disinapoyl glucose 64 as   “baits”   (Supplemental   Data Set 1). From the 15 flavonols and 22 phenylpropanoid derivatives obtained from the complete sub-networks (Supplemental Data Set 1), only a limited number of compounds are shown to improve the visibility. The phenylpropanoid subnetwork also shows oligolignols/(neo)lignans that were highly correlated with sinapoyl malate and/or sinapoyl glucose. Nodes and edges represent chromatogram peaks and (bio)chemical conversions (see Table 1 for conversion types). Node labels are based on the XCMS integration and alignment algorithm. Whenever the similarity between the MS2 spectra of candidate substrate and product exceeds 0.8, the edge label is black. The color brightness of the edge reflects the Pearson productmoment correlation coefficient   between   the   levels   of   the   CSPP   “substrate”   and   “product”   (blue   and   yellow represent a negative and positive correlation). Based on the MS n spectra, the identification of hexose, deoxyhexose and pentose residues is not possible. However, up to now, only 3–O- and/or 7–Olinked glucose (Glc), rhamnose (Rha) and arabinose (Ara) flavonols have been observed in Arabidopsis (Yonekura-Sakakibara et al., 2008). Shorthand naming of oligolignols/(neo)lignans is based on (Morreel et al., 2004): units derived from coniferyl alcohol, sinapyl alcohol and ferulic acid are abbreviated as G (guaiacyl unit), S (syringyl unit) and FA, whereas the linkage type is indicated between brackets (see footnote Supplemental Data Set 1). Ara arabinose, Glc glucose, Hex hexose, ISF ion source fragment, Kae kaempherol, Mal-Glc malonyl glucose, Que quercetin, Rha rhamnose.

12

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Supplemental Figure 6. CSPP Sub-network of Highly Correlated CSPPs. The sub-network represents mainly the (neo)lignans/oligolignols. An enlarged part of the major cluster is displayed whereas an overview is shown in the upper right corner. Nodes and edges represent chromatogram peaks and (bio)chemical conversions (see Table 1 for conversion types). The darkness of the edge label and the color brightness of the edge reflect the MS2 spectral similarity and the Pearson product-moment correlation coefficient (see Supplemental Figure 5). Shorthand naming of (neo)lignans/oligolignols is based on (Morreel et al., 2004) (see footnote Supplemental Data Set 1).

13

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Supplemental Figure 7. Synthesis of G(8–O–4)FA Glu.

14

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Supplemental Figure 8. Gas Phase Fragmentation Pathways of Simple Phenolics and Phenylpropanoids (A), 5΄-O-β-D-glucosyl Dihydroascorbigen (B) and Corchoionoside C Anions (C).

15

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

16

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Supplemental Figure 8. Gas Phase Fragmentation Pathways of Simple Phenolics and Phenylpropanoids (A), 5΄-O-β-D-glucosyl Dihydroascorbigen (B) and Corchoionoside C Anions (C).

17

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Supplemental Figure 9. Effect of Shared Control on Correlation. When the levels of two metabolites are controlled by e.g., two enzymes and the co-response on both metabolite levels of a change in enzymatic rate is the same for both enzymes, a high correlation will result (left plot). Otherwise a moderate correlation appears (right plot).

18

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

SUPPLEMENTAL TABLES Supplemental Table 1. MS2 spectra of simple phenolics 3-hydroxy-benzoic acid

35

35

30

35

35

35

35

35

123(21)

123(0)

153(100)

151(0)

181(0)

121(0)

137(0)

137(0)

138(31)

136(100)

166(100)

3,5-dimethoxy4-hydroxybenzoic acid

4-hydroxy-benzoic acid

Hydroxybenzoic acids

benzoic acid

4-hydroxy-3,5dimethoxybenzaldehyde

4-hydroxy-3methoxybenzaldehyde

4-hydroxy-3methoxybenzyl alcohol

2-hydroxybenzyl alcohol

Hydroxybenzaldehydes

3-hydroxy-4methoxybenzoi c acid

CE (%) [M-H] a [M-H-CH3] [M-H-H2O] [M-H-CH2O]

4-hydroxybenzyl alcohol

neutral (g/mol)

loss

Hydroxybenzyl alcohols

15 18

105(100)

30

93(3)

93(100)

CE (%) [M-H] a [M-H-CH3] [M-H-H2O] [M-H-CO2] [M-H-CO2-CH3]

35

28

30

30

30

30

35

153(1)

153(100)

153(23)

153(9)

167(7)

167(10)

197(0)

152(67)

152(100)

182(100)

123(100)

123(65)

153(59)

108(10)

108(5)

15 18

135(100)

135(82)

44

109(83)

109(33)

59

4-hydroxy-3methoxybenzoi c acid

2,5-dihydroxybenzoic acid

2,3-dihydroxybenzoic acid

2,6-dihydroxybenzoic acid

2,4-dihydroxybenzoic acid

neutral loss (g/mol)

Hydroxybenzoic acids

109(100)

109(100)

Relative intensity of the daughter ions as compared to the base peak is given between brackets. a, Elimination of a methyl radical from methoxylated benzene has been described by Reeks et al. (1993). CE, collision energy.

19

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Supplemental Table 2. MS2 spectra of monolignol-related compounds.

CE (%) [M-H] a [M-H-CH3] [M-H-CH4] b [M-H-H2O] b [M-H-CH2O] [M-H-CH3OH] [M-H-H2O-CH3] [M-H-CO2] [M-H-CO2-H2O] [M-H-CO2-CO]

15

3,4-dihydroxyhydrocinnamic acid

4-hydroxyhydrocinnamic acid

Hydroxycinnamic acids 3,5-dimethoxy-4hydroxyhydrocinnamyl alcohol

3,5-dimethoxy-4hydroxycinnamyl alcohol

4-hydroxy-3methoxy-cinnamyl alcohol

Hydroxycinnamyl alcohols cinnamyl alcohol

4,5-dihydroxy-3methoxycinnamaldehyde

4-hydroxy-3methyoxycinnamaldehyde

neutral (g/mol)

loss

Hydroxycinnamaldehydes

35

35

35

35

35

35

35

28

177(5)

193(0)

133(0)

179(0)

209(0)

211(0)

165(0)

181(3)

162(100)

178(100)

164(100)

194(100)

196(100)

121(100)

137(100)

16

193(14)

18

115(100)

30

103(33)

191(33)

32

177(4)

33

176(9)

44 62

119(5)

72

93(16)

CE (%) [M-H] a 15 [M-H-CH3] 44 [M-H-CO2] [M-H-CO2-CH3] 59

ferulic acid ethyl ester

4-hydroxy-3,5dimethoxycinnamic acid

3-hydroxy-4methoxycinnamic acid

4-hydroxy-3methoxycinnamic acid

3,4-dihydroxycinnamic acid

2-hydroxycinnamic acid

4-hydroxycinnamic acid

cinnamic acid

neutral loss (g/mol)

Hydroxycinnamic acids

35

30

35

36

32

34

35

35

147(0)

163(3)

163(0)

179(0)

193(1)

193(2)

223(0)

221(0)

178(48)

178(100)

208(100)

206(100)

103(100)

119(100)

119(100)

135(100)

149(100)

149(4)

179(42)

134(12)

134(3)

164(26)

Relative intensity of the daughter ions as compared to the base peak is given between brackets. a, Elimination of a methyl radical from methoxylated benzene has been described by Reeks et al. (1993). b, The collision-induced dissociation of ionized primary alcohols has been described by Bowie (1990). CE, collision energy. 20

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

SUPPLEMENTAL METHODS Chemicals Coniferaldehyde, sinapyl alcohol, dihydrocaffeic acid, p-coumaric acid, ferulic acid, 2hydroxybenzyl

alcohol,

2,6-dihydroxybenzoic

acid,

2,3-dihydroxybenzoic

acid,

3,5-

dihydroxybenzoic acid, 2,5-dihydroxybenzoic acid, indole-3-carboxylate, Trp and abscisic acid were purchased from Aldrich (Steinheim, Germany); vanillic acid, isovanillic acid, 4hydroxybenzyl alcohol, 4-hydroxy-3-methoxybenzyl alcohol, coniferyl alcohol and isoferulic acid were obtained from Acros (Geel, Belgium). Caffeic acid and 4-hydroxy-3,5dimethoxybenzaldehyde were bought at Janssen (Beerse, Belgium) and Roth (Karlsruhe, Germany), respectively. Direct Infusion MSn Analysis of Standards A 100 M solution of each standard, flowing at a rate of 10 l/min, was mixed with a flow of 300 l/min (water : methanol, 50:50 (v:v), 0.1% acetate) before entering a LCQ Classic ion trap MS (IT-MS) upgraded to a LCQ Deca (Thermo Fisher Scientific, Waltham, MA). With this MS instrument, the standards were more sensitively detected using Atmospheric Pressure Chemical Ionization (APCI) than when using ESI. Analytes were negatively ionized by APCI using the following parameter values: capillary temperature 150°C, vaporizer temperature 350°C, sheath gas 25 (arb), aux gas 3 (arb), source current 5 A. MSn analysis was performed by collision induced dissociation (CID) using He as the collision gas. The MSn spectra were analyzed with Xcalibur vs 1.2. MS2 Spectra of Benzenoid and Phenylpropanoid Standards The gas phase-based fragmentation of negatively ionized molecules is much less understood than that of positively ionized molecules. The main reasons are that anions fragment much more via charge-remote fragmentations, homolytic cleavages, rearrangements and ion-neutral complexes than cations (Bowie, 1990; Eichinger et al., 1994; Cheng and Gross, 2000). However, the various textbook organic chemistry reactions are much more reflected in the gas-phase fragmentation behavior of anions than in those of cations (Born et al., 1997; DePuy, 2000; Gronert, 2001,

21

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

2005). Therefore, studies that embark on elucidating the fragmentation pathways of anions increasingly appear in the literature. Among the four major classes of compounds (Supplemental Data Set 1) displayed by the CSPP networks, the low-energy collision-induced dissociation (CID) of the anions from only three compound classes have been extensively studied: glucosinolates (e.g. Fabre et al., 2007; Rochfort et al., 2008; Bialecki et al., 2010; Cataldi et al., 2010), flavonoids (e.g. Justesen, 2000; Fabre et al., 2001; Hughes et al., 2001; Hvattum and Ekeberg, 2003; Cuyckens and Claeys, 2004; Ferreres et al., 2004; Morreel et al., 2006; Yan et al., 2007) and (neo)lignans/oligolignols (e.g. Morreel et al., 2004; Eklund et al., 2008; Ricci et al., 2008; Schmidt et al., 2008; Morreel et al., 2010b; Morreel et al., 2010a; Ricci et al., 2010). Literature data for the fragmentation of the fourth major class displayed by the CSPP networks, benzenoids and phenylpropanoid anions, are available but their gas-phase fragmentation behavior upon low-energy CID has not been comprehensively analyzed. Therefore, various benzenoids and phenylpropanoids were subjected to negative ionization low-energy CID in an IT-MS instrument. Although both charge-driven, i.e., fragmentations that start from the most acidic site (Thevis et al., 2003), and charge-remote reactions might be responsible for the fragmentations, the former type will occur whenever possible (Cheng and Gross, 2000). Therefore charge-driven fragmentation pathways are proposed in this study. In the absence of a carboxylic acid function, ionization of benzenoids and phenylpropanoids will lead to a phenoxide anion from which the charge-driven CID pathways will proceed. In case both a carboxylic acid and a phenolic function are present, the carboxylic acid function will mainly take up the charge as it is more acidic in the gas-phase as compared to the phenolic function (e.g., 348.2 and 349.2 kcal/mol for acetic acid and phenol, respectively) (Harrison, 1992). In that case, the acid function will be mainly responsible for the fragmentation initiation. The MS2 product ions of various hydroxybenzoic acid, hydroxybenzyl alcohol and hydroxybenzaldehyde anions are listed in Supplemental Table 1. Upon fragmentation in the negative mode, alcohols may lose water and/or formaldehyde (Bowie, 1990). Both neutral losses are also observed with hydroxybenzyl alcohols (Supplemental Table 1), but their importance depends on the relative position of the phenol and the alcohol. When the phenol is in the para position of the hydroxymethyl functionality, water loss is favored (Supplemental Figure 8A.A), whereas formaldehyde loss is favored when the phenol is in the ortho-position (Supplemental Figure 8A.B). In case of a phenol in the para position of the hydroxymethyl, water loss is 22

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

initiated when the phenoxide anion of 4-hydroxybenzyl alcohol is converted to a quinone methide with the subsequent elimination of a hydroxide anion. Prior to the dissociation of this neutral/anion complex, the hydroxide anion will abstract a proton from the quinone methide (Bowie, 1990). The spectrum of 2-hydroxybenzylalcohol is clearly dominated by a so-called ortho effect (Supplemental Figure 8A.B): the phenoxide anion abstracts a proton from the alcohol function via a six-membered cyclic transition state in a McLafferty-type rearrangement (Grossert et al., 2006). The resulting alkoxide anion then undergoes a 1,2-elimination (Bowie, 1990). Ionization of hydroxybenzoic acids yields carboxylate anions. The spectra of these anions are characterized by a neutral loss of 44 g/mol due to decarboxylation (Supplemental Table 1), which was also observed by Levsel et al. (2007). Nevertheless, decarboxylation was hardly observed for benzoic acid or monohydroxybenzoic acids. This agrees with the conclusion of Bandu et al. (2004) that multiple electron-withdrawing groups should be present on the benzene ring before decarboxylation occurs. Interestingly, a product ion arising from the loss of water was only observed for 2,6- and 2,4-dihydroxybenzoic acid. A reaction mechanism for this water loss due to an ortho effect is proposed in Supplemental Figure 8A.C. Following the proton transfer from the C2 hydroxyl group to the carboxylate anion in a McLafferty-type rearrangement involving a six-membered cyclic transition state (Bandu et al., 2006; Grossert et al., 2006), an internal nucleophilic acyl substitution occurs between the C2 oxyanion and the carboxylic acid in which a hydroxide anion is expelled (Attygalle et al., 2006). Water elimination is then mediated by an anion/neutral complex (Bowie, 1990). Because of the absence of a product peak associated with water loss in the mass spectra of other positional isomers of dihydroxybenzoic acid, this reaction can only proceed if the proton can be taken from a hydroxyl function at C 4 or C6; due to the electron-withdrawing carbonyl function attached to C1, hydroxyl functions at C4 or C6 are more acidic than those at the C3 or C5 position. In the absence of a carboxylic acid function, the major loss observed in the spectra of monolignols and monolignol-related compounds was the charge-remote elimination of a neutral methyl radical (Supplemental Table 2). The most elaborate MS2 spectrum was observed for 4hydroxy-3,5-dimethoxycinnamyl alcohol. Besides methyl radical loss, a major MS2 peak corresponding with dehydration was observed. This dehydration likely proceeds by a reaction mechanism (Supplemental Figure 8A.D) that is similar to that described for hydroxybenzyl 23

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

alcohols (Supplemental Figure 8A.A). The absence of water loss in the spectrum of 4-hydroxy3,5-dimethoxyhydrocinnamyl alcohol supports the involvement of the aliphatic double bond in the loss of water in 4-hydroxy-3,5-dimethoxycinnamyl alcohol. The loss of water is observed when both ortho positions of the phenol group are methoxylated and not when only one ortho position is methoxylated, proving that the substitution of the aromatic ring influences the stability of the anion. Minor abundant product ions in the spectrum of 4-hydroxy-3,5dimethoxycinnamyl alcohol originate from the loss of methanol, methane and the combined elimination of water and a methyl radical. Methanol loss likely occurs by a SN2-type mechanism in which the expelled hydroxide anion acts as the nucleophile (Supplemental Figure 8A.F). The proposed fragmentation mechanism for the loss of methane is given in Supplemental Figure 8A.E. Decarboxylation was the main fragmentation pathway of the hydroxycinnamic acids (Supplemental Table 2). Although carboxylate anions are reported to loose water upon conversion to their enolate anions (Bowie, 1990) (Supplemental Figure 8A.G) with subsequent ketene formation and the elimination of a hydroxide anion, this pathway was not observed in the IT-MS obtained fragmentation spectra of hydroxycinnamic acids. Likely, such a pathway is more favored under high-energy CID, although it has been sporadically suggested to occur under lowenergy CID (Kanawati et al., 2007; Kanawati et al., 2008; Kanawati and Schmitt-Kopplin, 2010). Methyl radical loss of 4-hydroxy-3-methoxycinnamic acid was less pronounced than for 3hydroxy-4-methoxycinnamic acid owing to the greater degree of radical delocalization in the latter. Finally, the specific loss of 62 Da in the spectrum of dihydrocaffeic acid indicates that a flexible side chain is necessary for this fragmentation. As the corresponding m/z 119 ion is also formed upon MS3 of the m/z 137 first product ion, this loss represents a combined decarboxylation and dehydration. The additional water loss likely occurs by a charge-remote process involving the ortho-dihydroxybenzene moiety of the compound. In general, the major fragmentation reactions observed for all of these phenolics were decarboxylation when a carboxylic acid was present and a demethylation when an aromatic methoxygroup was present. In the absence of ortho effects, water and formaldehyde eliminations were associated with aliphatic alcoholgroups. No dissociation of the phenolic function itself was observed. Hydroxycinnamaldehydes did not show specific fragmentation mechanisms.

24

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Structural Elucidation of MSn Spectra Structural elucidation was based on knowledge obtained from the curated CSPP network (see Results) and further supported by interpretation of the MSn spectra whenever possible. In metabolomics, a full identification (based on purification of the unknown compound followed by NMR analysis, or by spiking a purchased or synthesized standard) of all molecules is not possible. Therefore, two other levels of structural elucidation, i.e., structural annotation and structural characterization, have been defined (Sumner et al., 2007). Below, a structural annotation (Sumner et al., 2007) is obtained whenever MSn information was used in addition to CSPP network information (see Supplemental Data Set 1 for information obtained from the CSPP network). In the absence of MSn data, a structural characterization, based solely on the information from the CSPP network, is performed (Sumner et al., 2007). Below, the structural annotations and characterizations are described for the 145 compounds of which the structure could be predicted. The MSn elucidation approach is more elaborately explained and referenced for the first representative compounds from each structural type. Glucosinolates 1. 4-methylthiobutyl Gluc The chemical formula of the anion of compound 1 was C12H22O9NS3 (m/z 420.04612). The base peak in its MS2 spectrum appeared at m/z 259, a first product ion which is characteristic for glucosinolates and that is derived from the common moiety in glucosinolates consisting of a glucose moiety attached to a sulphated thiohydroximate. This ion at m/z 259 is formed via a rearrangement and represents a sulphated glucose moiety (Rochfort et al., 2008; Bialecki et al., 2010; Cataldi et al., 2010). Further support was obtained from the MS3 spectrum of the m/z 259 ion that was identical to the one published previously (Rochfort et al., 2008). Other first product ions associated with the common glucose moiety in glucosinolates were observed as well at m/z 291, 275, 241 and 195 (Fabre et al., 2007; Bialecki et al., 2010; Cataldi et al., 2010). First product ions associated with the variable side-chain of glucosinolates (Fabre et al., 2007) were observed at m/z 340 (-80 Da, sulphite loss), 242 (-178 Da, loss of thio-glucose fragment), 224 (196 Da, thio-glucose loss) and 178 (-242 Da, combined loss of glucose and sulphite). Based on the chemical formula, this compound is a saturated, methionine-derived glucosinolate. Therefore, compound 1 is 4-methylthiobutyl glucosinolate (Gluc) or glucoerucin. Matching to MassBank 25

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

returned a hit and score of 4 and 0.42, respectively. The low score arose from the difference in the applied collision energy to obtain the library spectrum and the one used in current study. 2. 5-methylthiopentyl Gluc The ion of compound 2 had C13H24O9NS3 (m/z 434.06176) as chemical formula and was associated with 4-methylthiobutyl Gluc 1 in the CSPP networks. Its MS2 spectrum was very similar to that of 4-methylthiobutyl Gluc 1. However, the first product ions associated with the variable side-chain were all shifted by 14 amu in the MS2 spectrum of compound 2 as compared to those in the MS2 spectrum of 4-methylthiobutyl Gluc 1. Consequently, this compound was elucidated as 5-methylthiopentyl glucosinolate or glucoberteroin. Further support was obtained from its retention time: compound 2 eluted 2.4 min later than 4-methylthiobutyl Gluc 1, but 2.4 min earlier than the next member in this homologous series, i.e., 6-methylthiobutyl Gluc 3. 3. 6-methylthiohexyl Gluc Also for this compound with chemical formula C14H26O9NS3 (m/z 448.07705), a similar MS2 spectrum was obtained as that of 5-methylthiopentyl Gluc 2. Again, the main differences were the m/z values of the first product ions associated with the variable side-chain; all were shifted by 14 amu as compared to those in the MS2 spectrum of 5-methylthiopentyl Gluc 2. Compound 3 was characterized as 6-methylthiohexyl glucosinolate or glucolesquerellin. 4. 7-methylthioheptyl Gluc The anion of 4 had C15H28O9NS3 (m/z 462.09239) as chemical formula and its MS2 spectrum was very similar to that of 4-methylthiobutyl Gluc 1, mainly differing by the first product ions associated with the variable side-chain: they were all shifted with 42 amu as compared to the corresponding m/z peaks in the MS2 spectrum of 4-methylthiobutyl Gluc 1. Therefore, this compound was annotated as 7-methylthioheptyl glucosinolate. Matching to MassBank rendered a score and hit of 0.84 and 7, respectively. 5. 8-methylthiooctyl Gluc Again, the MS2 spectrum of the anion of 5, having a chemical formula equal to C16H30O9NS3 (m/z 476.10812), was very similar to that of compounds 1 and 4, essentially differing by a shift of 56 amu and 14 amu, respectively, for all first product ions representing the variable sidechain. Compound 5 is 8-methylthiooctyl glucosinolate. A score and hit of 0.58 and 7 were obtained when matching to MassBank. 26

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

6. 3-methylsulfinylpropyl Gluc The ion of compound 6 had a chemical formula of C11H20O10NS3 (m/z 422.02544). Together with first product ions at m/z 259, 275 and 291 in its MS2 spectrum, this compound is a methionine-derived glucosinolate. Opposite to the MS2 spectra of the methylthioalkyl glucosinolates (compounds 1, 4 and 5), the first product ion at m/z 358 was the base peak rather than the m/z 259 ion. This base peak arises from a loss of 64 Da which is derived from a methylsulfinyl endgroup (Fabre et al., 2007). The methylsulfinylalkyl side-chain looses also a methyl radical yielding the first product ion at m/z 407. Therefore, compound 6 is 3methylsulfinylpropyl glucosinolate or glucoiberin. Matching to MassBank returned a score and hit of 0.63 and 9. 7. 4-methylsulfinylbutyl Gluc The MS2 spectrum of the anion of 7, having a chemical formula of C12H22O10NS3 (m/z 436.04079), was similar to that of 3-methylsulfinylpropyl Gluc 6, but the base peak was shifted with 14 amu, leading to the annotation of compound 7 as 4-methylsulfinylbutyl glucosinolate or glucoraphanin. A score and hit of 0.72 and 6 were obtained when matching to MassBank. 8. 4-methylsulfinylbutyl Gluc The chemical formula (C12H22O10NS3, m/z 436.04080) and the MS2 spectrum of the anion of 8 were identical to 4-methylsulfinylbutyl Gluc 7 and, thus, compound 8 is characterized as an isomer. Matching to MassBank yielded a score and hit of 0.72 and 6. 9. 5-methylsulfinylpentyl Gluc The ion of compound 9 had C13H24O10NS3 (m/z 450.05654) as chemical formula and its MS2 spectrum was similar to that of compounds 6, 7 and 8. However, compared to the MS2 spectrum of compounds 7 and 8, the base peak and the first product ion arising from methyl radical loss were shifted with 14 amu. Compound 9 is annotated as 5-methylsulfinylpentyl glucosinolate or glucoalyssin. A score and hit of 0.79 and 5 were obtained when matching to MassBank. 10. 6-methylsulfinylhexyl Gluc With a chemical formula of C14H26O10NS3 (m/z 464.07213) and a MS2 spectrum similar to those of compounds 6, 7, 8 and 9 except for the m/z shifts of the base peak and of the first product ion associated with methyl radical loss, this compound is annotated as 6-methylsulfinylhexyl glucosinolate or glucohesperin. Matching to the MassBank did not yield any result. Searching for 27

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

glucohesperin showed that it was not present in the database. Structure prediction with MetFrag (Wolf et al., 2010) using the PubChem database explained the variable side-chain-characteristic first product ions at m/z 449 and 400, but could not explain any of the glucose moiety-based first product ions that are characteristic for the CID spectra of glucosinolates. This underscores the weakness of current MS2 structural prediction programs in the case of negative ions (Heinonen et al., 2008). 11. 7-methylsulfinylheptyl Gluc The chemical formula of the anion was C15H28O10NS3 (m/z 478.08779) and the MS2 spectrum indicates that this compound is also a methylsulfinylalkyl glucosinolate. More specifically, compound 11 was annotated as 7-methylsulfinylheptyl glucosinolate or glucoibarin. Matching to MassBank returned a score and hit of 0.77 and 7. 12. 8-methylsulfinyloctyl Gluc Based on the MS2 spectrum and the chemical formula (C16H30O10NS3, m/z 492.10339), this compound belonged also to the methylsulfinylalkyl glucosinolates. Compound 12 was annotated as 8-methylsulfinyloctyl Gluc or glucohirsutin. Comparison to the corresponding MassBank spectrum rendered a score and hit of 0.71 and 7. 13. 9-methylsulfinylnonyl Gluc The chemical formula (C17H32O10NS3, m/z 506.11965) and the MS2 spectrum were representative for methylsulfinylalkyl glucosinolates, and aided the structural annotation of compound 13 as 9-methylsulfinylnonyl glucosinolate or glucoarabin. Glucoarabin was not traced in the MassBank database. 14. hydroxy-4-(methylsulfinyl)-butyl Gluc The ion of this compound had a chemical formula of C12H22O11NS3 (m/z 452.03595) and was closely associated with the methylsulfinylalkyl glucosinolates in the CSPP networks. Its structure was characterized as hydroxy-4-(methylsulfinyl)-butyl glucosinolate, yet no MS2 spectrum was available for confirmation. 15. 3-hydroxy-5-(methylsulfinyl)-pentyl Gluc Based on its close association with hydroxy-4-(methylsulfinyl)-butyl Gluc 14 in the CSPP networks  and  its  anion’s  chemical  formula  of  C13H24O11NS3 (m/z 466.05157), compound 15 was characterized as hydroxy-5-(methylsulfinyl)-pentyl glucosinolate. In the plant kingdom, a similar 28

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

compound has been purified in which the hydroxyl group was attached to the 3-position (Fabre et al., 2007). 16. hydroxy-8-(methylsulfinyl)-octyl Gluc The ion of compound 16 had a chemical formula of C16H30O11NS3 (m/z 508.09845) and was the first compound in the hydroxy-(methylsulfinyl)-alkyl glucosinolate series for which MSn spectra were obtained, supporting as well the structural characterizations of compounds 14, 15 and 17. The MS2 spectrum of compound 16 showed first product ions at m/z 259 and 291 reminiscent of the common glucosinolate moiety, yet the base peak was observed at m/z 444 owing to a loss of 64 Da that is characteristic for a methylsulfinyl endgroup. Based on its MS2 spectrum, the additional oxygen in its chemical formula as compared to the chemical formula of the corresponding methylsulfinyloctyl glucosinolate, should be present as a hydroxyl function on the variable side-chain of the glucosinolate. Therefore, compound 16 was annotated as hydroxy-8(methylsulfinyl)-octyl glucosinolate. No hit with the same chemical formula was obtained when matching to MassBank. 17. hydroxy-6-(methylsulfinyl)-hexyl Gluc Based on the chemical formula, C14H26O11NS3 (m/z 480.06348), and its close association with the methylsulfinylalkyl glucosinolates in the CSPP networks, this compound was structurally characterized as hydroxy-6-(methylsulfinyl)-hexyl glucosinolate. No MS2 spectrum was obtained to verify this structure. 18. hydroxy-8-(methylsulfinyl)-octyl Gluc The ion of compound 18 had the same chemical formula (C16H30O11NS3, m/z 508.09845) as hydroxy-8-(methylsulfinyl)-octyl Gluc 16. Also its MS2 spectrum was similar except that m/z 291 was the base peak rather than the m/z 444 first product ion. In addition, a new first product ion at m/z 391 appeared. The differences in the fragmentation pattern of the glucosinolate common moiety as compared to those of other glucosinolates arises from the effect of the additional hydroxyl function, suggesting that the hydroxyl function is attached close to the glucose thiohydroximate moiety. No hit with the same chemical formula was obtained when matching to MassBank. 19. 3-methylbutyl Gluc

29

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

The chemical formula of anion 19 was C12H22O9NS2 (m/z 388.07390). Its MS2 spectrum was similar to those of the methylthioalkyl glucosinolates, i.e., dominated by first product ions due to fragmentations of the common moiety. The absence of fragmentations typical for the variable side-chain moiety suggested that no functional groups were present on the variable side-chain. Therefore, compound 19 is a leucine-derived glucosinolate and was annotated as 3-methylbutyl glucosinolate. No hit with the same chemical formula was obtained when matching to MassBank. 20. 4-methylpentyl Gluc The ion of compound 20 had C13H24O9NS2 (m/z 402.08964) as chemical formula and its MS2 spectrum was similar to that of 3-methylbutyl Gluc 19 except that the first product ion due to sulphite loss (-80 Da) was shifted by 14 amu. Therefore, this compound was annotated as 4methylpentyl glucosinolate. No hit with the same chemical formula was obtained when matching to MassBank. 21. methylpentyl Gluc Compound 21 had the same chemical formula (ion at m/z 402.08973, C13H24O9NS2) as 4methylpentyl Gluc 20, but no MS2 spectrum was obtained. Compound 21 was characterized as methylpentyl glucosinolate. 22. 5-methylhexyl Gluc Based  on  its  anion’s  chemical  formula  (C14H26O9NS2, m/z 416.10553) and its MS2 spectrum that was similar to the other methylalkyl glucosinolates except for the m/z shift of the first product ion resulting from sulphite loss (-80 Da), compound 22 was annotated as 5-methylhexyl glucosinolate. No hit with the same chemical formula was obtained when matching to MassBank. 23. 4-hydroxyglucobrassicin The ion of compound 23 had a chemical formula of C16H19O10N2S2 (m/z 463.04780) in which the additional nitrogen, as compared to the chemical formulae of the glucosinolates discussed above, should be present in the variable side-chain. Its characterization as a glucosinolate was indeed confirmed by the first product ions at m/z 259 and 275 that result from rearrangement reactions at the glucose thiohydroximate moiety. However, the most abundant first product ions were observed at m/z 285 and 267 due to losses of a thioglucose fragment and thioglucose (Fabre et 30

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

al., 2007), respectively. Subsequent losses of HCN and sulphite from the latter first product ion lead to the first product ions observed at m/z 240 and 160. This was confirmed in the MS3 spectrum of the m/z 267 first product ion. The latter gas phase fragmentation reactions are typical for indolic glucosinolates (Fabre et al., 2007) and compound 23 was annotated as 4hydroxyindol-3-ylmethyl glucosinolate or 4-hydroxyglucobrassicin. No hit with the same chemical formula was obtained when matching to MassBank. 24. glucobrassicin The chemical formula of this anion was C16H19O9N2S2 (m/z 447.05346) and its MS2 spectrum showed the same type of first product ions as the spectrum of 4-hydroxyglucobrassicin 23, yet the most abundant MS2 ions were observed at m/z 259 and 275 and also the other first product ions typical for the glucosinolate common moiety were observed, i.e., at m/z 195, 241 and 291. However, indicative for indolic glucosinolates were the ions at m/z 269 and 251 due to losses of a thioglucose fragment and thioglucose, and, from the latter ion, the further HCN and sulphite losses yielding the first product ions at m/z 244 and 144. Therefore, this compound was annotated as indol-3-ylmethyl glucosinolate or glucobrassicin. Matching to MassBank yielded a score and hit of 0.60 and 10, respectively. 25. hydroxy-methoxyglucobrassicin The ion of compound 25 had C17H21O11N2S2 (m/z 493.05834) as chemical formula and its MS2 spectrum showed the m/z 315 ion (loss of thioglucose fragment) as a very abundant peak, characteristic for indolic glucosinolates, yet the base peak was at m/z 259. This compound was annotated as hydroxy-methoxyglucobrassicin. No hit with the same chemical formula was obtained when matching to MassBank. 26. 4-methoxyglucobrassicin The chemical formula of the ion of compound 26 was C17H21O10N2S2 (m/z 477.06391). The most abundant MS2 ions were observed at m/z 259, 275 and 291, characteristic for the glucose thiohydroximate moiety. However, first product ions resulting from the losses of a thioglucose fragment and of thioglucose, typical for the indolic moiety were observed as well at m/z 299 and 281. The latter ion fragmented by HCN loss yielding the m/z 254 first product ion. This compound

31

was

annotated

as

4-methoxyindol-3-ylmethyl

glucosinolate

or

4-

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

methoxyglucobrassicin. Matching to MassBank rendered a score and hit of 0.87 and 8, respectively. 27. neoglucobrassicin The ion of this compound had the same chemical formula (C17H21O10N2S2, m/z 477.06398) as 4methoxyglucobrassicin 26. Its MS2 spectrum showed first product ions at m/z 259, reminiscent of the common moiety in glucosinolates, and at m/z 299 due to the loss of a thioglucose fragment typical for indolic glucosinolates. However, the base peak was observed at m/z 446 resulting from methoxyl radical loss. Such a homolytic fragmentation would be expected when the methoxyl function is linked to the indole nitrogen. Therefore, this compound is 1-methoxyindol3-ylmethyl glucosinolate or neoglucobrassicin. No hit with the same chemical formula was obtained when matching to MassBank. 28. 2-phenylethyl Gluc The anion with chemical formula of C15H20O9NS2 (m/z 422.05842) had a MS2 spectrum similar to that of the methylthioalkyl and the methylalkyl glucosinolates, i.e., in which the fragmentations typical for the glucose thiohydroximate moiety prevailed. Based on the high RDB value (ring and double bonds=6.5) as compared to that of other glucosinolates, this compound was annotated as 2-phenylethyl glucosinolate or gluconasturtiin. No hit with the same chemical formula was obtained when matching to MassBank. Flavonoids 29. 3-Glc(2←1)Rha-7-Rha-Que The ion of compound 29 had C33H39O20 (m/z 755.20427) as chemical formula. Its MS2 spectrum was dominated by the loss of 146 Da due to expelling a deoxyhexose residue yielding the ion at m/z 609. MS3 fragmentation of this first product ion lead to second product ions at m/z 463 and 447 indicating that a second deoxyhexose and a hexose were lost, respectively. The occurrence of both m/z 463 and 447 revealed that the deoxyhexose and the hexose were attached to two different sites on the aglycone. The aglycone was represented by second product ions at m/z 300 and 301. The former represents the aglycone radical anion resulting from homolytic cleavage of the O-glycosidic bond (Hvattum and Ekeberg, 2003). MS4 fragmentation of the aglycone yielded third product ions due to the loss of CO or both CO and CO2, indicative for a flavonol whereas the third product ions at m/z 151 and 179 arose from a Retro Diels-Alder (RDA) cleavage of the 32

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

C ring. Both ions are annotated as the

1,3

A- and

1,2

A- ions of quercetin in Fabre et al. (2001).

Other groups have also pinpointed the typical RDA cleavages occurring during gas phase fragmentations of flavonoids (Justesen, 2000; Hughes et al., 2001). Because all glycosides were linked to the 3- and/or 7-position, because diglycosides were more prominent on the 3-position than on the 7-position and because all deoxyhexoses and hexoses were rhamnose and glucose, respectively, in a thorough study of Arabidopsis flavonoids (Yonekura-Sakakibara et al., 2008), compound 29 was annotated as 3-rhamnosyl(1→2)glucoside-7-rhamnoside-quercetin. The rhamnose  is  1→2  linked  to  glucose  (neohesperidose)  rather  than  1→6  as  in  rutinose  because  the   cross-ring cleavage

0,2

X0- first and second product ions at m/z 489 are observed (Cuyckens and

Claeys, 2004). Furthermore, the interglycosidic bond in a neohesperidose moiety is much more fragile than that in a rutinose moiety, rendering information on the sugar sequence for the former (Ferreres et al., 2004; Yan et al., 2007). Matching to MassBank yielded a score and hit of 0.80 and 9. 30. 2-Glc-2-Rha-Kae The chemical formula of this anion was C39H49O24 (m/z 901.26343). The MS2 spectrum showed a loss of 308 Da as major fragmentation pathway leading to the ion at m/z 593. This is due to the loss of a disaccharide composed of a deoxyhexose, likely rhamnose, and a hexose, likely glucose. Because no first product ion due to  interglycosidic  cleavage  was  observed,  a  1→6  rather   than   a   1→2   glycosidic   linkage   might   be   inferred   (Cuyckens and Claeys, 2004; Ferreres et al., 2004; Yan et al., 2007). MS3 of the first product ion at m/z 593 yields then a second product ion at m/z 447 due to the loss of a second deoxyhexose. Subsequent dissociation of this ion eliminates another hexose and yields the second product ions at m/z 285 and 284, representing the aglycone. This aglycone eliminates CO yielding the ion at m/z 257, whereas the second product ion at m/z 179 can be annotated as the 1,2A- ion arising from RDA cleavage of a flavonol C ring (Fabre et al., 2001). Therefore, this compound was characterized as kaempherol to which two disaccharides, each comprising a rhamnosyl and a glucosyl unit, are linked. No hit with the same chemical formula was obtained when matching to MassBank. Using MetFrag, 20 biological compounds with the same chemical formula were downloaded from the Pubchem database of which 5 were predicted via in silico fragmentation to match equally likely the MS2 spectrum (containing only one first product ion at m/z 593, see above) of compound 30: two of those 5 compounds were kaempherol glycosides and two others were apigenin glycosides. 33

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

31. 3-Glc(2←1)Rha-7-Rha-Kae Anion 31 had C33H39O19 (m/z 739.20820) as chemical formula. The MS2 and MS3 spectra were similar to those of 3-Glc(2←1)Rha-7-Rha-Que 29 except that the product ion for the aglycone was detected 16 amu lower, i.e., at m/z 285. The CO and combined CO and CO2 losses observed in the MS4 spectrum were also at m/z values 16 amu lower than the corresponding losses in the MS4 spectrum of 3-Glc(2←1)Rha-7-Rha-Que 29. Therefore, this compound is 3rhamnosyl(1→2)glucoside-7-rhamnoside-kaempherol. A score and hit of 0.77 and 1 was obtained when matching to MassBank. 32. 3-Glc-7-Rha-Que The  chemical  formula  of  this  compound’s  anion  was  C27H29O16 (m/z 609.14553). The MS2 base peak at m/z 463 indicated a deoxyhexose (rhamnose) loss. However, also a hexose (glucose) loss was evident from the first product ion at m/z 447. Both sugars are attached to different positions of the aglycone. A first product ion representing a quercetin aglycone was observed at m/z 301. This was confirmed by MS3 fragmentation of the m/z 463 first product ion rendering both the aglycone anion and the aglycone radical anion at m/z 301 and 300, respectively (Hvattum and Ekeberg, 2003). Likely, the rhamnose and glucose moieties are linked to the 7-O- and 3-Opositions because glycosidic cleavage in negative ionization mode occurs presumably more readily at the 7-O-position since the reverse is true in positive ionization mode (Cuyckens and Claeys, 2004). This was indeed verified for some flavonol di-O-glycosides from Farsetia aegyptia (Shahat et al., 2005). Compound 32 was annotated as 3-glucosyl-7-rhamnosylquercetin. Matched to MassBank, a score and hit of 0.33 and 2 was obtained with a library spectrum recorded on a quadrupole-time-of-flight MS. 33. 3-Rha-7-Glc-Que The chemical formula (C27H29O16, m/z 609.14527) of the anion was the same as for 3-Glc-7Rha-Que 32. The MS2 spectrum was similar as well. The major difference was that hexose loss was more facile than rhamnose loss based on the abundance of the corresponding first product ions at m/z 447 and 463, respectively. Based on the same reasoning as for 3-Glc-7-Rha-Que 32, this compound was annotated as 3-rhamnosyl-7-glucosyl-quercetin. MassBank matching returned a score and hit of 0.72 and 7 with a library spectrum of 3-glucosyl-7-rhamnosidequercetin (see compound 32). However, 3-rhamnosyl-7-glucoside-quercetin was not present in the database. 34

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

34. 3-Glc(6←1)Glc-7-Rha-Kae The anion had C33H39O20 (m/z 755.20409) as chemical formula. The base peak in its MS2 spectrum was at m/z 609 due to rhamnose loss. A minor peak was observed at m/z 432 resulting from the elimination of a dihexoside. Thus, the rhamnose is likely attached to the 7-O-position, whereas the dihexoside is 3-O-linked (Cuyckens and Claeys, 2004; Shahat et al., 2005). MS3 fragmentation of the m/z 609 first product ion rendered a second product ion at m/z 285, representing the kaempherol aglycone. This was verified by MS4 fragmentation of the second product ion at m/z 285, which was characterized by CO loss (m/z 257), CO2 loss (m/z 241), combined CO/CO2 loss (m/z 213), 2CO loss (m/z 229), C2H2O loss (m/z 243) and by the RDAderived

1,3

A- ion at m/z 151 (Fabre et al., 2001; Hughes et al., 2001). As no interglycosidic

cleavage was observed in the MS3 spectrum,   the   dihexoside   is   1→6   linked   and   represents   a   gentiobiose moiety (Ferreres et al., 2004; Yan et al., 2007). Therefore, this compound was annotated as 3-glucosyl-(6←1)-glucosyl-7-rhamnosyl-kaempherol. No hit with the same chemical formula and the same aglycone was obtained when matching to MassBank. Using MetFrag, 110 biological compounds were retrieved with an equal chemical formula of which 22, via in silico fragmentation, equally likely matched with the MS2 spectrum (containing one first product ion) of compound 34: they comprised a diverse set of flavonol and flavone aglycone structures. 35. 3-Rha-7-Rha-Que The chemical formula for the anion was C27H29O15 (m/z 593.15133). The MS2 base peak appeared at m/z 447 indicating a rhamnose loss. Further MS3 fragmentation of the MS2 base peak lead to a second rhamnose loss yielding the peaks at m/z 300 and 301 corresponding to a quercetin radical anion and a quercetin anion, respectively. The aglycone structure was verified by MS4 fragmentation. This compound was annotated as 3-rhamnosyl-7-rhamnosyl-quercetin. A score and hit value of 0.58 and 2 were obtained when matching to MassBank. 36. 3-Ara(2←1)Rha-7-Rha-Kae This anion had a chemical formula of C32H37O18 (m/z 709.19849). The MS2 spectrum had a base peak at m/z 563 owing to rhamnose loss (-146 Da). Further MS3 fragmentation rendered a base peak at m/z 417 due to a second rhamnose loss. A further pentose loss (likely arabinose) in the MS3 spectrum explained the kaempherol anion peak at m/z 285. The MS3 ion at m/z 417 expelled a water molecule yielding the second product ion at m/z 399. Together with the 35

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

abundant second product ion representing the kaempherol radical anion at m/z 284, this suggested   a   rhamnose   1→2   linked   to   arabinose   (Hvattum and Ekeberg, 2003; Ferreres et al., 2004; Yan et al., 2007). The latter disaccharide moiety cleaved off leading to the low abundant radical anion at m/z 430 in the MS2 spectrum. Therefore, this compound was structurally elucidated as 3-arabinosyl-(2←1)-rhamnosyl-7-rhamnosyl-kaempherol. No hit with the same chemical formula was obtained when matching to MassBank. Using MetFrag, 18 compounds with the same chemical formula were retrieved from the Pubchem database of which 5, via in silico fragmentation, equally likely matched the MS2 (containing one peak) spectrum of compound 36: 3 kaempherol and 2 apigenin glycosides. Among the kaempherol glycosides, 3arabinosyl-(2←1)-rhamnosyl-7-rhamnosyl-kaempherol was present. 37. 3-Glc-7-Rha-Kae As chemical formula C27H29O15 (m/z 593.15055) was obtained for this anion. The MS2 spectrum was characterized by glucose and rhamnose loss leading to the first product ions at m/z 431 and 447, the latter ion being the most abundant (Cuyckens and Claeys, 2004; Shahat et al., 2005). The spectrum showed also the aglycone peak at m/z 285. MS3 fragmentation of the m/z 447 ion rendered second product ions typical for kaempherol (see description of MSn spectra from compound 34). This compound was annotated as 3-glucosyl-7-rhamnosyl-kaempherol. The best fit when matching to MassBank was obtained for 3-rhamnosyl-7-rhamnosyl-quercetin (score=0.74, hit=4). However, the second best fit was for 3-glucosyl-7-rhamnosyl-kaempherol (score=0.48, hit=5). 38. 3-Rha-7-Ara-Kae This anion had C26H27O14 (m/z 563.14006) as chemical formula. Its MS2 spectrum was dominated by the base peak at m/z 431 and the less abundant ion at m/z 417 due to pentose (arabinose) and rhamnose losses. The aglycone anion at m/z 285 was reminiscent of a kaempherol. Based on Cuyckens and Claeys (2004) and on Shahat et al. (2005), this compound was annotated as 3-rhamnosyl-7-arabinosyl-kaempherol. No hit with the same chemical formula was obtained when matching to MassBank. Entering the MS2 data in MetFrag returned 137 biological compounds with the same chemical formula as compound 38 from the Pubchem database. Following in silico fragmentation, the best hit was obtained for the flavone 7xylosyl(1→4)rhamnosyl-scutellarein (score=1, # explained peaks=4). Among the next 4 best hits (score=0.985, # explained peaks=3) were the flavonols 3-arabinosyl-7-rhamnosyl-kaempherol, 336

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

rhamnosyl-7-arabinosyl-kaempherol, 3-rhamnosyl-4’-arabinosyl-kaempherol and the flavone 6xylosyl-7-rhamnosyl-scutellarein. 39. 3-Rha-7-Mal-Glc-Kae The chemical formula for this anion was C29H31O16 (m/z 635.16151). Upon MS2 fragmentation, the loss of rhamnose and malonylglucose yield the first product ions at m/z 489 and 431. The first product ion at m/z 285 suggested a kaempherol structure as aglycone. As the m/z 431 ion was the most abundant (Cuyckens and Claeys, 2004; Shahat et al., 2005), this compound was annotated as 3-rhamnosyl-7-malonylglucosyl-kaempherol. No hit with the same chemical formula was obtained when matching to MassBank. No biological compounds with this chemical formula could be retrieved from either the Pubchem or the ChemSpider databases using MetFrag. 40. 3-Rha-7-Rha(4←1)Glc-Kae The anion had a chemical formula of C33H39O19 (m/z 739.20920). The MS2 base peak was observed at m/z 431 due to the combined loss of glucose and rhamnose. Although a very minor peak at m/z 577 was observed resulting from glucose loss, the major loss as a disaccharide moiety  indicates  that  no  1→2  linkage  was  involved  (Ferreres et al., 2004; Yan et al., 2007) and that this disaccharide moiety was present at the 7-position (Cuyckens and Claeys, 2004; Shahat et al., 2005). A rhamnose loss from the 3-position renders the MS2 ion at m/z 593. The aglycone was represented by the first product ion at m/z 285. In the MS3 spectrum of the first product ion at m/z 431, the flavonoid-specific RDA cleavage

1,3

A- and

1,2

A- ions at m/z 151 and 179 were

observed (Fabre et al., 2001; Hughes et al., 2001). Therefore, the aglycone was annotated as a kaempherol and the compound as 3-rhamnosyl-7-rhamnosyl-(4←1)-glucosyl-kaempherol. Matching

to

MassBank

retrieved

3-rhamnosyl-(2←1)-glucosyl-7-rhamnosyl-kaempherol

(score=0.77, hit=2). Our proposed structure was not present in this database. 41. 3-Rha(4←1)Rha-7-Rha-Kae The chemical formula for this anion was C33H39O18 (m/z 723.21490). No MS2 spectrum was obtained for this low abundant compound, but this compound was connected to 3-Ara(2←1)Rha7-Rha-Kae 36 via  a  “methylation”  conversion  in the CSPP networks. Furthermore, the levels of both compounds were highly correlated. The possible candidate structure, i.e., 3-rhamnosyl-

37

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

(4←1)-rhamnosyl-7-rhamnosyl-kaempherol, has been previously observed in the plant kingdom (Rasoanaivo et al., 1990). 42. 3-Rha-7-Rha-Kae The   anion’s   chemical   formula   was   C27H29O14 (m/z 577.15646). MS2 fragmentation yielded a base peak at m/z 431 (rhamnose loss) and a minor peak at m/z 285 representing the aglycone following two rhamnose losses. The MS3 spectrum of the m/z 431 ion yielded ions at m/z 151 and 179 corresponding to the flavonoid RDA cleavage

1,3

A- and

1,2

A- ions (Justesen, 2000;

Yonekura-Sakakibara et al., 2008), thus, pinpointing to a kaempherol aglycone. Both rhamnoses, connected in a disaccharide moiety via   a   2←1   linkage,   could   have   been   attached   to   the   3-Oposition or the rhamnoses were separately attached at two different positions on the kaempherol moiety. As this compound was one of the three most abundant, already identified (Veit and Pauli, 1999), flavonols in leaves, this compound was annotated as 3-rhamnosyl-7-rhamnosylkaempherol or kaempferitrin. Matching to MassBank returned a score and hit of 0.52 and 2. 43. 3-sinapoyl-Rha-7-Rha-Kae The chemical formula of the anion was C38H39O18 (m/z 783.21356). This compound was connected to the previous compound, i.e., 3-Rha-7-Rha-Kae 42,  via  a  “sinapoylation”  conversion   in the CSPP networks. Rhamnose elimination was the main fragmentation pathway rendering the MS2 base peak at m/z 637. Smaller peaks at m/z 577 and 431 were also visible, the first derived from cleavage of the sinapoyl ester bond, the second due to the loss of a second rhamnose moiety. This was further verified by MS3 fragmentation of the m/z 637 ion. In both the MS2 and MS3 spectra, the ion at m/z 285 indicated that kaempherol was the aglycone part. Therefore, this compound was identified as 3-sinapoyl-rhamnosyl-7-rhamnosyl-kaempherol. No hit with the same chemical formula was obtained when matching to MassBank. Eight hits were returned from the Pubchem database using MetFrag. In silico fragmentation provided non-zero scores for only three of them; all were flavone glycosides bearing a hydroxycinnamic acid further supporting our proposed structure for compound 43. Opposite to the observation for glucosinolate anions, MetFrag predicted reasonable structures that were all close to the true structure for all the flavonol glycoside anions. This seemed to be mainly due to the accurate prediction of interglycosidic cleavages and the use of the parent ion molecular weight for searching the Pubchem database.

38

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Benzenoids 44. protocatechoyl Glc The chemical formula for the ion was C13H15O9 (m/z 315.07229). The MS2 base peak at m/z 153 resulted from a hexose loss (-162 Da) and the MS3 spectrum showed that decarboxylation, yielding the ion at m/z 109, was the main fragmentation pathway for the m/z 153 ion. MS2 ions at m/z 225 and 165, resulting from hexose cross-ring cleavages, indicated that the hexose was connected as a hexoside with a free reducing end (Carroll et al., 1995; Mulroney et al., 1999; March and Stadey, 2005) or via an ester bond (Vanholme et al., 2012). The aglycone is a dihydroxybenzoic acid. In Arabidopsis, both 3,4-dihydroxybenzoic acid (protocatechuic acid) and 2,4-dihydroxybenzoic acid (homogentisic acid) occur. However, in the case of a homogentisic acid moiety, the MS3 spectrum of this moiety should show a base peak due to water loss (Supplemental Table 1 and Supplemental Methods), which is not observed here. Therefore, protocatechoyl glucose was proposed as structure for this molecule. Further support for the ester bond was provided by searching the complementary pairs of ions associated with the two characteristic cleavages that esters undergo (Debrauwer et al., 1992; Fournier et al., 1993; Fournier et al., 1995; Stroobant et al., 1995). The first charge-remote cleavage type produces a carboxylate anion and a neutral which remain initially together in an ion-dipole complex. The complex can then dissociate or might be preceded by a proton transfer between the carboxylate anion and the neutral, hence, leading to the loss of a neutral carboxylic acid. From the resulting complementary pair of ions, only the carboxylate ion is observed at m/z 153. A second cleavage type, which occurs to a lesser extent, produces a neutral ketene and an alkoxide anion that remain together in an ion-dipole complex. Again the complex can dissociate as such or might be preceded by a proton transfer yielding a ynolate ion and the neutral alcohol. This complementary pair of ions was observed at m/z 135 (ynolate ion) and 179 (alkoxide ion). No hit with the same chemical formula was obtained when matching to MassBank. However, when using MetFrag, 23 biological compounds were retrieved from the Pubchem database and, following in silico fragmentation, the best hit returned our proposed structure for compound 44. 45. protocatechoyl Xyl The  ion’s  chemical  formula  was  C12H13O8 (m/z 285.06195). The MS2 spectrum was very similar to that of protocatechoyl Glc, but the MS2 base peak (m/z 153) was due to a pentose loss instead of a hexose loss. This compound was characterized as protocatechoyl xylose. No hit with the 39

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

same chemical formula was obtained when matching to MassBank. Using MetFrag, Pubchem returned 47 biological compounds with the same chemical formula. After in silico fragmentation, the best hit returned our proposed structure. 46. p-hydroxybenzoic acid hex This ion had C13H15O8 (m/z 299.07748) as chemical formula. The MS2 spectrum showed a base peak at m/z 137 resulting from hexose loss and MS3-induced decarboxylation yielded the peak at m/z 93. Fortunately, decarboxylation of the aglycone could be pinpointed in the MS3 spectrum using a linear ion trap despite that this fragmentation pathway did not lead to a stable peak in the MS2 spectrum of the corresponding aglycone standard using a 3D ion trap (Supplemental Methods, Supplemental Table 1). This compound was annotated as p-hydroxybenzoic acid hexoside because no ester bond was evident based on its two types of characteristic cleavages and the hexose cross-ring cleavages (see protocatechoyl Glc 44). No hit with the same chemical formula was obtained when matching to MassBank. Using MetFrag, 84 biological compounds were retrieved from Pubchem. Following in silico fragmentation, the three best hits (score=1, # explained peaks=4/5) were the o-, m- and p-hydroxybenzoic acid hexosides. Phenylpropanoid derivates 47. 5-hydroxyferuloyl Glc The   ion’s   chemical   formula   was   C16H19O10 (m/z 371.09825). Elimination of a hexose moiety lead to the base peak at m/z 209. First product ions at m/z 251, 281 and 311 due to hexose crossring cleavages indicated that the hexose was linked in an ester bond (Vanholme et al., 2012) or that it was a hexoside with a free reducing end (Carroll et al., 1995). An ester bond is confirmed by its characteristic second type of cleavage in which an ynolate ion was formed at m/z 191. Further dissociation yielded the first product ion at m/z 176 via methyl radical loss. MS 3 fragmentation of the base peak at m/z 209 showed that methyl radical loss was the major fragmentation pathway (ion at m/z 194), but also decarboxylation (m/z 165) and a combined methyl radical loss / decarboxylation (m/z 150) were observed. As these losses are typical for hydroxycinnamic acids, e.g. MS2 spectrum of 4-hydroxy-3-methoxy-cinnamic (ferulic) and 4hydroxy-3,5-dimethoxy cinnamic (sinapic) acid in Supplemental Table 2 (see also Supplemental Methods), and because decarboxylation only occurred following hexose loss in the MS2 spectrum, this compound was annotated as 5-hydroxyferuloyl glucose. No hit with the same 40

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

chemical formula was obtained when matching to MassBank. From the Pubchem, 20 biological compounds having the same chemical formula were retrieved using MetFrag. None of them was a hydroxycinnamic acid. 48. sinapoyl malate hex This ion had C21H25O14 (m/z 501.12423) as chemical formula and was connected via a “hexosylation”  conversion  to  trans-sinapoyl malate 58. The levels of both compounds were also highly correlated across biological replicates. Its MS2 spectrum showed first product ions resulting from either hexose (m/z 339) or malate (m/z 385) loss. The loss of both moieties lead to the first product ion at m/z 223 representing a sinapate ion. Therefore, this compound was characterized as sinapoyl malate hexoside. 49. 5-hydroxyferuloyl malate The   ion’s   chemical   formula   was   C14H13O9 (m/z 325.05674). Although no MS2 spectrum was obtained for this compound, a MS2 spectrum was recorded for an in-source fragment resulting from the loss of 116 Da which corresponds with a malate moiety. The MS2 spectrum of the insource fragment was identical to the MS3 spectrum of 5-hydroxyferuloyl Glc 47. Therefore, this compound was elucidated as 5-hydroxyferuloyl malate. No hit with the same chemical formula was obtained when matching to MassBank. Using MetFrag, the MS2 spectrum of the in-sourceproduced 5-hydroxyferulic acid was analyzed. From the Pubchem database, 430 biological compounds having the same chemical formula were retrieved. However, the in silico obtained spectrum for 5-hydroxyferulic acid was not even included in the 150 best hits. In negative ionization mode, MetFrag rendered good results for smaller phenolics, i.e., the benzenoids, but was not efficient anymore for explaining the, more complicated, gas phase fragmentation behavior of larger phenolics such as the phenylpropanoids. 50. sinapoyl gentiobiose The chemical formula of this ion was C23H31O15 (m/z 547.16645). Although no MS2 spectrum was obtained for this compound, the CSPP network showed an association with cis-sinapoyl Glc 55 via   a   “hexosylation”   reaction.   Furthermore,   levels   of   both   compounds   were   very   highly   correlated across biological replicates. Therefore, this compound was characterized as a sinapoyl diglycoside. Because one or more sinapic acid moieties esterified to gentiobiose [glucosyl-

41

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

(1→6)-glucose] has been observed in other Brassicaceae (Supplemental Data Set 1), this compound was more narrowly defined as sinapoyl gentiobiose. 51. sinapoyl malate hex This ion had the same chemical formula as sinapoyl malate hex 48 and was also via a “hexosylation”   conversion   linked   to   trans-sinapoyl malate 58 in the CSPP networks. Additionally, the levels of the latter compound were highly correlated with those of compound 51 across biological replicates. Its MS2 spectrum was dominated by malate loss (m/z 385) and MS3 fragmentation of the m/z 385 first product ion yielded a sinapate ion due to hexose loss. Therefore, this compound is a sinapoyl malate hexoside isomer. 52. feruloyl glycerol The ion had C13H15O6 (m/z 267.08784) as chemical formula. The MS2 spectrum showed a base peak at m/z 149 and smaller peaks at m/z 134, 178 and 193. The first three peaks likely arise from the m/z 193 ion via decarboxylation, a combined decarboxylation and methyl radical loss, and a methyl radical loss, respectively. These losses are typically for the hydroxycinnamic acids and indicate that a ferulic acid moiety is present (see Supplemental Methods and Supplemental Table 2). The 74 Da loss leading to the first product ion at m/z 193 corresponds with an esterified glycerol moiety. Indeed, the second ester-specific ketene-producing cleavage yielded the first product ynolate ion at m/z 175. A rearrangement followed by a decarboxylation rendered the peak at m/z 223, whereas the peak at m/z 192 resulted from homolytic cleavage of the ester bond (Bowie, 1990). Therefore, this compound was characterized as feruloyl glycerol. No hit with the same chemical formula was obtained when matching to MassBank. 53. trans-sinapoyl Glc The  ion’s  chemical  formula  was  C17H21O10 (m/z 385.11406). The MS2 spectrum showed a base peak at m/z 223 due to hexose loss (-162 Da) which, upon MS3 fragmentation, yielded second product ions at m/z 208, 179 and 164 due to methyl radical loss, decarboxylation and a combined methyl radical loss and decarboxylation. Thus, this MS3 spectrum shows the typical collisioninduced dissociation fingerprint of sinapic acid (Supplemental Methods, Supplemental Table 2). The observation of first product ions at m/z 325, 295 and 265 arise from hexose cross-ring cleavages and occur whenever the hexose is connected as a hexoside with a free reducing end (Carroll et al., 1995) or when the hexose is linked as an ester (Vanholme et al., 2012). Since 42

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

decarboxylation only occurred after hexose loss and the second ester-characteristic cleavage produced the MS2 ynolate ion at m/z 205, this compound was characterized as sinapoyl glucose. Matching to MassBank rendered a score and hit of 0.51 and 5. 54. sinapoyl gentiobiose The ion had the same chemical formula as sinapoyl gentiobiose 50. It was connected to sinapoyl malate hexoside 48 via  a  “malate-hexoside  transesterification”  conversion  in  the  CSPP  networks   and the levels of the CSPP substrate and product peaks were highly correlated across biological replicates. Its MS2 spectrum was dominated by the first product ion at m/z 223 representing a sinapate ion. Other major first product ions were observed at m/z 385 (-162 Da, dehydrated hexose loss), m/z 367 (-180 Da, hexose loss), m/z 349 (-198 Da, combined loss of hexose and water), m/z 323 (-224 Da, sinapic acid loss), m/z 289 (-258 Da, combined loss of hexose, water and two molecules of formaldehyde). The latter first product ion arises from hexose cross-ring cleavages. Also the less abundant first product ions at m/z 325 and 295 arise from hexose crossring cleavage occurring on the m/z 385 first product ion. As already mentioned, cross-ring cleavages can occur when the hexose is linked as an ester (Vanholme et al., 2012). Consequently, the MS2 data suggest a disaccharide ester-linked to sinapic acid. For the same reasoning as described for compound 50, compound 54 is defined as a sinapoyl gentiobiose isomer. 55. cis-sinapoyl Glc The same chemical formula and similar MSn spectra were obtained for this compound as for trans-sinapoyl Glc 53. Because this compound eluted later and was much less abundant, it was characterized as the cis isomer of sinapoyl Glc, i.e., cis-sinapoyl glucose. The MassBank score and hit value were 0.48 and 5. 56. sinapoyl pen The chemical formula of this ion was C16H19O9 (m/z 355.10357). No MS2 spectrum was obtained,  but  the  compound  was  a  CSPP  substrate  for  a  “dihydroxybenzoylation”  conversion  to   the CSPP product dihydroxybenzoyl sinapoyl pen 62. Also the levels of both compounds were highly correlated across biological replicates. Thus, this compound was annotated as sinapoyl pentose. 57. disinapoyl butanoyl gentiobiose 43

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

The ion had C38H47O20 (m/z 823.26695) as chemical formula. No MS2 spectrum was recorded, but the compound was connected with trisinapoyl gentiobiose 66 via  a  “dihydroxybenzoylation”   in the CSPP networks. Both compound levels were highly correlated across biological replicates, strongly suggesting that compound 57 was a sinapoyl gentiobiose derivate as well. A putative candidate is disinapoyl butanoyl gentiobiose. 58. trans-sinapoyl malate The chemical formula of this ion was C15H15O9 (m/z 339.07220). The MS2 spectrum showed a base peak at m/z 223 due to malate loss (-116 Da) and MS3 fragmentation of this first product ion rendered second product ions at m/z 208, 179 and 164 which are characteristic for sinapic acid (Supplemental Table 2). The second ketene-producing cleavage was not observed for this ester, because the ester cleavage reaction leading to the sinapate anion is too favorable. Besides classical charge-remote ester cleavage (Stroobant et al., 1995), the neighboring carboxylic acid functions on the malate moiety facilitate proton transfer to the ester, enhancing a cleavage reaction  similar  to  a  β-keto acid decarboxylation in solution chemistry. Therefore, this compound is sinapoyl malate. Matching to MassBank returned a score and hit of 0.71 and 1. 59. cis-sinapoyl malate The chemical formula and MSn spectra were the same as for trans-sinapoyl malate 58 and, because compound 59 had lower levels and was eluting later, it was characterized as cis-sinapoyl malate. Matching to MassBank returned a score and hit of 0.71 and 1. 60. disinapoyl gentiobiose The  ion’s  chemical  formula  was C34H41O19 (m/z  753.22528).  It’s  MS2 spectrum was dominated by a peak at m/z 591 resulting from hexose loss (-162 Da). No neutral losses of -60, -90 and/or 120 Da were observed, indicating that this hexose was connected via its reducing end in a glycosidic rather than an ester bond (Carroll et al., 1995; March and Stadey, 2005; Vanholme et al., 2012). This ion dissociated further by expelling a sinapic acid residue leading to the second product ion at m/z 367. A second hexose loss (-144 Da) lead to the second product ion at m/z 223; the latter ion being reminiscent of a second sinapic acid moiety in the molecule. Second product ions at m/z 307 and 277 (-60 and -90 Da losses) indicated that this hexose was linked via an ester bond. This was further confirmed by the second product ion at m/z 349 representing the ynolate ion due to the second ketene-producing cleavage typical for esters. Consequently, this 44

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

compound was characterized as disinapoyl gentiobiose. Disinapoyl gentiobiose has been observed in other Brassicaceae in which both sinapoyl moieties were connected to the 1-O- and 2-O positions of gentiobiose (Supplemental Data Set 1). No hit with the same chemical formula was obtained when matching to MassBank. 61. disinapoyl gentiobiose This compound had the same chemical formula and similar MSn spectra as disinapoyl gentiobiose 60 and is, consequently, an isomer. This compound was characterized as disinapoyl gentiobiose. No hit with the same chemical formula was obtained when matching to MassBank. 62. dihydroxybenzoyl sinapoyl pen The chemical formula of this ion was C23H23O12 (m/z 491.11868). The MS2 spectrum showed two sets of complementary peaks, i.e., at m/z 337 and m/z 153 and at m/z 267 and m/z 223, suggesting that there were two ester bonds involved. Upon CID, ester bonds are subjected to a charge-remote cleavage producing a carboxylate anion and a neutral that remain in an ion-neutral complex (Debrauwer et al., 1992; Stroobant et al., 1995). Following dissociation of the ionneutral complex, the carboxylate ions provide the peaks at m/z 153 and m/z 223 representing dihydroxybenzoate and sinapate. Alternatively, before complex dissociation, the carboxylate ion can abstract a proton from the neutral of which the ion is visible by peaks at m/z 337 and m/z 267 representing the sinapoyl pentose and the dihydroxybenzoyl pentose moiety, respectively. Therefore, this compound was presumably dihydroxybenzoyl sinapoyl pentose. Further support was derived from the remaining MS2 first product ions. The second ester-characteristic keteneproducing cleavage is observed for the sinapoyl moiety rendering ions at m/z 285 and 205 representing the alkoxide and ynolate ions, respectively. Although this second cleavage type is not observed for the dihydroxybenzoyl moiety, the latter can initiate a rearrangement converting dihydroxybenzoyl sinapoyl pentose into (iso)vanilloyl 5-hydroxyferuloyl pentose. When the dihydroxybenzoyl moiety is deprotonated, the resulting phenoxide ion is stabilized by the orthohydroxy group. The phenoxide anion can then attack one of the methoxy groups on the sinapoyl system converting it into a 5-hydroxyferuloyl system in a similar SN2 reaction as observed for the methanol loss upon CID of sinapyl alcohol (3,5-dihydroxy-4-methoxycinnamyl alcohol; see Supplemental Methods, Supplemental Table 2 and Supplemental Figure 8A). This rearrangement is followed by charge-remote ester cleavages, providing the base peak at m/z 323 (vanillic acid loss, -168 Da), the ion at m/z 209 (5-hydroxyferulate ion), or the ketene-producing cleavage 45

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

providing the sinapic acid-derived ynolate ion at m/z 191. The latter first product ion can lose a methyl radical yielding the first product ion at m/z 176. The rearrangement can only occur when the dihydroxybenzoyl and sinapoyl moieties are close together, e.g., connected to the 1-O- and 2O-positions of the pentose ring. The first product ion at m/z 233 likely resulted from pentose cross-ring cleavage of the m/z 323 ion which represented the sinapoyl pentose moiety. Noticeable, assuming the rearrangement occurs in the reverse direction, i.e., producing dihydroxybenzoyl sinapoyl pentose from (iso)vanilloyl 5-hydroxyferuloyl pentose, all MSn peaks could be explained in exactly the same way. Therefore, it cannot be excluded that this compound is (iso)vanilloyl 5-hydroxyferuloyl pentose. When matching to MassBank, no hit with the same chemical formula as that of dihydroxybenzoyl sinapoyl pentose was obtained. 63. disinapoyl hexanetriol dihex The  ion’s  chemical  formula  was  C40H53O21 (m/z 869.31077). This compound was connected to disinapoyl butanoyl gentiobiose 57 via   a   “malate-hexose   transesterification”   conversion   in   the   CSPP networks. However, this CSPP connection is based on the mass difference and does not necessarily visualize a true biochemical transesterification. Nonetheless, as the levels of both compounds were highly correlated across biological replicates, they are likely structurally similar. In the full MS, the detection of a peak at m/z 753 might originate from the in-source loss of hexanetriol (-116 Da). Therefore, this compound was annotated as disinapoyl hexanetriol dihexos-e/-ide. 64. disinapoyl Glc The chemical formula of this ion was C28H31O14 (m/z 591.17154). Its MS2 spectrum was identical to the MS3 spectrum of the first product ion at m/z 591 of disinapoyl gentiobiose 60. Therefore, this compound was characterized as disinapoyl glucose. When matching to MassBank, no hit with the same chemical formula as compound 64 was obtained. 65. disinapoyl Glc The chemical formula and MS2 spectrum were identical to that of disinapoyl Glc 64. Therefore compound 65 was characterized as a disinapoyl glucose isomer. When matching to MassBank, no hit with the same chemical formula as compound 65 was obtained. 66. trisinapoyl gentiobiose

46

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

This ion had C45H51O23 (m/z 959.28407) as chemical formula. Its MS2 spectrum showed a base peak at m/z 797 due to hexose loss (-162 Da). Further MS3 fragmentation of this first product ion eliminated a sinapic acid-derived unit, rendering the peaks at m/z 591 and 573 corresponding with the two cleavages typical for esters (Debrauwer et al., 1992; Fournier et al., 1993; Fournier et al., 1995). The first product ions at m/z 501 and 349 are due to hexose cross-ring cleavage and the loss of a second sinapic acid-derived unit. This compound was characterized as trisinapoyl gentiobiose. When matching to MassBank, no hit with the same chemical formula as compound 66 was obtained. 67. disinapoyl Glc The chemical formula and MS2 spectrum were identical to that of disinapoyl Glc 64. Therefore compound 67 was characterized as a disinapoyl glucose isomer. When matching to MassBank, no hit with the same chemical formula as compound 67 was obtained. 68. trisinapoyl gentiobiose The chemical formula was the same as obtained for trisinapoyl gentiobiose 66. No MS2 spectrum was obtained, but it was connected to disinapoyl gentiobiose 60 via   a   “sinapic   acid   derivatization”   conversion   in   the   CSPP   networks.   Levels   of   both   compounds   were   also correlated across biological replicates. Therefore, this compound was annotated as a trisinapoyl gentiobiose isomer. (Neo)Lignans/Oligolignols 69. hex G(8–O–4)FA malate The   ion’s   chemical   formula   was   C30H35O17 (m/z 667.18775). No MS2 spectrum was obtained, but it was connected to G(8–O–4)FA malate 86 via   a   “hexosylation”   conversion   in   the   CSPP   networks. Levels of both compounds were highly correlated across biological replicates. Therefore, this compound was characterized as the hexoside of guaiacylglycerol 8–O–4 feruloyl malate ether. 70. G(8–O–4)FA hex The  ion’s  chemical  formula  was  C26H31O13 (m/z 551.17649). Upon CID, hexose loss (m/z 389, 162 Da) was the dominating fragmentation pathway. The MS3 spectrum of this base peak showed two small neutral losses, i.e., water loss (-18 Da) and a combined water / formaldehyde 47

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

loss (-48 Da), leading to the ion at m/z 371 and 341. These are characteristic for 1,4-propanediol or

1,3-propanediol   moieties   and,   thus,   for   a   dibenzylbutanediol   or   β-aryl

ether

(neo)lignan/oligolignol (Eklund et al., 2008; Morreel et al., 2010a).  In  the  case  of  a  β-aryl ether, cleavage of the 8–O–4-linkage provides information on the composing units (Morreel et al., 2010a; Morreel et al., 2010b). The ions at m/z 195 and 193 represent a guaiacyl unit connected to a unit derived from ferulic acid. The former ion fragments further to m/z 165 due to formaldehyde loss, whereas the latter ion shows the typical fragmentation pattern (m/z 178, 149 and 134) of ferulic acid (Supplemental Table 2). Therefore, this compound was characterized as guaiacylglycerol 8–O–4 ferulic acid ether hexoside. The shorthand name for the aglycone, G(8– O–4)FA, is based on Morreel et al. (2004) and is explained in the legend of Supplemental Data Set 1. When matching to MassBank, no hit with the same chemical formula as compound 70 was obtained, but the CID spectrum has been described previously (Vanholme et al., 2012). Using MetFrag, 51 biological compounds having the same chemical formula were retrieved from the Pubchem  database.  None  of  them  was  a  β-aryl ether. 71. lariciresinol dihex This ion had C32H43O16 (m/z 683.25571) as chemical formula. Dissociation of a hexose lead to the MS2 peak at m/z 521. Further MS3 fragmentation yielded a peak at m/z 359 indicating a second hexose loss. The second product ion at m/z 329 indicated a further formaldehyde loss and is typical for lariciresinol (Morreel et al., 2010a), i.e., pinoresinol or G(8–8)G in which one of the tetrahydrofuran rings is reduced. Its structural characterization was confirmed by its connection to pinoresinol dihex 79 via  a  “reduction”  conversion.  The  levels  of  both  compounds  were  very   highly correlated across biological replicates. When matching to MassBank, no hit with the same chemical formula as compound 71 was obtained. Using MetFrag, 51 biological compounds having the same chemical formula were retrieved from the Pubchem database. None of them was a resinol-derived compound. Because lariciresinol hexoside, represented by the first product ion at m/z 521, was expected to be present in the Pubchem database, the MS3 spectrum was entered into MetFrag. Both second product ions from lariciresinol hexoside were recognized leading to a perfect match. 72. G(8–O–4)G hex The  ion’s  chemical  formula  was  C28H37O14 (m/z 597.21896). Its MS2 spectrum showed the loss of an acetate adduct (-60 Da) rendering the peak at m/z 537. Another peak at m/z 375 indicated 48

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

that this compound was hexosylated. Furthermore, a first product ion at m/z 327 was found due to the combined loss of water and formaldehyde (-48 Da) from the aglycone (m/z 375). This is characteristic   for   β-aryl ethers (Morreel et al., 2010a). Therefore, this compound is guaiacylglycerol 8–O–4 coniferyl ether hexoside. When matching to MassBank, no hit with the same chemical formula as compound 72 was obtained. Using MetFrag, 57 biological compounds having the same chemical formula were retrieved from the Pubchem database. Two hexosides isomers of G(t8–O–4)G hexoside were present among the 10 best hits. The MS 3 spectrum of the first product ion at m/z 375 represented the fragmentation of the aglycone. When the accurate mass of this aglycone was entered into MetFrag, 388 hits with the Pubchem database were returned.   However,   β-aryl ethers were not among the top 50 hits as only three of the four fragment ions (m/z 327, m/z 195 and m/z 179, but not m/z 165) were recognized. 73. hex G(8–O–4)FA malate The  ion’s  chemical  formula (C30H35O17; m/z 667.18877) was the same as for hex G(8–O–4)FA malate 69. Its MS2 spectrum showed a peak at m/z 551 due to malate loss. Following further MS3 dissociation of the first product ion at m/z 551, a peak appeared at m/z 389 due to hexose loss and also at m/z 341 due to the combined loss of water and formaldehyde from the aglycone (m/z 389) which is typical   for   β-aryl ethers. Therefore, this compound is an isomer of guaiacylglycerol 8–O–4 feruloyl malate ether hexoside. When matching to MassBank, no hit with the same chemical formula as compound 73 was obtained. No MetFrag search was performed because the aglycone is not present in the Pubchem database. 74. G(t8–O–4)G hex The ion had C26H33O12 (m/z 537.19774) as chemical formula. Its MS2 spectrum was dominated by a peak at m/z 375 due to hexose loss. The latter ion was further fragmented to the ions at m/z 327, 195 and 179 in the MS3 spectrum. These are due to a combined water/formaldehyde loss, characteristic for the 8–O–4-linkage in (neo)lignans/oligolignols and cleavage of this linkage resulting in second product ions representing each of the units in this dimer (Morreel et al., 2010a). The fragmentations in this MS3 spectrum are well-documented and the aglycone-based MS2 spectrum has been published before (Morreel et al., 2010a). This compound is the threo isomer of guaiacylglycerol 8–O–4 coniferyl ether hexoside. When matching to MassBank, no hit with the same chemical formula as compound 74 was obtained. MetFrag-based structural elucidation rendered the same result as for G(t8–O–4)G hex 72. 49

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

75. G(8–O–4)G(8–O–4)G hex This ion had C36H45O16 (m/z 733.27194) as chemical formula. No MS2 spectrum was obtained, but the compound was connected to G(t8–O–4)G hex 74 via   a   “G   unit   addition”   conversion.   Furthermore, both compound levels were very highly correlated across biological replicates. Therefore, this compound was annotated as guaiacylglycerol 8–O–4 guaiacylglycerol ether 8–O– 4 coniferyl ether hexoside. 76. S(8–O–4)FA hex This ion had C27H33O14 (m/z 581.18709) as chemical formula. Upon CID, a base peak at m/z 419 was observed due to hexose loss. MS3 fragmentation yielded the type I small neutral losses at m/z 401 (water loss) and  371  (combined  water/formaldehyde  loss)  that  are  characteristic  for  βaryl ethers (Morreel et al., 2010a). The type II cleavage of the linkage reveals the type of units involved (Morreel et al., 2010a): peaks at m/z 225 and 193 indicate a syringyl unit coupled to a unit derived from ferulic acid. Therefore, this compound is syringylglycerol 8–O–4 ferulic acid ether hexoside. When matching to MassBank, no hit with the same chemical formula as compound 76 was obtained. The aglycone was not present in the Pubchem database, rendering MetFrag-based structural verification redundant. 77. G(8–O–4)FA hex The ion had the same chemical formula (C26H31O13, m/z 551.17679) and very similar MSn spectra as G(8–O–4)FA hex 70. The additional MS3 peak at m/z 195 confirmed that this compound contains a guaiacyl unit and is an isomer of guaiacylglycerol 8–O–4 ferulic acid ether hexoside. When matching to MassBank, no hit with the same chemical formula as compound 77 was obtained. As the aglycone is not present in the Pubchem database, no MetFrag-based annotation was performed. 78. G(8–O–4)FA Glu The chemical formula of this ion was C25H28O11N (m/z 518.16698). No MS2 spectrum was recorded, but the compound was connected to G(8–5)FA Glu 112 via  a  “hydration”  conversion.   The levels of both compounds were very highly correlated across biological replicates. This compound was annotated as guaiacylglycerol 8–O–4 feruloyl glutamic acid ether. As the aglycone is not present in the Pubchem database, no MetFrag-based annotation was performed. However, as glutamate derivates of (neo)lignans/oligolignols were not expected, the standard 50

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

was synthesized allowing the authentication of these compound structures. Therefore, this proved that (neo)lignans/oligolignols can be derivatized with glutamate. 79. pinoresinol dihex The   ion’s   chemical   formula   was   C32H41O16 (m/z 681.24042). Upon CID, a hexose loss was observed leading to the ion at m/z 519. MS3 fragmentation of this first product ion rendered the peak at m/z 357 due to a second hexose loss. The MS4 spectrum of this second product ion was identical to the MS2 spectrum of pinoresinol or G(8–8)G (Ye et al., 2005; Guo et al., 2007; Eklund et al., 2008; Ricci et al., 2008; Morreel et al., 2010a; Hanhineva et al., 2012). Therefore, this compound is pinoresinol dihexoside. When matching to MassBank, no hit with the same chemical formula as compound 79 was obtained. Using MetFrag, 37 molecules with the same chemical formula were retrieved from the Pubchem database of which pinoresinol diglucoside was among the best hits upon in silico fragmentation. To determine whether MetFrag would be suitable to determine the resinol aglycone structure, the MS4 spectrum of the m/z 357 second product ion (representing pinoresinol) was entered. Almost 1000 biological molecules were retrieved from the PubChem database and, upon in silico fragmentation, pinoresinol was among the top 25 best hits. All fragment ions were recognized by the software. 80. hex G(8–5)FA Glu The chemical formula of this ion was C31H36O15N (m/z 662.20968). Its MS2 spectrum was dominated by a peak at m/z 500 due to hexose loss. When subjected to MS3 fragmentation, water and formaldehyde losses rendered the peaks at m/z 482 and 470. The second product ion at m/z 371 was formed by glutamic acid loss (-129 Da). It also fragmented further by water and formaldehyde which explains the peaks at m/z 353 and 341. These losses are typical for 8–5linked neolignans/oligolignols (Morreel et al., 2010a). Taking the chemical formula into account, this compound was characterized as dihydroconiferyl alcohol 8–5 feruloyl glutamic acid hexoside. When matching to MassBank, no hit with the same chemical formula as compound 80 was obtained. Using MetFrag, 4 biological compounds having the same chemical formula were retrieved from the Pubchem database. None of them was a phenylcoumaran. 81. G(8–5)FA dihex The chemical formula of this ion was C32H39O17 (m/z 695.21981). No MS2 spectrum was recorded, but the compound was connected to G(8–5)FA hex 104 in the CSPP network via a 51

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

“hexosylation”  reaction.  Levels  of  both  compounds  were  very  highly  correlated  across  biological   replicates. Therefore, this compound was annotated as dihydroconiferyl alcohol 8–5 ferulic acid dihexoside. 82. G(8–O–4)G(8–O–4)FA hex This ion had C36H43O17 (m/z 747.25195) as chemical formula. This compound was linked to G(8–O–4)FA hex 70 via  a  “G  unit  addition”  in  the  CSPP  network.  Moreover, the abundances of both compounds were highly correlated. Its MS2 spectrum indicated the presence of a hexose moiety (m/z 585, -162 Da). Therefore, this compound was annotated as guaiacylglycerol 8–O–4 guaiacylglycerol ether 8–O–4 ferulic acid ether hexoside. 83. G(8–O–4)lariciresinol dihex The chemical formula of the ion was C42H55O20 (m/z 879.33274). The abundance was too low to obtain a MS2 spectrum. However, the compound was connected to lariciresinol dihex 71 via  a  “G   unit   addition”   conversion   in   the CSPP networks and both compound levels were highly correlated across biological replicates. This compound was annotated as guaiacylglycerol 8–O–4 lariciresinol ether dihexoside. 84. G(8–O–4)FA Glu This compound had the same chemical formula and showed the same CSPP network connections as G(8–O–4)FA Glu78. The abundances of both compounds were mutually highly correlated across biological samples. 85. G(8–O–4)SA hex The chemical formula of this ion was C27H33O14 (m/z 581.18713). The base peak in the MS2 spectrum was at m/z 419 indicating a hexose loss (-162 Da). MS3 fragmentation of this MS2 base peak   yielded   the   small   neutral   losses   characteristic   for   β-aryl ether neolignan/oligolignols (Morreel et al., 2010a), i.e., the so-called type I fragmentations: water (-18 Da) and a combined water/formaldehyde (-48 Da) loss leading to ions at m/z 401 and 471. The type II fragmentations start by cleavage of the 8–O–4-linkage providing product ions representing the connecting units. These ions, i.e., at m/z 223 and 195 indicated that a guaiacyl unit was coupled to a unit derived from sinapic acid. This compound was characterized as guaiacylglycerol 8–O–4 sinapic acid ester hexoside. When matching to MassBank, no hit with the same chemical formula as

52

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

compound 85 was obtained. Using MetFrag, 27 biological compounds having the same chemical formula  were  retrieved  from  the  Pubchem  database.  None  of  them  was  a  β-aryl ether. 86. G(8–O–4)FA malate The  ion’s  chemical  formula  was  C24H25O12 (m/z 505.13495). The MS2 base peak was at m/z 389 due to malate loss (-116 Da). Subjecting the ion to MS3 fragmentation showed peaks at m/z 341 (-48   Da   loss,   type   I   cleavage   of   β-aryl   ethers)   and   at   m/z   193   and   195   (due   to   β-aryl ethercharacteristic type II cleavages) corresponding with a unit derived from ferulic acid and a guaiacyl unit. Therefore, this compound was characterized as guaiacylglycerol 8–O–4 feruloyl malate ether. When matching to MassBank, no hit with the same chemical formula as compound 86 was obtained. 87. G(8–O–4)lariciresinol dihex The chemical formula was the same at that of G(8–O–4)lariciresinol dihex 83. The abundance was too low to obtain a MS2 spectrum. However, the compound was connected to lariciresinol dihex 71 via   a   “G   unit   addition”   conversion   in   the   CSPP   networks and both compound levels were highly correlated across biological replicates. This compound was annotated as another guaiacylglycerol 8–O–4 lariciresinol ether dihexoside isomer. 88. lariciresinol hex This ion had C26H33O11 (m/z 521.20286) as chemical formula. The MS2 spectrum indicated a hexose loss (-162 Da) rendering the base peak at m/z 359. Its MS3 spectrum was dominated by formaldehyde loss leading to the peak at m/z 329. This spectrum is typical for the CID spectrum of lariciresinol, i.e., reduced pinoresinol [G(8–8)G] (Eklund et al., 2008; Morreel et al., 2010a; Hanhineva et al., 2012). Therefore, this compound was characterized as lariciresinol hexoside. When matching to MassBank, no hit with the same chemical formula as compound 88 was obtained. Using MetFrag, 107 compounds were returned from the Pubchem database of which, upon in silico fragmentation, the best hit was our proposed structure. 89. hex G(8–5)FA malate The chemical formula of this ion was C30H33O16 (m/z 649.17719). Elimination of malate lead to the MS2 base peak at m/z 533 which, upon MS3 fragmentation, expelled a hexose group leading to the second product ion at m/z 371. This aglycone underwent losses of 18 and 30 Da representing water and formaldehyde and providing the second product ions at m/z 353 and 341. 53

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Both losses are characteristic for phenylcoumaran neolignans/oligolignols (Morreel et al., 2010a; 2010b). Therefore, this compound is dihydroconiferyl alcohol 8–5 feruloyl malate hexoside. When matching to MassBank, no hit with the same chemical formula as compound 89 was obtained. No structural elucidation via MetFrag was performed as G(8–5)FA is not present in the PubChem database. 90. G(8–O–4)G(8–O–4)SA hex The ion had C37H45O18 (m/z 777.26334) as chemical formula. No MS2 fragmentation occurred, but the compound was connected to G(8–O–4)FA hex 70 via  a  “S  unit  addition”  conversion  in   the CSPP network. The levels of both compounds are highly correlated across biological replicates. Although this would suggest the compound to be S(8–O–4)G(8–O–4)FA hex, the 8– O–4 coupling of a sinapyl alcohol radical to a oligolignol radical is not favored owing to oxidation potential differences (Syrjänen and Brunow, 1998). Therefore, this compound was annotated as guaiacylglycerol 8–O–4 guaiacylglycerol ether 8–O–4 sinapic acid ether hexoside. 91. G(8–O–4)pinoresinol dihex The   ion’s   chemical   formula   was   C42H53O20 (m/z 877.31705). No MS2 fragmentation was recorded, but the compound was connected to pinoresinol dihex 79 via   a   “G   unit   addition”   conversion in the CSPP network. Levels of both compounds were correlated across biological replicates and, thus, this compound was annotated as guaiacylglycerol 8–O–4 pinoresinol ether dihexoside. 92. G(8–O–4)SA hex The chemical formula of this ion was C27H33O14 (m/z 581.18785). Its MS2 spectrum showed a base peak at m/z 419 due to hexose loss. Upon MS3, the latter ion expelled water and formaldehyde (-48   Da)   yielding   the   second   product   ion   at   m/z   371.   In   addition   to   this   β-aryl ether-characteristic type I fragmentation, type II fragmentations lead to the ions at m/z 223 and m/z 195 representing the units derived from sinapic acid (supported by the observation of a second product ion at m/z 208 due to methyl radical loss from the m/z 223 ion) and from coniferyl alcohol (this ion fragmented further by formaldehyde loss yielding the second product ion at m/z 165). Therefore, this compound was characterized as guaiacylglycerol 8–O–4 sinapic acid ether hexoside. When matching to MassBank, no hit with the same chemical formula as

54

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

compound 92 was obtained. MetFrag-based structural characterization was not performed as G(8–O–4)SA is not present in the PubChem database. 93. G(8–O–4)SA Glu The ion had C26H30O12N (m/z 548.17812) as chemical formula, but no MS2 spectrum was obtained. In the CSPP network, the compound was linked to G(8–O–4)FA Glu 78 via a “methoxylation”  conversion.  Furthermore,  the  levels  of  both  compounds  were  correlated  across   biological replicates. Therefore, this compound is annotated as guaiacylglycerol 8–O–4 sinapoyl glutamic acid ether. 94. G(8–5)G hex The chemical formula of the acetate adduct of this compound was C28H35O13 (m/z 579.20835). MS2 fragmentation yielded the deprotonated compound at m/z 519. Other second product ions at m/z 357 (due to hexose loss) and at m/z 339 and 327 (losses of water and formaldehyde) – type I fragmentations typically observed in the spectrum of phenylcoumarans – pointed to a hexoside of a phenylcoumaran. The type II ion at m/z 221 indicated the presence of a guaiacyl unit and, thus this compound was characterized as (8–5)-dehydrodiconiferyl alcohol hexoside. When matching to MassBank, no hit with the same chemical formula as compound 94 was obtained. To observe the extent that the aglycone could be characterized using MetFrag, the MS 3 spectrum of the m/z 357 first product ion was entered. 988 compounds with the same chemical formula were retrieved from the PubChem database, yet G(8–5)G was not among the 50 best hits upon in silico fragmentation. Only four out of the five second product ions (m/z 339, 327, 221 and 203 but not m/z 191) could be explained by MetFrag. Furthermore, MetFrag returned G(8–8)G as a better hit rendering explanations for all of the second product ions. Although MetFrag recognized most of the product ions upon CID of any of the guaiacyl dimers, i.e., G(8–O–4)G, G(8–8)G and G(8–5)G, the large number of structural isomers (belonging to various biochemical classes) from the PubChem database that were equally likely good hits, was very confusing. Additionally, the MetFrag-proposed fragments were often not in agreement with previously published gas phase fragmentation reactions for these compounds (Eklund et al., 2008; Morreel et al., 2010a; Morreel et al., 2010b). Because of this lack of specificity and the absence of most of the (neo)lignan/oligolignol core structures in the PubChem database, MetFrag-based structural elucidation was not considered anymore for the remainder of these compounds. 55

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

95. G(8–O–4)G(8–O–4)SA hex The   ion’s   chemical   formula   was   C37H45O18 (m/z 777.26288). No MS2 spectrum was obtained, but  the  compound  was  connected  via  a  “S  unit  addition”  conversion  to  G(8–O–4)FA hex 70 in the CSPP network. The levels of both compounds were very highly correlated across biological replicates. In agreement with the same reasoning as mentioned for compound 90, compound 95 was annotated as guaiacylglycerol 8–O–4 guaiacylglycerol ether 8–O–4 sinapic acid ether hexoside. 96. lariciresinol hex This ion had the same chemical formula and similar MSn spectra as those of lariciresinol hex 88. Therefore, this compound was characterized as a lariciresinol hexoside isomer. When matching to MassBank, no hit with the same chemical formula as compound 96 was obtained. 97. G(8–O–4)FA malate The chemical formula of this ion was the same as the one of G(8–O–4)FA malate 86 and also the MSn spectra were similar. Consequently, this compound was characterized as an isomer of guaiacylglycerol 8–O–4 feruloyl malate ether. When matching to MassBank, no hit with the same chemical formula as compound 97 was obtained. 98. G(8–5)FA hex This ion had C26H29O12 (m/z 533.16673) as chemical formula. No MS2 spectrum was recorded, but a connection with G(8–O–4)FA hex 70 via   a   “hydration”   conversion   was   observed   in   the   CSPP network. Levels of both compounds were highly correlated and this compound was annotated as dihydroconiferylalcohol 8–5 ferulic acid hexoside or glycosmisic acid hexoside. 99. G(8–O–4)SA hex The chemical formula of this ion was the same as that of G(8–O–4)SA hex 92. The MSn spectra were very similar to those of G(8–O–4)SA hex 92. Thus, this compound was characterized as a guaiacylglycerol 8–O–4 sinapic acid ether hexoside isomer. When matching to MassBank, no hit with the same chemical formula as compound 99 was obtained. 100. G(8–O–4)G(8–O–4)FA hex The ion had C36H43O17 (m/z 747.25195) as chemical formula. No MS2 spectrum was recorded, but a connection with G(8–O–4)FA hex 70 via  a  “G  unit  addition”  conversion  was  observed  in  

56

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

the CSPP network. Levels of both compounds were correlated and this compound was annotated as guaiacylglycerol 8–O–4 guaiacylglycerol ether 8–O–4 ferulic acid ether hexoside. 101. G(8–O–4)pinoresinol dihex The ion had C42H53O20 (m/z 877.31685) as chemical formula. No MS2 spectrum was recorded, but a connection with pinoresinol dihex 79 via  a  “G  unit  addition”  conversion  was  observed  in   the CSPP network. Levels of both compounds were correlated and this compound was annotated as guaiacylglycerol 8–O–4 pinoresinol ether dihexoside. 102. G(8–O–4)G(8–O–4)FA hex The chemical formula of this ion was the same as for G(8–O–4)G(8–O–4)FA hex 100 (C36H43O17, m/z 747.25195). No MS2 spectrum was obtained for this precursor ion, but an insource hexose elimination rendered the ion at m/z 585.19801 (C30H33O12) in the full MS spectrum. The MS2 spectrum obtained for the latter ion showed ions at m/z 567 and 535 due to water loss (-18 Da) and the combined loss of water and formaldehyde (-48 Da). Both dissociations  are  typical  type  I  fragmentations  of  β-aryl ethers (Morreel et al., 2010a; Morreel et al., 2010b; Hanhineva et al., 2012). Type II fragmentations were evident as well (Morreel et al., 2010a; Morreel et al., 2010b; Hanhineva et al., 2012). A neutral loss of 196 Da, indicating the presence of a guaiacylglycerol moiety, lead to the ion at m/z 389 representing a dimeric moiety. The  m/z  341  ion  could  be  explained  by  another  β-aryl ether-specific type I-associated combined water/formaldehyde loss. The ion at m/z 193 suggests the presence of a unit derived from ferulic acid. Therefore, this compound was characterized as guaiacylglycerol 8–O–4 guaiacylglycerol ether 8–O–4 ferulic acid ether hexoside. When matching to MassBank, no hit with the same chemical formula as compound 102 was obtained. 103. G(8–O–4)S(8–8)G dihex The ion had C36H43O17 (m/z 747.25195) as chemical formula. No MS2 spectrum was obtained but   the   compound   was   connected   via   a   “S   unit   addition”   conversion   to   G(8–8)G dihex or pinoresinol dihex 79 in the CSPP network. The levels of both compounds were highly correlated across biological replicates. Although this would suggest the 8–O–4-coupling of a sinapyl alcohol radical to a radical from pinoresinol leading to the S(8–O–4)G(8–8)G aglycone, this reaction is unfavorable (Syrjänen and Brunow, 1998). Therefore, the most logical structure to be annotated was guaiacylglycerol 8–O–4 medioresinol ether dihexoside or buddlenol E dihexoside. 57

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

104. G(8–5)FA hex The ion had the same chemical formula (C26H29O12, m/z 533.16580) as G(8–5)FA hex 98. The MS2 spectrum showed a base peak at m/z 371 due to hexose loss (-162 Da). MS3 dissociation of the latter first product ion yielded the typical type I-associated neutral losses (water loss, formaldehyde loss and a combined water/methyl radical loss) of a phenylcoumaran neolignan/oligolignol (Morreel et al., 2010a; Morreel et al., 2010b) providing the second product ions at m/z 353, 341 and 338. A decarboxylation was evident from the peak at m/z 327. Type IIassociated peaks were observed at m/z 235 and 191 (Morreel et al., 2010a) resulting from cleavage of the phenylcoumaran into its composing units. Therefore, this compound was characterized as dihydroconiferylalcohol 8–5 ferulic acid hexoside or glycosmisic acid hexoside. When matching to MassBank, no hit with the same chemical formula as compound 104 was obtained. 105. G(8–O–4)lariciresinol hex The chemical formula of this ion was C36H45O15 (m/z 717.27758). Upon CID, the type I fragmentations  of  a  β-aryl ether were observed as first product ions at m/z 699 and 669 (Morreel et al., 2010a; Morreel et al., 2010b). Hexose loss resulted in the first product ion at m/z 555. The MS3 spectrum  of  the  latter  ion  showed  also  a  β-aryl ether-associated type I peak at m/z 507. The peak  at  m/z  195  is  due  to  a  β-aryl ether-associated type II cleavage and indicated the presence of a guaiacylglycerol 8–O–4 ether moiety. A second product ion at m/z 329 results from the further formaldehyde loss from a lariciresinol moiety, which has been previously observed to be a dominating pathway upon CID of lariciresinol (Morreel et al., 2010a). Therefore, this compound was characterized as guaiacylglycerol 8–O–4 lariciresinol ether hexoside. When matching to MassBank, no hit with the same chemical formula as compound 105 was obtained. 106.G(8–O–4)SA malate This ion had C25H27O13 (m/z 535.14570) as chemical formula. The MS2 spectrum was dominated by a base peak at m/z 419 due to malate loss. Further MS3 fragmentation of this ion yielded  a  βaryl ether type I pathway-associated peak at m/z 371 (Morreel et al., 2010a; 2010b). Furthermore,  type  II  fragmentation  of  the  β-aryl ether linkage rendered ions at m/z 223 and 195 representing moieties derived from sinapic acid and guaiacylglycerol. Therefore, this compound was characterized as guaiacylglycerol 8–O–4 sinapoyl malate ether. When matching to MassBank, no hit with the same chemical formula as compound 106 was obtained. 58

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

107. G(8–O–4)G(8–5)G hex The chemical formula of the acetate adduct of this compound was C38H47O17 (m/z 775.28366). This compound was connected to G(8–5)G hex 94 via   a   “G   unit   addition”   conversion   in   the   CSPP network. Both compound levels were highly correlated across biological replicates. Its MS2 spectrum was dominated by the combined loss of the acetate and the hexoside yielding the first product ion at m/z 553. Further MS3 fragmentation of this first product ion lead to second product ions at m/z 535, 523 and 505 due to the loss of water, formaldehyde and the combined loss of water and formaldehyde, respectively. These second product ions are characteristic type I fragmentations of β-aryl ethers (Morreel et al., 2010a). Thus, this compound was annotated as guaiacylglycerol 8–O–4 dehydrodiconiferyl alcohol ether hexoside. 108. G(8–O–4)G(8–O–4)SA hex This   ion’s  chemical   formula  was  C37H45O18 (m/z 777.26346). No MS2 spectrum was obtained, but the compound was connected to G(8–O–4)FA hex 70 via  a  “S  unit  addition”  conversion  in   the CSPP network. Both compound levels were correlated across biological replicates. However, as mentioned above (see compounds 95 and 103), a sinapyl alcohol radical will not readily couple via an 8–O–4-linkage to a guaiacyl-derived phenolic function. Therefore, this compound should be annotated as guaiacylglycerol 8–O–4 guaiacylglycerol ether 8–O–4 sinapic acid ether hexoside. 109. S(8–5)FA hex The chemical formula of this ion was C27H31O13 (m/z 563.17746). No MS2 spectrum was obtained, but the compound was connected to G(8–5)FA hex 98 via   a   “methoxylation”   conversion in the CSPP network. Both compound levels were correlated across biological replicates. Thus, this compound was annotated as dihydrosinapyl alcohol 8–5 ferulic acid hexoside. 110. G(8–O–4)S(8–8)G dihex The ion had C43H55O21 (m/z 907.32659) as chemical formula. Its MS2 spectrum showed a peak at m/z 745 indicative for a hexose loss (-162 Da). No further MSn information was obtained, but the compound was connected to pinoresinol dihex 79 via   a   “S   unit   addition”   conversion.   Both   compound levels were very highly correlated across biological replicates. Because of the resilience of the sinapyl alcohol radical to form an 8–O–4-linkage to a guaiacyl phenolic function 59

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

(see compounds 95 and 103), this compound should be annotated as guaiacylglycerol 8–O–4 medioresinol ester dihexoside. When matching to MassBank, no hit with the same chemical formula as compound 110 was obtained. 111. G(8–O–4)G(8–O–4)S hex This ion had as chemical formula C37H47O17 (m/z 763.28446). No MS2 spectrum was obtained, but the compound was connected to G(t8–O–4)G hex 74 via  a  “S  unit  addition”  conversion  in  the   CSPP network. Both compound levels were highly correlated across biological replicates. Because a sinapyl alcohol radical does not readily form an 8–O–4-linkage to a guaiacyl phenolic function (see compounds 95 and 103), this compound was annotated as guaiacylglycerol 8–O–4 guaiacylglycerol ether 8–O–4 sinapyl alcohol ether hexoside. 112. G(8–5)FA Glu The chemical formula of this ion was C25H26O10N (m/z 500.15606). MS2 dissociation lead to the m/z 371 ion due to a neutral loss of 129 Da, indicating the presence of a glutamic acid-derived moiety. Fragmentation of the first product ion at m/z 371 rendered the first product ions at m/z 353, 341, 327, 235 and 191 in a similar way as observed upon MS 3 fragmentation of the m/z 371 ion in the MS2 spectrum of G(8–5)FA hex 104. This suggested dehydroconiferyl alcohol 8–5 feruloyl glutamic acid as structure for compound 112. The remaining first product ions provided further evidence. The phenylcoumaran-associated type II cleavage (Morreel et al., 2010a) produced the first product ion at m/z 364 representing a moiety derived from the feruloyl glutamic acid unit of compound 112. Interestingly, as has been shown earlier for dicarboxylic acids (Kanawati and Schmitt-Kopplin, 2010), a combined loss of water and carbondioxide from the glutamate moiety yielding the first product ion at m/z 438 (-62 Da) was observed. Therefore, this compound was characterized as dehydroconiferyl alcohol 8–5 feruloyl glutamic acid. When matching to MassBank, no hit with the same chemical formula as compound 112 was obtained. 113. pinoresinol hex The chemical formula of this ion was C26H31O11 (m/z 519.18705). The MS2 spectrum was dominated by the ion at m/z 357 due to hexose loss. MS3 fragmentation of this base peak yielded a spectrum identical to the MS2 spectrum observed for pinoresinol (Morreel et al., 2010a). Therefore, this compound is G(8–8)G hexoside or pinoresinol hexoside. When matching to MassBank, no hit with the same chemical formula as compound 113 was obtained. 60

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

114. G(8–O–4)G(8–O–4)lariciresinol hex This ion had C46H57O19 (m/z 913.35306) as chemical formula. The MS2 base peak at m/z 751 arose from hexose loss. A water loss (-18 Da) and a combined water/formaldehyde loss (-48 Da), i.e., the   type   I   fragmentations   of   a   β-aryl ether (Morreel et al., 2010a; Morreel et al., 2010b), occurred both from the precursor ion (yielding the first product ions at m/z 895 and 865) as well as from the MS2 base peak (leading to the first product ions at m/z 733 and 703). The first product ion at m/z 555 is formed by expelling a G unit from the MS2 ion at m/z 751; a type II fragmentation known  to  occur  in  β-aryl ethers. The further structural characterization was solely based on the CSPP network in which this compound was linked to G(8–O–4)lariciresinol hex 105 via   a   “G   unit   addition”   conversion.   Both   compound   levels   were   very   highly   correlated across biological replicates. Compound 114 was characterized as guaiacylglycerol 8–O–4 guaiacylglycerol ether 8–O–4 lariciresinol ether hexoside. When matching to MassBank, no hit with the same chemical formula as compound 114 was obtained. 115. G(8–O–4)SA malate The ion of this compound had C25H27O13 (m/z 535.14509) as chemical formula. No MS2 spectrum was obtained, but the compound was connected to trans-sinapoyl malate 58 via  a  “G   unit   addition”   conversion   in   the   CSPP   network.   Both   compound   levels   were highly correlated across biological replicates. Therefore, this compound was annotated as guaiacylglycerol 8–O–4 sinapoyl malate ester, i.e., an isomer of compound 106. 116. G(8–5)FA hex The chemical formula of this ion was identical and its MS2 spectrum highly similar to those of G(8–5)FA

hex

104.

Therefore,

this

compound

was

characterized

as

another

dihydroconiferylalcohol 8–5 ferulic acid hexoside or glycosmisic acid hexoside isomer. When matching to MassBank, no hit with the same chemical formula as compound 116 was obtained. 117. G(8–O–4)G(8–5)FA hex This ion had C36H41O16 (m/z  729.24154)  as  chemical  formula.  It’s  MS2 spectrum was dominated by the peak at m/z 567 resulting from hexose loss. The MS3 spectrum of this MS2 base peak showed ions at m/z 549   and   519   resulting   from   β-aryl ether characteristic type I cleavages leading to water (-18 Da) and a combined water/formaldehyde (-48 Da) loss (Morreel et al., 2010a; Morreel et al., 2010b). The type II fragmentations yielded ions at m/z 195 and 371 61

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

representing a guaiacylglycerol moiety and a dimeric moiety. The latter first product ion lost water and formaldehyde rendering the peaks at m/z 353 and 341 and can, thus, be pinpointed as a phenylcoumaran (Morreel et al., 2010a). Therefore, this compound was characterized as guaiacylglycerol 8–O–4 glycosmisic acid ether hexoside. When matching to MassBank, no hit with the same chemical formula as compound 117 was obtained. 118. G(8–5)FA Glu The ion had the same chemical formula and almost identical MSn spectra as G(8–5)FA Glu 112 and is, thus, an isomer of dehydroconiferyl alcohol 8–5 feruloyl glutamic acid. When matching to MassBank, no hit with the same chemical formula as compound 118 was obtained. 119. G(8–O–4)G(8–O–4)FA The chemical formula and the MS2 spectrum were the same as obtained for the in-source fragment ion of G(8–O–4)G(8–O–4)FA hex 102. Therefore, this compound is an isomer of guaiacylglycerol 8–O–4 guaiacylglycerol ether 8–O–4 ferulic acid ether. When matching to MassBank, no hit with the same chemical formula as compound 119 was obtained. 120. G(8–8)S hex The ion had C27H33O12 (m/z 549.19770) as chemical formula. The base peak in its MS2 spectrum was the ion at m/z 387 resulting from a hexose loss. When this ion was subjected to MS3 fragmentation, the second product ions at m/z 372 and 341 arising from methyl radical (-15 Da) and formic acid (-46 Da) losses were those expected from the type I fragmentation of resinol lignans (Morreel et al., 2010a; see also references mentioned for pinoresinol dihex 79). The type II cleavages of the resinol structure itself provided the ions at m/z 181, 166, 151 and 136, indicating that this compound is medioresinol hexoside. When matching to MassBank, no hit with the same chemical formula as compound 120 was obtained. 121. G(8–O–4)G(8–O–4)S hex The chemical formula of this ion was C37H47O17 (m/z 763.28385). No MS2 spectrum was obtained,  but   this  compound   was  connected  via  a  “S  unit  addition”  conversion  to   G(t8–O–4)G hex 74 in the CSPP network. Moreover, both compound levels were very highly correlated across biological replicates. Because 8–O–4-coupling of a sinapyl alcohol radical to the phenolic function of a G unit is not favored (Syrjänen and Brunow, 1998), only guaiacylglycerol 8–O–4

62

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

guaiacylglycerol ether 8–O–4 sinapyl alcohol ether hexoside is possible as annotation for this (neo)lignan/oligolignol. 122. G(8–O–4)pinoresinol hex The  ion’s  chemical   formula  was  C36H43O15 (m/z 715.26152). The compound was connected to the acetate adduct of G(8–5)G hex 94 via   the   spurious   “vanillyl   alcohol   condensation”   conversion in the CSPP network. The addition of vanillyl alcohol does not occur in Arabidopsis to our knowledge, but the levels of both peaks were highly correlated across biological replicates. In addition, compound 122 was   also   connected   via   a   “G   unit   addition”   to   the   deprotonated form of G(8–5)G hex 94. Its MS2 spectrum clearly indicated that it was a hexoside (first product ion at m/z 553, -162 Da loss). The most straightforward annotation for this compound was G(8–O–4)G(8–5)G hex, yet the MS3 spectrum of this compound showed the characteristic   type   I   fragmentations   for   a   β-aryl ether (-18, -30 and -48 Da losses yielding the second product ions at m/z 535, 523 and 505; Morreel et al., 2010a).   Furthermore,   the   β-aryl ether-specific type II fragmentations yielded the second product ions at m/z 357, 343 and 327 which are typical for a resinol moiety (Morreel et al., 2010b). Therefore, G(8–O–4)pinoresinol hex was taken as annotation for compound 122. 123. G(8–O–4)S(8–5)FA hex The chemical formula of this ion was C37H43O17 (m/z 759.25082). The base peak at m/z 597 in its MS2 spectrum was derived from a hexose loss. Further MS3 fragmentation of this first product ion   showed   the   typical   type   I   cleavages   of   a   β-aryl ether (m/z 579 and 549) (Morreel et al., 2010a; see structural elucidation of e.g. compounds 70, 74, 76, 85), but also a decarboxylation was  evident  from  the  ion  at  m/z  553.  The  characteristic  type  II  cleavages  of  a  β-aryl ether yielded the ions at m/z 195 and m/z 401 representing a guaiacylglycerol moiety and the remaining dimeric moiety. Further dissociation of m/z 401 lead to the second product ions at m/z 383 and 371 due to water and formaldehyde losses and correspond with the type I cleavages of a phenylcoumaran. As this dimeric moiety (m/z 401) underwent also a decarboxylation (m/z 357), compound 123 was characterized as guaiacylglycerol 8–O–4 dehydrosinapyl alcohol ether 8–5 ferulic acid hexoside. When matching to MassBank, no hit with the same chemical formula as compound 123 was obtained. 124. G(8–5)FA malate 63

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

This ion had C24H23O11 (m/z 487.12414) as chemical formula. The MS2 spectrum showed a base peak at m/z 371 due to malate loss (-116 Da). MS3 fragmentation of the MS2 base peak yielded a spectrum identical to the MS3 spectrum described for G(8–5)FA hex 104. Therefore, this compound was characterized as dihydroconiferylalcohol 8–5 feruloyl malate or glycosmisoyl malate. When matching to MassBank, no hit with the same chemical formula as compound 124was obtained. 125. G(8–O–4)G(8–5)FA malate The chemical formula of this ion was C34H35O15 (m/z 683.19815). Its MS2 spectrum was dominated by malate loss (-116 Da) leading to the peak at m/z 567. Upon MS3 dissociation, the β-aryl ether-characteristic type I losses of 18 and 48 Da were observed at m/z 549 and 519 (Morreel et al., 2010a; see structural elucidation of e.g. compounds 70, 74, 76, 85).   A   β-aryl ether-associated type II cleavage explained the origin of the ion at m/z 371. Other second product ions at m/z 353 and 341 corresponded with the phenylcoumaran-associated type I losses occurring from the dimeric moiety represented by the m/z 371 ion. Therefore, this compound was characterized as guaiacylglycerol 8–O–4 glycosmisoyl malate ether. When matching to MassBank, no hit with the same chemical formula as compound 125 was obtained. 126. G(8–O–4)S(8–8)G hex This ion had C37H45O16 (m/z 745.27271) as chemical formula. No MS2 spectrum was recorded, but this compound was connected to G(8–8)G hex or pinoresinol hex 113 via  a  “S  unit  addition”   conversion in the CSPP network. The levels of both compounds were correlated across biological replicates. Although the most logical structure would be S(8–O–4)G(8–8)G hex, the 8–O–4-radical radical coupling of sinapyl alcohol to a guaiacyl phenolic endgroup of an oligolignol/(neo)lignan is not favored (Syrjänen and Brunow, 1998). Therefore, this compound was annotated as guaiacylglycerol 8–O–4 medioresinol ether hexoside. 127. G(8–O–4)S(8–8)G hex This ion had C37H45O16 (m/z 745.27233) as chemical formula. The MS2 base peak was observed at m/z 583 and resulted from hexose loss. The MS3 spectrum of this first product ion was showing type I losses (at m/z 565 and 535) and type II cleavages (at m/z 387 and 195) typical for a  β-aryl ether (Morreel et al., 2010a; see structural elucidation of e.g. compounds 70, 74, 76, 85). The quartet of peaks at m/z 387, 373, 357 and 343 are typical for a resinol-containing oligolignol 64

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

(Morreel et al., 2010b) and, thus, this compound was characterized as guaiacylglycerol 8–O–4 medioresinol ether hexoside. When matching to MassBank, no hit with the same chemical formula as compound 127 was obtained. 128. S(8–5)FA hex The chemical formula of this ion was C27H31O13 (m/z 563.17473). No MS2 spectrum was recorded, but this compound was connected to G(8–5)FA hex 104 via   a   “methoxylation”   conversion in the CSPP network. The levels of both compounds were highly correlated across biological replicates and this compound was annotated as dihydrosinapyl alcohol 8–5 ferulic acid hexoside. 129. G(8–O–4)S(8–5)FA malate This ion had C35H37O16 (m/z  713.20870)  as  chemical  formula.  It’s  MSn spectra were similar to those of G(8–O–4)G(8–5)FA malate 125, but the m/z values of all MSn peaks were increased with 30 amu. Therefore, this compound was characterized as guaiacylglycerol 8–O–4 dihydrosinapyl alcohol 8–5 feruloyl malate ether. When matching to MassBank, no hit with the same chemical formula as compound 129 was obtained. 130. G(8–O–4)G(8–5)FA The chemical formula of this ion was C30H31O11 (m/z 567.18674). No MS2 spectrum was generated, but the compound was connected in the CSPP network to the ion of G(8–5)FA malate 124 representing the dilignol core structure resulting from an in-source malate loss. Both compounds were  connected  via  a  “G  unit  addition”  conversion  and  their  levels  were  correlated   across biological replicates. This compound was annotated as guaiacylglycerol 8–O–4 glycosmisic acid ether. 131. S(8–5)FA hex The ion had C27H31O13 (m/z 463.17432) as chemical formula. No MS2 spectrum was recorded, but this compound was connected to G(8–5)FA hex 104 via  a  “methoxylation”  conversion  in  the   CSPP network. The levels of both compounds were highly correlated across biological replicates and this compound was annotated as dihydrosinapyl alcohol 8–5 ferulic acid hexoside. Indolics 132. 6-hydroxyindole-3-carboxylate dihex 65

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

This ion had C21H26O13N (m/z 500.14021) as chemical formula. Its MS2 spectrum had a base peak at m/z 338 due to hexose loss. MS3 fragmentation of the latter ion yielded a peak at m/z 176 due to a second hexose loss. The second product ion at m/z 132 arose due to a decarboxylation occurring from the m/z 176 ion and indicated that the aglycone was hydroxyindole-3carboxylate. In the MS2 spectrum, abundant peaks were present due to hexose cross-ring cleavages (ions at m/z 440, 410 and 380) reminding of an ester-linked hexose. However, also in the MS3 spectrum, although much less abundant, these hexose cross-ring cleavages were evident (ions at m/z 278, 248 and 218) despite that no second carboxyl acid function was available. This indicates that one of the two hexoses is esterified whereas the other one is present as a phenolic hexoside. Therefore, this compound is 6-hydroxyindole-3-oyl hexose hexoside. The position of the hydroxyl group was derived from its connection to 6-hydroxyindole-3-carboxylate hex 133 in the CSPP network. When matching to MassBank, no hit with the same chemical formula as compound 132 was obtained. Using MetFrag, 8 biological compounds having the same chemical formula were retrieved from the Pubchem database. None of the returned hits was an indolic compound. 133. 6-hydroxyindole-3-carboxylate hex The  ion’s  chemical  formula  was  C15H16O8N (m/z 338.08829). The MS2 spectrum showed a base peak at m/z 176 due to hexose loss (-162 Da). Because no cross-ring cleavages were observed, this hexose is attached as a hexoside. The MS3 spectrum of the m/z 176 first product ion rendered a second product ion at m/z 132 due to a decarboxylation. This MS3 spectrum was similar to that of indole-3-carboxylate hex 135 except for the 16 amu shift to higher m/z values of first and second product ions. This compound was more precisely characterized as 6hydroxyindole-3-carboxylate hexoside because this compound has been previously observed in Arabidopsis leaf extracts. When matching to MassBank, no hit with the same chemical formula as compound 133 was obtained. Via MetFrag, 54 structural isomers were returned from the PubChem database. After in silico fragmentation, two methoxyindolic compounds were found among the five best hits. 134. indole-3-carboxylate dihex The chemical formula of this ion was C21H26O12N (m/z 484.14531). No MS2 spectrum was recorded, but this compound was connected to indole-3-carboxylate hex 135 via a “hexosylation”   conversion   in   the   CSPP   network.   The   levels   of   both   compounds   were   highly   66

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

correlated across biological replicates and this compound was annotated as indole-3-carboxylate dihexoside. 135. indole-3-carboxylate hex This ion had C15H16O7N (m/z 322.09348) as chemical formula. The major peak (m/z 160) in its MS2 spectrum originates from hexose loss (-162 Da). Additional first product ions at m/z 262, 232 and 202, corresponding with losses of 60, 90 and 120 Da, respectively, result from hexose cross-ring cleavages and occur whenever the reducing end of the hexose is free (Carroll et al., 1995) or when the hexose is linked via an ester bond (Vanholme et al., 2010; see also explanation MSn spectra of protocatechoyl Glc 44). Evidence that the hexose is linked in an ester bond arises as well from the observation of a decarboxylation occurring upon MS3 fragmentation of the aglycone (second product ion at m/z 116) which does not occur from the hexosylated compound. In addition, an ester bond might as well fragment via formation of a ketene neutral (Debrauwer et al., 1992; see other references in the explanation for protocatechoyl Glc 44). This yields the hexose-associated second product ion at m/z 179. Based on the chemical formula of the aglycone, a nitrogen-containing aromatic molecule is expected. As the ynolate ion is not observed in the MS2 spectrum (see explanation for protocatechoyl Glc 44), this suggests that the ester function is directly connected to the aromatic system. Therefore, this compound was characterized as indole-3-oyl hexose and the aglycone structure was confirmed by MSn analysis of an indole-3-oyl standard. When matching to MassBank, no hit with the same chemical formula as compound 135 was obtained. With MetFrag, 157 structural isomers were downloaded from the PubChem database. Following in silico fragmentation, our proposed structure belonged to the four best hits. All of the product ions could be explained by the software. 136. 6-hydroxyindole-3-carboxylate dihex sinA The chemical formula of this ion was C32H36O17N (m/z 706.19914). The MS2 base peak at m/z 544 resulted from hexose loss (-162 Da). An additional loss of the sinapic acid moiety as a ketene (-206 Da) yielded the first product ion at m/z 338. Ester cleavage of the precursor ion produced the complementary peaks at m/z 223, representing sinapate, and 482 originating from sinapic acid loss. The ion at m/z 367 corresponded with the sinapoyl hexose moiety. The MS3 spectrum of the first product ion at m/z 544 rendered the complementary ions at m/z 367 and 176 representing sinapoyl hexose and 6-hydroxyindole-3-carboxylate. The presence of 6hydroxyindole-3-carboxylate was further supported in the MS3 fragmentation of the first product 67

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

ion at m/z 338 which rendered ions at m/z 176 and 132 due to a hexose loss and a subsequent decarboxylation from the 6-hydroxyindole-3-carboxylate aglycone. Furthermore, the presence of a sinapoyl hexose moiety was proven by MS4 dissociation of the second product ion at m/z 367, yielding the third product ion at m/z 223. Therefore, this compound was characterized as 6hydroxyindole-3-oyl sinapoyl dihexoside. When matching to MassBank, no hit with the same chemical formula as compound 136 was obtained. Using MetFrag, two structural isomers were returned from the PubChem database, but none was retained as a valuable candidate after in silico fragmentation. Indolic glucosinolate catabolites 137.  5’-Glc-dihydroascorbigen The ion had C21H26O11N (m/z 468.15056) as chemical formula. Its MS2 spectrum was not similar to any of the compounds described up to now, yet a clear connection with the indolics was evident from the CSPP network. The compound was linked with indole-3-carboxylate dihex 134 via  an  “oxygenation”  conversion.  Both  compound  levels  were  highly  correlated  across  biological   replicates. Searching indolics with the same chemical formula in the CAS database yielded one candidate, i.e., 5’-O-β-D-glucosyl dihydroascorbigen, of which the structure is expected to follow similar dissociation channels in the gas phase as those represented by the MS 2 spectrum of our unknown compound 137. An outline of the various gas phase fragmentation reactions is shown in Supplemental Figure 8B. Although both charge-driven and charge-remote pathways can occur during the gas phase fragmentations of negative ions, the former type of pathways will prevail whenever possible (Thevis et al., 2003). Therefore, in the explanation of the MSn dissociations of compound 137 (Supplemental Figure 8B), charge-remote pathways were considered whenever no charge-driven pathway could be deduced. Arguably, the most acidic site is the C1´-position which is allylic to the indole moiety. Following deprotonation, a proton transfer from the 3´-OH group via a fivecenter intermediate (Supplemental Figure 8B, pathway a) or from the 3´-OH group via an eightcenter intermediate (Supplemental Figure 8B, pathway c) can be envisaged. Alternatively, an E1cb-like reaction might open the lactone ring (Supplemental Figure 8B, pathway b). Subsequent fragmentation along pathway a also opens up the lactone ring. Dependent whether the negative charge then ends up at the C2´-position via cleavage of the 3´–4´-linkage or leads to 68

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

the cleavage of the 2´–3´-linkage, a 2´-hydroxyindolylpropanyl aldehyde (m/z 188) or 2´hydroxyindolylpropanoic acid (m/z 204) anion is formed. The latter structure resembles that of Trp leading to a similar fragmentation behavior (Supplemental Figure 8B, table insert). Interestingly, because ammonia is a less electronegative species than water, the peak at m/z 186 (ammonia loss) in the MS2 spectrum of Trp is much smaller than the same peak in the MS3 spectrum of compound 137 (due to water loss in the latter case). Furthermore, the peak at m/z 158 is due to decarboxylation in the case of Trp fragmentation but formic acid loss upon dissociation   of   the   2’-hydroxyindolylpropanoyl   anion.   Formic   acid   loss   is   characteristic   for   αhydroxy acids (Bandu et al., 2006). Pathway b explains the ions at m/z 440, 406, 226, 208 and 179. The E1cb reaction opens the lactone ring via carbon monoxide loss (m/z 440) followed by cleavage of the 3´–4´-bond, hence, providing a second pathway leading to the ion at m/z 188. Decarbonylation from the ester group  in  an  α-hydroxy ester moiety has been previously suggested to occur in negative ionization tandem-in-space MS/MS (Mancel et al., 2004). However, instead of decarbonylation, the E1cb reaction might as well lead to decarboxylation of the lactone ring. Concomitantly with the decarboxylation, desaturation of the 3´–4´-bond or the 4´–5´-bond will occur. Whereas the former desaturation leads to water loss providing the ion at m/z 406, desaturation of the 4´–5´bond will result in the elimination of a glucose anion (m/z 179). However, as the glucose anion is transiently withhold in an ion-neutral complex (Bowie, 1990), complex dissociation might be preceded by a proton transfer from the 3´-position to the glucose anion and the subsequent elimination of water from the indolyl-containing anion. This yields the highly conjugated anion at m/z 226. Expelling another water molecule renders then the peak at m/z 208. Following initial proton transfer from the 6´-OH group (pathway c), the resulting alkoxide anion could attack the C6´-position in an internal SN2 reaction, cleaving the 5´–6´-bond and, thus, providing the indole-bearing anion at m/z 246 and a neutral epoxide. A further decarbonylation from the lactone ring delivers the ion at m/z 218. Internal S N2 reactions leading to epoxide formation have been previously described to occur in the gas phase (Binkley et al., 1996; Mancel et al., 2004). Finally, a charge-remote glycosidic bond cleavage explains the loss of 162 Da yielding the peak at m/z 306 (Carroll et al., 1995), whereas a subsequent water loss produces the peak at m/z 288 (Mulroney et al., 1999). Because the most abundant product ions agreed with the proposed structure, this compound was characterized as 5’-O-β-D-glucosyl 69

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

dihydroascorbigen. When matching to MassBank, no hit with the same chemical formula as compound 137 was obtained. Via MetFrag, 9 structural isomers were returned from the PubChem database. After in silico fragmentation, one indolic compound was found among the retained hits, yet its aglycone was not dihydroascorbigen. However, a high resolution tandem-inspace MS/MS spectrum has been obtained previously (Montaut and Bleeker, 2010) and was highly similar to the tandem-in-time MS2 spectrum obtained in our study (S. Montaut, personal communication). 138. 5΄-Glc-dihydroneoascorbigen This ion had C22H28O12N (m/z 498.16129) as chemical formula. Its MSn spectra were similar to those of 5΄-Glc-dihydroascorbigen 137 except that most of the product ions appeared at m/z values that were 30 Th higher. Therefore, this compound was characterized as 5΄-O-β-D-glucosyl dihydroneoascorbigen. When matching to MassBank, no hit with the same chemical formula as compound 138 was obtained. Via MetFrag, 12 structural isomers were returned from the PubChem database. After in silico fragmentation, three indolic compounds were found among the retained hits, yet their aglycones were not similar to dihydroneoascorbigen. 139. hydroxy-dihydroascorbigen hex The chemical formula of this ion was C21H26O12N (m/z 484.14515). Its MS2 spectrum was characterized by the loss of hexose (-162 Da) yielding the base peak at m/z 322. MS3 fragmentation of the latter first product ion yielded second product ions at m/z 188 and 204 that were reminiscent of the 2´-hydroxyindolylpropanyl aldehyde or 2´-hydroxyindolylpropanoic acid anions that are produced upon MS2 fragmentation of 5´-Glc-dihydroascorbigen 137. Therefore, compound 139 was structurally characterized as hydroxyl-dihydroascorbigen hex. Apocarotenoids 140. corchoionoside C This ion had C19H29O8 (m/z 385.18684) as chemical formula. The MS2 spectrum of compound 140 showed a hexose loss leading to the ion at m/z 223. Possible biological candidate compound classes containing representatives with the chemical formula of the aglycone in the CAS database, were the jasmonates and the apocarotenoids. The similarity between the MS 2 spectrum of abscissic acid (an apocarotenoid) and the MS3 spectrum of the aglycone of compound 140 suggested the latter to be an apocarotenoid. In both cases, the major product ion was observed at 70

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

m/z 153 corresponding with the apocarotenoid ring structure. Blumenol A glucoside or corchoionoside C has already been observed in the Brassicaceae (Cutillo et al., 2005) and had the correct chemical formula. Indeed, all MSn fragmentations could be explained based on the anion of this candidate structure. In case of the base peak, the presumed charge-driven mechanism starting from deprotonation of the 6-position is given in Supplemental Figure 8C (pathway a). A proton transfer from the 3´-OH function, succeeded by ethyne elimination, explains the m/z 153 ion production. This reaction is driven by the neutral loss of two highly electronegative species, i.e., ethyne and acetaldehyde, and the resonance-stabilized product ion. Such an ion might readily loose a methyl radical as both 5-linked methyl groups are allylic to the conjugated double bonds. Indeed, this is verified by the observation of a base peak at m/z 138 in the MS4 spectrum of the m/z 153 second product ion. An alternative fragmentation pathway occurs when charge delocalization leads to the elimination of the 3´-OH function (pathway b, Supplemental Figure 8C). This results in a dehydration when an allylic proton is abstracted from the 3´-position via a hydroxide ion-neutral complex. Therefore, this compound was characterized as corchoionoside C. When matching to MassBank, no hit with the same chemical formula as compound 140 was obtained. Via MetFrag, 68 structural isomers were returned from the PubChem database. After in silico fragmentation, the four best hits in which all four first product ions (m/z 153, 161, 205 and 223) could be explained, were all apocarotenoids. Among these four, a stereomer of corchoionoside C (or 6S,9R blumenol A hexoside) was included, i.e., 6S,9Sroseoside. 141. blumenol A malonylhex The chemical formula of this ion was C22H31O11 (m/z 471.18690). No MS2 spectrum was obtained, but in-source decarboxylation yielded the ion at m/z 427.19708 (C21H31O9,  Δppm  =  0.65) for which a MS2 spectrum was recorded. The first product ion at m/z 385 due to 42 Da loss together with the in-source decarboxylation is typical for malonate moieties (Pollier et al., 2011). The base peak at m/z 367 resulted from a further water loss and, when subjected to MS 3 fragmentation, rendered the characteristic fragment ions for blumenol A. Therefore, this compound was characterized as blumenol A malonyl hexoside. 142. blumenol A acetylmalonylhex The ion had C24H33O12 (m/z 513.19789) as chemical formula. No MS2 spectrum was recorded, but in-source fragmentation produced the ion at m/z 469.20771 (C23H33O10,  Δppm  =  -0.45) that 71

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

dissociated into the first product ions at m/z 427 and 385 due to two subsequent 42 Da losses. Together with the in-source decarboxylation, these ions refer to the presence of a malonyl and an acetyl moiety. Water losses from both first product ions rendered the peaks at m/z 409 and 367. MS3 fragmentation of the m/z 409 ion showed a very similar spectrum as the MS3 spectrum of the m/z 223 ion of corchoionoside C 140. Therefore, this compound was characterized as blumenol A acetyl malonyl hexoside. 143. blumenol A acetylmalonylhex The ion had C24H33O12 (m/z 513.19761) as chemical formula. No MS2 spectrum was recorded, but the compound was connected to blumenol A malonylhex 141 via an “acetylation”  conversion   in the CSPP network. The levels of both compounds were highly correlated across biological replicates. Therefore, this compound was annotated as another isomer of blumenol A acetyl malonyl hexoside. Others 144. glutathione This ion had C10H16O6N3S (m/z 306.07695) as chemical formula. Matching its MS2 spectra with those in the MassBank database, yielded an almost perfect match to the spectrum of glutathione. 145. butyl acetylhex The chemical formula of the ion was C12H21O7 (m/z 277.12977). No MS2 spectrum was recorded, but an in-source fragment at m/z 235.11946 (C10H19O6,   Δppm   =   3.18)   indicated   the   loss of an acetyl group. A MS2 spectrum for the latter fragment ion was obtained that was very similar to that of a hexose (March and Stadey, 2005), except for the presence of the first product ion at m/z 191. The base peak at m/z 161 indicated that the charge mainly remained with the hexose moiety. Therefore, this in-source fragment is a hexose to which an apolar butyl moiety is attached. Taking into account that an acetyl group dissociated in-source, this compound is acetylbutyl hexoside. With MetFrag, 203 structural isomers of the in-source fragment were downloaded from the PubChem database. After in silico fragmentation, several butyl glucosides were found among the 20 best hits. For all of them, all 10 product ions were predicted.

Interpretation of CSPPs Associated with Moderate to High Correlation Coefficients 72

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

In   this   study,   CSPPs   in   which   the   abundances   of   the   “substrate”   and   “product”   peaks   were   moderately or highly correlated, could be classified into four groups. The first group contained CSPPs representing true enzymatic reactions (group 1). Examples were the oxygenations in aliphatic glucosinolate biosynthesis (Figures 4B and 5) and the glycosylations in flavonoid metabolism (Figure 5 and Supplemental Figure 5A). CSPPs from group 2 represented organic reactions that occur during the biosynthesis of the compound class, but more upstream in the biosynthetic pathway than suggested by the CSPP. For example, the sequence of methylations and oxygenations in the sub-network of the aliphatic glucosinolates did not reflect the biochemical reaction sequence. Methylene insertions (represented by methylation CSPPs) take place earlier in glucosinolate biosynthesis than the oxygenations (Figures 4A and 5). This side-chain elongation occurs before the glucosinolate is synthesized, i.e., at the level of the amino acid precursor of the glucosinolate. However, using CSPPs to characterize unknown compounds still yielded valid structures. In flavonoid metabolism (Figure 5 and Supplemental Figure 5A), kaempherol glycosides and their quercetin analogues, e.g., kaempherol 3-O-rhamnosyl-7-O-rhamnoside 42, kaempherol 3-O-glucosyl-7-Orhamnoside 37 and kaempherol 3-O-glucosyl(1→2)rhamnosyl-7-O-rhamnoside 31, and their respective quercetin analogues 35, 32 and 29 (see Supplemental Data Set 1 Online, Supplemental Figure 4), are associated with CSPPs representing oxygenations occurring more upstream in the pathway, i.e., those  performed  by  flavonoid  3΄-hydroxylase. Again, although the position of the reaction in the biosynthetic pathway cannot be unambiguously inferred from the CSPP network, the network still yields valid structural information about unknown metabolites. Confounding the CSPP-based structural elucidation were those CSPPs in which the “substrate”  was  connected  with  a  structurally  similar  isomer  of  the  expected  “product”  (group  3).   This can readily happen in, e.g., phenylpropanoid/(neo)lignan metabolism. In phenylpropanoid metabolism (Figure 5), the side-chain of feruloyl-CoA might be reduced or oxidized, leading to coniferyl alcohol (a hydroxycinnamyl alcohol) or ferulic acid (a hydroxycinnamic acid), respectively. By the addition of a methoxy group to the benzene ring, coniferyl alcohol and ferulic acid can be further converted to sinapyl alcohol (a hydroxycinnamyl alcohol) and sinapic acid (a hydroxycinnamic acid). During radical cross-coupling reactions, coniferyl and sinapyl alcohol provide the guaiacyl (G) and syringyl (S) units in lignins and (neo)lignans, whereas units derived from ferulic and sinapic acids are denoted as FA and SA. Noticeably, units derived from 73

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

both hydroxycinnamic acids (FA and SA) as well as from both hydroxycinnamyl alcohols (G and S) differ only by a methoxyl group. This has to be taken into account when characterizing the structures of their coupling products. For example, the guaiacyl (Gun) CSPP between sinapoyl glucose 55, and 76 (Supplemental Figure 5B), suggested G(8–O–4)SA hexose,  which  is   the  βaryl ether crossed dimer derived from a G monomer (coniferyl alcohol) coupling to sinapoyl hexose, as the structure for the latter compound. However, when the MS2 spectrum of compound 76 was recorded, it was deemed to be S(8–O–4)FA hexose, which is a structural isomer – a  βaryl ether crossed dimer derived from an S monomer (sinapyl alcohol) coupling to feruloyl hexose. In other words, a methoxy substituent had to be swapped to obtain an agreement between the CSPP-based and the MS2-based structural elucidation. In a fourth group of CSPPs, often no biochemical support for the presumed conversion type  was  obtained  based  on  the  molecular  structures  as  the  structural  moieties  of  the  “substrate”   were   shuffled   in   the   “product”   structure.   For   example, kaempherol 3-arabinosyl-rhamnosyl-7rhamnoside 36 and kaempherol 3-rhamnosyl-7-malonylglucoside 39 are both derived from the same aglycone precursor, yet follow different glycosylation paths. The sequence of glycosylation reactions for each resulted in a final mass difference corresponding with a glycerol moiety, i.e., the mass difference between the malonyl-glucosyl and arabinosyl-rhamnosyl moieties. Obviously,  structural   annotations  based  on  the   CSPP  network  in   which  “low-correlated”   edges   are removed, will be erroneous in this case, necessitating the inclusion of an MS2 spectral similarity algorithm. (Bio)chemical Validity  of  the  “High  CSPP”  Group Conversions Because of the higher fraction of true (bio)chemical CSPPs in the sub-network   of   the   “high CSPP”  group  versus  that  of  the  network  containing  all  CSPPs,  the  former  sub-network was, as expected, more similar to the topology of a metabolic network (Figures 2C and 2D). Furthermore,  the  levels  of  the  “substrate”  and  “product”  of  each  CSPP  of  the  “high  CSPP”  group   were   on   average   more   highly   correlated   than   those   of   the   “low   CSPP”   group.   Although   reductions (Red, Table 1) and the addition of syringyl (S) units to oligolignols/(neo)lignans (Sun, Table 1) are true conversions in Arabidopsis, they belonged   to   the   “low   CSPP”   group,   yet   the   correlations  found  for  their  CSPPs  were  still  higher  than  those  for  the  other  CSPPs  of  the  “low   CSPP”  group.  Furthermore,  when  selecting  the  node  file  for  “reduction”  peak  pairs  based  solely   74

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

on mass differences and checking subsequently their retention time differences (Figure 3E, Supplemental Figure 3),   the   “substrate”   and   “product”   peaks   often   eluted   too   closely   to   each   other to be annotated as a CSPP and, hence, the number of CSPPs was underestimated. The classification  of  the  “Sun”  conversion  as  a  “low  CSPP”  group  conversion  was  a  borderline  case   as it had an intermediate number of CSPPs (245, Table 1). Conversely, glycerol addition is not a well-known reaction in Arabidopsis metabolism, yet the conversion belonged to the  “high  CSPP   number”   conversions.   However,   MS2 data of some Arabidopsis (neo)lignans/oligolignols showed neutral losses corresponding to the expected mass of a dissociated glycerol moiety (e.g., feruloyl glycerol 52, Supplemental Data Set 1), indicating that this conversion was correctly classified   as   a   “high   CSPP   number”   conversion.   Finally,   the   use   of   correlations   as   support   for   true metabolism-based CSPPs emanates also from the observation that the sub-networks of highly correlated peaks in the CSPP network corresponded with biochemically related compounds, such as those of the glucosinolate, flavonoid, sinapate and oligolignol/(neo)lignan metabolism.

Searching Compound Classes and New Enzymatic Reactions with the CSPP Algorithm The information obtained from the CSPP algorithm not only aids structural characterization of compounds but provides a new metabolomics tool to analyze profiled data. Firstly, by using one or  a  few  representatives  of  a  compound  class  as  “bait”,  the  CSPP  network  strategy  illustrates the ease   of   teasing   out   all   compounds   with   similar   structures   from   the   “haystack”   of   chromatographic peaks at once. Comparing the MSn ion trees of all similar compounds aids in interpretation of the gas-phase fragmentation pathways and, thus, the spectral interpretation. The power of this approach is illustrated with, e.g., the trisinapoylgentiobiose isomers 66 and 68 (Supplemental Data Set 1, Supplemental Figure 4) for which the structure would likely never have been proposed using solely the complex MS2 spectrum. Perhaps most importantly, the CSPP-based annotation procedure allows (tentative) assignments of structures for unknowns that are only present in minute amounts that compromise their purification and, thus, their identification by NMR. Examples of such structurally characterized low-abundance compounds are G(8–O–4)SA malate 106 and G(8–O–4)FA malate hexoside 69 (Supplemental Data Set 1, Supplemental Figure 4). The metabolic class of some sub-networks could not be elucidated as 75

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

none of their members could be structurally annotated. Purification followed by NMR-based identification of the more abundant compounds would allow further characterization of these sub-networks of unknown compound classes. Secondly, the use of CSPPs enabled the pinpointing of as yet unknown enzymatic reactions. For example, the levels of kaempherol 3-O-arabinosyl-rhamnosyl-7-O-rhamnoside 36 and kaempherol 3-O-arabinosyl-7-O-rhamnoside 38 were mutually strongly correlated (Pearson r2 = 0.92; Supplemental Data Set 1), and the CSPP conversion between them suggests a rhamnosyltransferase-catalyzed reaction. Furthermore, the presence of a sinapic acid derivative of a kaempherol dirhamnoside (compound 43) indicates that hydroxycinnamate derivatization within flavonoid metabolism is not restricted to the anthocyanins (Nakabayashi et al., 2009). Overall, the CSPP algorithm allowed the structural annotation of 145 compounds in Arabidopsis leaves from a total of 229 profiled compounds. Consequently, 60% of all profiled compounds were annotated/characterized, a percentage that has never been obtained before in a metabolomics experiment. Searching the Scifinder database revealed that only 40 compounds had been previously detected in Arabidopsis leaves (Supplemental Data Set 1), although another 14 compounds had been described in other Arabidopsis tissues. Ten compounds were found before in one or more Brassicaceae species but were observed here for the first time in Arabidopsis. From the remainder of the annotated compounds, 20 have been described in other plant families and 61 structures were not found in the database at all. Compound classes that are already well-known in Arabidopsis leaves were the phenylpropanoids, flavonoids and glucosinolates (Figure 5). Nevertheless, the subclass of the hydroxy-(methylsulfinyl)-alkyl glucosinolates has not yet been described in Arabidopsis and only a few members have been found in the plant kingdom (Fahey et al., 2001). Because no hydroxy-(methylthio)-alkyl glucosinolates were detected, this strongly suggests that the hydroxy-(methylsulfinyl)-alkyl glucosinolates arise by hydroxylation of the methylsulfinylalkyl glucosinolate which is itself formed from the oxidation of the corresponding methylthioalkyl glucosinolate (Figure 4A). Thus, CSPP networks also provide information on biosynthetic routes. In addition to the indolic glucosinolates, their breakdown products were detected as well (Supplemental Data Set 1). These breakdown products are formed whenever indolic glucosinolates are released from the vacuole and encounter myrosinase. For the first time dihydroascorbigen-based glucosinolate catabolites were observed in Arabidopsis. Their existence was supported by the high MS2 similarity  of  5΄76

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Glc-dihydroascorbigen 137 with a previously recorded MS fragmentation spectrum for this compound (Montaut and Bleeker, 2010). Strikingly, a high number of (neo)lignans/oligolignols (Morreel et al., 2004; 2010a; 2010b) were present in the leaves. These structures displayed the various units and combinatorial linkages that are found in lignin, yet they were all adorned with hexose, malate and/or glutamate which classifies them as (neo)lignans. Some of these compounds have recently been described in flax stem tissues (Huis et al., 2012) and were also found in Arabidopsis stems (Vanholme et al., 2012). Finally, apocarotenoids implicated here were unknown to exist in Arabidopsis to date.

SUPPLEMENTAL REFERENCES Al-Shehbaz, I.A., and Al-Shammary, K.I. (1987). Distribution and chemotaxonomic significance of glucosinolates in certain Middle-Eastern cruciferae. Biochem. Syst. Ecol. 15: 559-569. Attygalle, A., Ruzicka, J., Varughese, D., and Sayed, J. (2006). An unprecedented ortho effect in mass spectrometric fragmentation of even-electron negative ions from hydroxyphenyl carbaldehydes and ketones. Tetrahedron Lett. 47: 4601-4603. Bandu, M.L., Grubbs, T., Kater, M., and Desaire, H. (2006). Collision induced dissociation of alpha hydroxy acids: Evidence of an ion-neutral complex intermediate. Int. J. Mass Spectrom. 251: 40-46. Bandu, M.L., Watkins, K.R., Bretthauer, M.L., Moore, C.A., and Desaire, H. (2004). Prediction of MS/MS Data. 1. A focus on pharmaceuticals containing carboxylic acids. Anal. Chem. 76: 1746-1753. Bialecki, J.B., Ruzicka, J., Weisbecker, C.S., Haribal, M., and Attygalle, A.B. (2010). Collision-induced dissociation mass spectra of glucosinolate anions. J. Mass Spectrom. 45: 272-283. Binkley, R.W., Binkley, E.R., Duan, S.M., Tevesz, M.J.S., and Winnik, W. (1996). Negativeion mass spectrometry of carbohydrates. A mechanistic study of the fragmentation reactions of dideoxy sugars. Journal of Carbohydrate Chemistry 15: 879-895. Born, M., Ingemann, S., and Nibbering, N.M.M. (1997). Formation and chemistry of radical anions in the gas phase. Mass Spectrom. Rev. 16: 181-200.

77

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Bowie, J.H. (1990). The fragmentations of even-electron organic negative ions. Mass Spectrom. Rev. 9: 349-379. Carroll, J.A., Willard, D., and Lebrilla, C.B. (1995). Energetics of cross-ring cleavages and their relevance to the linkage determination of oligosaccharides. Anal. Chim. Acta 307: 431-447. Cataldi, T.R.I., Lelario, F., Orlando, D., and Bufo, S.A. (2010). Collision-induced dissociation of the A+2 isotope ion facilitates glucosinolates structure elucidation by electrospray ionization-tandem mass spectrometry with a linear quadrupole ion trap. Anal. Chem. 82: 5686-5696. Chapple, C.C.S., Vogt, T., Ellis, B.E., and Somerville, C.R. (1992). An Arabidopsis mutant defective in the general phenylpropanoid pathway. Plant Cell 4: 1413-1424. Cheng, C., and Gross, M.L. (2000). Applications and mechanisms of charge-remote fragmentation. Mass Spectrom. Rev. 19: 398-420. Cutillo, F., Dellagreca, M., Previtera, L., and Zarrelli, A. (2005). C13 norisoprenoids from Brassica fruticulosa. Nat. Prod. Res. 19: 99-103. Cuyckens, F., and Claeys, M. (2004). Mass spectrometry in the structural analysis of flavonoids. J. Mass Spectrom. 39: 1-15. Dauwe, R., Morreel, K., Goeminne, G., Gielen, B., Rohde, A., Van Beeumen, J., Ralph, J., Boudet, A.-M., Kopka, J., Rochange, S.F., Halpin, C., Messens, E., and Boerjan, W. (2007). Molecular phenotyping of lignin-modified tobacco reveals associated changes in cell-wall metabolism, primary metabolism, stress metabolism and photorespiration. Plant J. 52: 263-285. Debrauwer, L., Paris, A., Rao, D., Fournier, F., and Tabet, J.-C. (1992). Mass spectrometric studies  on  17β-estradiol-17-fatty acid esters: Evidence for the formation of anion-dipole intermediates. Organic Mass Spectrometry 27: 709-719. De Kok, L.J., and Graham, M. (1989). Levels of pigments, soluble proteins, amino acids and sulfhydryl compounds in foliar tissue of Arabidopsis thaliana during dark-induced and natural senescence. Plant Physiol. Biochem. 27: 203-209. DePuy, C.H. (2000). An introduction to the gas phase chemistry of anions. Int. J. Mass Spectrom. 200: 79-96.

78

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Deyama, T., Ikawa, T., Kitagawa, S., and Nishibe, S. (1986). The constituents of Eucommia ulmoides Oliv. IV. Isolation of a new sesquilignan glycoside and iridoids. Chem. Pharm. Bul. 34: 4933-4938. Deyama, T., Ikawa, T., and Nishibe, S. (1985). The constituents of Eucommia ulmoides Oliv. II. Isolation and structures of three new lignan glycosides. Chem. Pharm. Bull. 33: 36513657. Eichinger, P.C.H., Dua, S., and Bowie, J.H. (1994). A comparison of skeletal rearrangement reactions of even-electron anions in solution and in the gas phase. Int. J. Mass Spectrom. Ion Processes 133: 1-12. Eklund, P.C., Backman, M.J., Kronberg, L.Å., Smeds, A.I., and Sjöholm, R.E. (2008). Identification of lignans by liquid chromatography-electrospray ionization ion-trap mass spectrometry. J. Mass Spectrom. 43: 97-107. Fabre, N., Rustan, I., de Hoffmann, E., and Quetin-Leclercq, J. (2001). Determination of flavone, flavonol, and flavanone aglycones by negative ion liquid chromatography electrospray ion trap mass spectrometry. J. Am. Soc. Mass Spectrom. 12: 707-715. Fabre, N., Poinsot, V., Debrauwer, L., Vigor, C., Tulliez, J., Fourasté, I., and Moulis, C. (2007). Characterisation of glucosinolates using electrospray ion trap and electrospray quadrupole time-of-flight mass spectrometry. Phytochem. Anal. 18: 306-319. Fahey, J.W., Zalcmann, A.T., and Talalay, P. (2001). The chemical diversity and distribution of glucosinolates and isothiocyanates among plants. Phytochem. 56: 5-51. Fernández-Arroyo, S., Barrajón-Catalán, E., Micol, V., Segura-Carretero, A., and Fernández-Gutiérrez, A. (2010). High-performance liquid chromatography with diode array detection coupled to electrospray time-of-flight and ion-trap mass spectrometry to identify phenolic compounds from a Cistus ladanifer aqueous extract. Phytochem. Anal. 21: 307-313. Ferreres, F., Sousa, C., Valentão, P., Seabra, R.M., Pereira, J.A., and Andrade, P.B. (2007). Tronchuda cabbage (Brassica oleracea L. var. costata DC) seeds: phytochemical characterization and antioxidant potential. Food Chem. 101: 549-558. Ferreres, F., Llorach, R., and Gil-Izquierdo, A. (2004). Characterization of the interglycosidic linkage in di-, tri-, tetra- and pentaglycosylated flavonoids and differentiation of

79

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

positional isomers by liquid chromatography/electrospray ionization tandem mass spectrometry. J. Mass Spectrom. 39: 312-321. Fournier, F., Perlat, M.-C., and Tabet, J.-C. (1995). Control of internal proton transfers on ion-dipole complexes from [M-H]− ions of diphenol esters. Rapid Commun. Mass Spectrom. 9: 13-17. Fournier, F., Remaud, B., Blasco, T., and Tabet, J.C. (1993). Ion-dipole complex formation from deprotonated phenol fatty acid esters evidenced by using gas-phase labeling combined with tandem mass spectrometry. J. Am. Soc. Mass Spectrom. 4: 343-351. Fraser, C.M., Thompson, M.G., Shirley, A.M., Ralph, J., Schoenherr, J.A., Sinlapadech, T., Hall, M.C., and Chapple, C. (2007). Related Arabidopsis serine carboxypeptidase-like sinapoylglucose acyltransferases display distinct but overlapping substrate specificities. Plant Physiol. 144: 1986-1999. Goujon, T., Sibout, R., Pollet, B., Maba, B., Nussaume, L., Bechtold, N., Lu, F., Ralph, J., Mila, I., Barrière, Y., Lapierre, C., and Jouanin, L. (2003). A new Arabidopsis thaliana mutant deficient in the expression of O-methyltransferase impacts lignins and sinapoyl esters. Plant Mol. Biol. 51: 973-989. Gronert, S. (2001). Mass spectrometric studies of organic ion/molecule reactions. Chem. Rev. 101: 329-360. Gronert, S. (2005). Ouadrupole ion trap studies of fundamental organic reactions. Mass Spectrom. Rev. 24: 100-120. Grossert, J.S., Cook, M.C., and White, R.L. (2006). The influence of structural features on facile McLafferty-type, even-electron rearrangements in tandem mass spectra of carboxylate anions. Rapid Commun. Mass Spectrom. 20: 1511-1516. Guo, H., Liu, A.-H., Ye, M., Yang, M., and Guo, D.-A. (2007). Characterization of phenolic compounds in the fruits of Forsythia suspensa by high-performance liquid chromatography coupled with electrospray ionization tandem mass spectrometry. Rapid Commun. Mass Spectrom. 21: 715-729. Hagemeier, J., Schneider, B., Oldham, N.J., and Hahlbrock, K. (2001). Accumulation of soluble and wall-bound indolic metabolites in Arabidopsis thaliana leaves infected with virulent or avirulent Pseudomonas syringae pathovar tomato strains. Proc. Natl. Acad. Sci. U.S.A. 98: 753-758. 80

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Hanhineva, K., Rogachev, I., Aura, A.-M., Aharoni, A., Poutanen, K., and Mykkänen, H. (2012). Identification of novel lignans in the whole grain rye bran by non-targeted LCMS metabolite profiling. Metabolomics 8: 399-409. Hansen, C.H., Wittstock, U., Olsen, C.E., Hick, A.J., Pickett, J.A., and Halkier, B.A. (2001). Cytochrome

P450

CYP79F1

from

Arabidopsis

catalyzes

the

conversion

of

dihomomethionine and trihomomethionine to the corresponding aldoximes in the biosynthesis of aliphatic glucosinolates. J. Biol. Chem. 276: 11078-11085. Harrison, A.G. (1992). Chemical ionization mass spectrometry. (Boca Raton, Florida: CRC Press). Haughn, G.W., Davin, L., Giblin, M., and Underhill, E.W. (1991). Biochemical genetics of plant secondary metabolites in Arabidopsis thaliana. Plant Physiol. 97: 217-226. Heinonen, M., Rantanen, A., Mielikäinen, T., Kokkonen, J., Kiuru, J., Ketola, R.A., and Rousu, J. (2008). FiD: a software for ab initio structural identification of product ions from tandem mass spectrometric data. Rapid Commun. Mass Spectrom. 22: 3043-3052. Ho, J.-C., Chen, C.-M., and Row, L.-C. (2003). Neolignans from the parasitic plants. Part 1. Aeginetia indica. J. Chin. Chem. Soc. 50: 1271-1274. Hou, S., Zhu, J., Ding, M., and Lv, G. (2008). Simultaneous determination of gibberellic acid, indole-3-acetic acid and abscisic acid in wheat extracts by solid-phase extraction and liquid chromatography-electrospray tandem mass spectrometry. Talanta 76: 798-802. Hughes, R.J., Croley, T.R., Metcalfe, C.D., and March, R.E. (2001). A tandem mass spectrometric study of selected characteristic flavonoids. Int. J. Mass Spectrom. 210-211: 371-385. Huis, R., Morreel, K., Fliniaux, O., Lucau-Danila, A., Fenart, S., Grec, S., Neutelings, G., Chabbert, B., Mesnard, F., Boerjan, W., and Hawkins, S. (2012). Natural hypolignification is associated with extensive oligolignol accumulation in flax stems. Plant Physiol. 158: 1893-1915. Hvattum, E., and Ekeberg, D. (2003). Study of the collision-induced radical cleavage of flavonoid glycosides using negative electrospray ionization tandem quadrupole mass spectrometry. J. Mass Spectrom. 38: 43-49.

81

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Justesen, U. (2000). Negative atmospheric pressure chemical ionisation low-energy collision activation mass spectrometry for the characterisation of flavonoids in extracts of fresh herbs. J. Chromatogr. A 902: 369-379. Kanawati, B., and Schmitt-Kopplin, P. (2010). Exploring rearrangements along the fragmentation of glutaric acid negative ion: a combined experimental and theoretical study. Rapid Commun. Mass Spectrom. 24: 1198-1206. Kanawati, B., Joniec, S., Winterhalter, R., and Moortgat, G.K. (2007). Mass spectrometric characterization of small oxocarboxylic acids and gas phase ion fragmentation mechanisms studied by electrospray triple quadrupole-MS/MS-TOF system and DFT theory. Int. J. Mass Spectrom. 266: 97-113. Kanawati, B., Herrmann, F., Joniec, S., Winterhalter, R., and Moortgat, G.K. (2008). Mass spectrometric characterization of β-caryophyllene ozonolysis products in the aerosol studied using an electrospray triple quadrupole and time-of-flight analyzer hybrid system and density functional theory. Rapid Commun. Mass Spectrom. 22: 165-186. Kitamura, S., Matsuda, F., Tohge, T., Yonekura-Sakakibara, K., Yamazaki, M., Saito, K., and

Narumi,

I.

(2010).

Metabolic

profiling

and

cytological

analysis

of

proanthocyanidins in immature seeds of Arabidopsis thaliana flavonoid accumulation mutants. Plant J. 62: 549-559. Kjær, A., and Schuster, A. (1970). Glucosinolates in Erysimum hieracifolium L.; three new, naturally occurring glucosinolates. Acta Chem. Scand. 24: 1631-1638. Kliebenstein, D.J., Kroymann, J., Brown, P., Figuth, A., Pedersen, D., Gershenzon, J., and Mitchell-Olds, T. (2001). Genetic control of natural variation in Arabidopsis glucosinolate accumulation. Plant Physiol. 126: 811-825. Kuang, H., Sun, S., Yang, B., Xia, Y., and Feng, W. (2009). New megastigmane sesquiterpene and indole alkaloid glucosides from the aerial parts of Bupleurum chinense DC. Fitoterapia 80: 35-38. Le Gall, G., Metzdorff, S.B., Pedersen, J., Bennett, R.N., and Colquhoun, I.J. (2005). Metabolite profiling of Arabidopsis thaliana (L.) plants transformed with an antisense chalcone synthase gene. Metabolomics 1: 181-198. Leplé, J.-C., Dauwe, R., Morreel, K., Storme, V., Lapierre, C., Pollet, B., Naumann, A., Kang, K.-Y., Kim, H., Ruel, K., Lefèbvre, A., Joseleau, J.-P., Grima-Pettenati, J., De 82

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Rycke, R., Andersson-Gunnerås, S., Erban, A., Fehrle, I., Petit-Conil, M., Kopka, J., Polle, A., Messens, E., Sundberg, B., Mansfield, S.D., Ralph, J., Pilate, G., and Boerjan, W. (2007). Downregulation of cinnamoyl-coenzyme A reductase in poplar: multiple-level phenotyping reveals effects on cell wall polymer metabolism and structure. Plant Cell 19: 3669-3691. López-Carbonell, M., and Jáuregui, O. (2005). A rapid method for analysis of abscisic acid (ABA) in crude extracts of water stressed Arabidopsis thaliana plants by liquid chromatography-mass spectrometry in tandem mode. Plant Physiol. Biochem. 43: 407411. Levsel, K., Schiebel, H.-M., Terlouw, J.K., Jobst, K.J., Elend, M., Preiß, A., Thiele, H., and Ingendoh, A. (2007). Even-electron ions: a systematic study of the neutral species lost in the dissociation of quasi-molecular ions. J. Mass Spectrom. 42: 1024-1044. Mancel, V., Sellier, N., Lesage, D., Fournier, F., and Tabet, J.-C. (2004). Gas phase enantiomeric distinction of (R)- and (S)-aromatic hydroxy esters by negative ion chemical ionization mass spectrometry using a chiral reagent gas. Int. J. Mass Spectrom. 237: 185195. March, R.E., and Stadey, C.J. (2005). A tandem mass spectrometric study of saccharides at high mass resolution. Rapid Commun. Mass Spectrom. 19: 805-812. Marzouk, M.M., Al-Nowaihi, A.-S.M., Kawashty, S.A., and Saleh, N.A.M. (2010). Chemosystematic studies on certain species of the family Brassicaceae (Cruciferae) in Egypt. Biochem. Syst. Ecol. 38: 680-685. Matsubara, Y., Kumamoto, H., Sawabe, A., Iizuka, Y., and Okamoto, K. (1985). Structure and physiological activity of phenylpropanoid glycosides in lemon, unshui and orange peelings. Kinki Daigaku Igaku Zasshi 10: 51-58. Matsuda, F., Hirai, M.Y., Sasaki, E., Akiyama, K., Yonekura-Sakakibara, K., Provart, N.J., Sakurai, T., Shimada, Y., and Saito, K. (2010). AtMetExpress development: a phytochemical atlas of Arabidopsis development. Plant Physiol. 152: 566-578. Meyermans, H., Morreel, K., Lapierre, C., Pollet, B., De Bruyn, A., Busson, R., Herdewijn, P., Devreese, B., Van Beeumen, J., Marita, J.M., Ralph, J., Chen, C., Burggraeve, B., Van Montagu, M., Messens, E., and Boerjan, W. (2000). Modifications in lignin and accumulation of phenolic glucosides in poplar xylem upon down-regulation of 83

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

caffeoyl-coenzyme A O-methyltransferase, an enzyme involved in lignin biosynthesis. J. Biol. Chem. 275: 36899-36909. Ming, D.-S., Jiang, R.-W., But, P.P.-H., Towers, G.H.N., and Yu, D.-Q. (2002). A new compound from Geum rivale L. J. Asian Nat. Prod. Res. 4: 217-220. Miyaichi, Y., and Tomimori., T. (1998). Studies on constituents of Scutellaria species. XIX. Lignan glycosides of roots of Scutellaria baicalensis GEORGI. Nat. Med. 52: 82-86. Montaut, S., and Bleeker, R.S. (2010). Isolation and   structure   elucidation   of   5′-O-β-dglucopyranosyl-dihydroascorbigen from Cardamine diphylla rhizome. Carbohydr. Res. 345: 1968-1970. Morreel, K., Ralph, J., Kim, H., Lu, F., Goeminne, G., Ralph, S., Messens, E., and Boerjan, W. (2004). Profiling of oligolignols reveals monolignol coupling conditions in lignifying poplar xylem. Plant Physiol. 136: 3537-3549. Morreel, K., Goeminne, G., Storme, V., Sterck, L., Ralph, J., Coppieters, W., Breyne, P., Steenackers, M., Georges, M., Messens, E., and Boerjan, W. (2006). Genetical metabolomics of flavonoid biosynthesis in Populus: a case study. Plant J. 47: 224-237. Morreel, K., Kim, H., Lu, F., Dima, O., Akiyama, T., Vanholme, R., Niculaes, C., Goeminne, G., Inzé, D., Messens, E., Ralph, J., and Boerjan, W. (2010a). Mass spectrometry-based fragmentation as an identification tool in lignomics. Anal. Chem. 82: 8095-8105. Morreel, K., Dima, O., Kim, H., Lu, F., Niculaes, C., Vanholme, R., Dauwe, R., Goeminne, G., Inzé, D., Messens, E., Ralph, J., and Boerjan, W. (2010b). Mass spectrometrybased sequencing of lignin oligomers. Plant Physiol. 153: 1464-1478. Mulroney, B., Peel, J.B., and Traeger, J.C. (1999). Theoretical study of deprotonated glucopyranosyl disaccharide fragmentation. J. Mass Spectrom. 34: 856-871. Nagatani, Y., Warashina, T., and Noro, T. (2002). Studies on the constituents from the aerial part of Baccharis dracunculifolia DC. II. Chem. Pharm. Bull. 50: 583-589. Nakabayashi, R., Kusano, M., Kobayashi, M., Tohge, T., Yonekura-Sakakibara, K., Kogure, N., Yamazaki, M., Kitajima, M., Saito, K., and Takayama, H. (2009). Metabolomics-oriented isolation and structure elucidation of 37 compounds including two anthocyanins from Arabidopsis thaliana. Phytochem. 70: 1017-1029.

84

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Pedras, M.S.C., and Zheng, Q.-A. (2010). Metabolic responses of Thellungiella halophila/salsuginea to biotic and abiotic stresses: metabolite profiles and quantitative analyses. Phytochemistry 71: 581-589. Petersen, B.L., Chen, S., Hansen, C.H., Olsen, C.E., and Halkier, B.A. (2002). Composition and content of glucosinolates in developing Arabidopsis thaliana. Planta 214: 562-571. Plumb, G.W., Price, K.R., Rhodes, M.J.C., and Williamson, G. (1997). Antioxidant properties of the major polyphenolic compounds in broccoli. Free Radical Res. 27: 429435. Pollier, J., Morreel, K., Geelen, D., and Goossens, A. (2011). Metabolite profiling of triterpene saponins in Medicago truncatula hairy roots by liquid chromatography Fourier transform ion cyclotron resonance mass spectrometry. J. Nat. Prod. 74: 1462-1476. Rasoanaivo, P., Ratsimamanga-Urverg, S., Messana, I., De Vicente, Y., and Galeffi, C. (1990). Cassinopin, a kaempferol trirhamnoside from Cassinopsis madagascariensis. Phytochemistry 29: 2040-2043. Reeks, L.B., Eichinger, P.C.H., Bowie, J.H. (1993). Ortho-rearrangements of Oalkylphenoxide anions. Rapid Commun. Mass Spectrom. 7: 286-287. Reichelt, M., Brown, P.D., Schneider, B., Oldham, N.J., Stauber, E., Tokuhisa, J., Kliebenstein, D.J., Mitchell-Olds, T., and Gershenzon, J. (2002). Benzoic acid glucosinolate esters and other glucosinolates from Arabidopsis thaliana. Phytochemistry 59: 663-671. Ricci, A., Fiorentino, A., Piccolella, S., D'Abrosca, B., Pacifico, S., and Monaco, P. (2010). Structural discrimination of isomeric tetrahydrofuran lignan glucosides by tandem mass spectrometry. Rapid Commun. Mass Spectrom. 24: 979-985. Ricci, A., Fiorentino, A., Piccolella, S., Golino, A., Pepi, F., D'Abrosca, B., Letizia, M., and Monaco, P. (2008). Furofuranic glycosylated lignans: a gas-phase ion chemistry investigation by tandem mass spectrometry. Rapid Commun. Mass Spectrom. 22: 33823392. Rochfort, S.J., Trenerry, V.C., Imsic, M., Panozzo, J., and Jones, R. (2008). Class targeted metabolomics: ESI ion trap screening methods for glucosinolates based on MSn fragmentation. Phytochemistry 69: 1671-1679.

85

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Rohde, A., Morreel, K., Ralph, J., Goeminne, G., Hostyn, V., De Rycke, R., Kushnir, S., Van Doorsselaere, J., Joseleau, J.-P., Vuylsteke, M., Van Driessche, G., Van Beeumen, J., Messens, E., and Boerjan, W. (2004). Molecular phenotyping of the pal1 and pal2 mutants of Arabidopsis thaliana reveals far-reaching consequences on phenylpropanoid, amino acid, and carbohydrate metabolism. Plant Cell 16: 2749-2771. Sakushima, A., Coskun, M., Tanker, M., and Tanker, N. (1994) A sinapic acid ester from Boreava orientalis. Phytochemistry 35: 1481-1484. Schmidt, T.J., Alfermann, A.W., and Fuss, E. (2008). High-performance liquid chromatography/mass spectrometric identification of dibenzylbutyrolactone-type lignans: insights into electrospray ionization tandem mass spectrometric fragmentation of lign-7eno-9,9'-lactones and application to the lignans of Linum usitatissimum L. (Common Flax). Rapid Commun. Mass Spectrom. 22: 3642-3650. Shahat, A.A., Cuyckens, F., Wang, W., Abdel-Shafeek, K.A., Husseiny, H.A., Apers, S., Van Miert, S., Pieters, L., Vlietinck, A.J., and Claeys, M. (2005). Structural characterization of flavonol di-O-glycosides from Farsetia aegyptia by electrospray ionization and collision-induced dissociation mass spectrometry. Rapid Commun. Mass Spectrom. 19: 2172-2178. Shimomura, H., Sashida, Y., and Mimaki, Y. (1987) Phenolic glycerides from Lilium auratum. Phytochemistry 26: 844-845. Stroobant, V., Rozenberg, R., Bouabsa, e.M., Deffense, E., and de Hoffmann, E. (1995). Fragmentation of conjugate bases of esters derived from multifunctional alcohols including triacylglycerols. J. Am. Soc. Mass Spectrom. 6: 498-506. Sumner, L.W., Amberg, A., Barrett, D., Beale, M.H., Beger, R., Daykin, C.A., Fan, T.W.M., Fiehn, O., Goodacre, R., Griffin, J.L., Hankemeier, T., Hardy, N., Harnly, J., Higashi, R., Kopka, J., Lane, A.N., Lindon, J.C., Marriott, P., Nicholls, A.W., Reily, M.D., Thaden, J.J., and Viant, M.R. (2007). Proposed minimum reporting standards for chemical analysis. Metabolomics 3: 211-221. Syrjänen, K., and Brunow, G. (1998). Oxidative cross coupling of p-hydroxycinnamic alcohols with   dimeric  arylglycerol   β-aryl ether lignin model compounds. The effect of oxidation potentials. J. Chem. Soc., Perkin Trans. 1: 3425-3429.

86

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Thevis, M., Schänzer, W., and Schmickler, H. (2003). Effect of the location of hydrogen abstraction on the fragmentation of diuretics in negative electrospray ionization mass spectrometry. J. Am. Soc. Mass Spectrom. 14: 658-670. Tohge, T., Nishiyama, Y., Hirai, M.Y., Yano, M., Nakajima, J., Awazuhara, M., Inoue, E., Takahashi, H., Goodenowe, D.B., Kitayama, M., Noji, M., Yamazaki, M., and Saito, K. (2005), Functional genomics by integrated analysis of metabolome and transcriptome of Arabidopsis plants over-expressing an MYB transcription factor. Plant J. 42: 218-235. Tohge, T., Yonekura-Sakakibara, K., Niida, R., Watanabe-Takahashi, A., and Saito, K. (2007). Phytochemical genomics in Arabidopsis thaliana: a case study for functional identification of flavonoid biosynthesis genes. Pure Appl. Chem. 79: 811-823. Vanholme, R., Storme, V., Vanholme, B., Sundin, L., Christensen, J.H., Goeminne, G., Halpin, C., Rohde, A., Morreel, K., and Boerjan, W. (2012). A systems biology view of responses to lignin biosynthesis perturbations in Arabidopsis. Plant Cell 24: 35063529. Veit, M., and Pauli, G.F. (1999). Major flavonoids from Arabidopsis thaliana leaves. J. Nat. Prod. 62: 1301-1303. Wang, H., Leach, D.N., Thomas, M.C., Blanksby, S.J., Forster, P.I., and Waterman, P.G. (2008). Bisresorcinols and arbutin derivatives from Grevillea banksii R. Br. Nat. Prod. Commun. 3: 57-64. Wolf, S., Schmidt, S., Müller-Hannemann, M., and Neumann, S. (2010). In silico fragmentation for computer assisted identification of metabolite mass spectra. BMC Bioinformatics 11: 148. Woodhead, S., and Cooper-Driver, G. (1979). Phenolic acids and resistance to insect attack in Sorghum bicolor. Biochem. Syst. Ecol. 7: 309-310. Yan, C., Liu, S., Zhou, Y., Song, F., Cui, M., and Liu, Z. (2007). A study of isomeric diglycosyl flavonoids by SORI CID of Fourier transform ion cyclotron mass spectrometry in negative ion mode. J. Am. Soc. Mass Spectrom. 18: 2127-2136. Ye, M., Yan, Y.N., and Guo, D.-a. (2005). Characterization of phenolic compounds in the Chinese herbal drug Tu-Si-Zi by liquid chromatography coupled to electrospray ionization mass spectrometry. Rapid Commun. Mass Spectrom. 19: 1469-1484.

87

Supplemental Data. Morreel et al. Plant Cell (2014). 10.1105/tpc.113.122242.

Yonekura-Sakakibara, K., Tohge, T., Matsuda, F., Nakabayashi, R., Takayama, H., Niida, R., Watanabe-Takahashi, A., Inoue, E., and Saito, K. (2008). Comprehensive flavonol profiling and transcriptome coexpression analysis leading to decoding gene-metabolite correlations in Arabidopsis. Plant Cell 20: 2160-2176.

88

Systematic Structural Characterization of Metabolites in Arabidopsis via Candidate Substrate-Product Pair Networks Kris Morreel, Yvan Saeys, Oana Dima, Fachuang Lu, Yves Van de Peer, Ruben Vanholme, John Ralph, Bartel Vanholme and Wout Boerjan Plant Cell 2014;26;929-945; originally published online March 31, 2014; DOI 10.1105/tpc.113.122242 This information is current as of May 27, 2014 Supplemental Data

http://www.plantcell.org/content/suppl/2014/03/14/tpc.113.122242.DC1.html

References

This article cites 47 articles, 6 of which can be accessed free at: http://www.plantcell.org/content/26/3/929.full.html#ref-list-1

Permissions

https://www.copyright.com/ccc/openurl.do?sid=pd_hw1532298X&issn=1532298X&WT.mc_id=pd_hw1532298X

eTOCs

Sign up for eTOCs at: http://www.plantcell.org/cgi/alerts/ctmain

CiteTrack Alerts

Sign up for CiteTrack Alerts at: http://www.plantcell.org/cgi/alerts/ctmain

Subscription Information

Subscription Information for The Plant Cell and Plant Physiology is available at: http://www.aspb.org/publications/subscriptions.cfm

© American Society of Plant Biologists ADVANCING THE SCIENCE OF PLANT BIOLOGY