Untargeted Metabolomics Profiling of Arabidopsis WT, lbr-2-2 and bak1-4 Mutants Following Treatment with Two LPS Chemotypes

Plants perceive pathogenic threats from the environment that have evaded preformed barriers through pattern recognition receptors (PRRs) that recognise microbe-associated molecular patterns (MAMPs). The perception of and triggered defence to lipopolysaccharides (LPSs) as a MAMP is well-studied in mammals, but little is known in plants, including the PRR(s). Understanding LPS-induced secondary metabolites and perturbed metabolic pathways in Arabidopsis will be key to generating disease-resistant plants and improving global plant crop yield. Recently, Arabidopsis LPS-binding protein (LBP) and bactericidal/permeability-increasing protein (BPI)-related proteins (LBP/BPI related-1) and (LBP/BPI related-2) were shown to perceive LPS from Pseudomonas aeruginosa and trigger defence responses. In turn, brassinosteroid insensitive 1 (BRI1)-associated receptor kinase 1 (BAK1) is a well-established co-receptor for several defence-related PRRs in plants. Due to the lack of knowledge pertaining to LPS perception in plants and given the involvement of the afore-mentioned proteins in MAMPs recognition, in this study, Arabidopsis wild type (WT) and mutant (lbr2-2 and bak1-4) plants were pressure-infiltrated with LPSs purified from Pseudomonas syringae pv. tomato DC3000 (Pst) and Xanthomonas campestris pv. campestris 8004 (Xcc). Metabolites were extracted from the leaves at four time points over a 24 h period and analysed by UHPLC-MS, generating distinct metabolite profiles. Data analysed using unsupervised and supervised multivariate data analysis (MVDA) tools generated results that reflected time- and treatment-related variations after both LPS chemotypes treatments. Forty-five significant metabolites were putatively annotated and belong to the following groups: glucosinolates, hydroxycinnamic acid derivatives, flavonoids, lignans, lipids, oxylipins, arabidopsides and phytohormones, while metabolic pathway analysis (MetPA) showed enrichment of flavone and flavanol biosynthesis, phenylpropanoid biosynthesis, alpha-linolenic acid metabolism and glucosinolate biosynthesis. Distinct metabolite accumulations depended on the LPS chemotype and the genetic background of the lbr2-2 and bak1-4 mutants. This study highlights the role of LPSs in the reprogramming Arabidopsis metabolism into a defensive state, and the possible role of LBR and BAK1 proteins in LPSs perception and thus plant defence against pathogenic bacteria.


Introduction
Lipopolysaccharides (LPSs) as microbe-associated molecular patterns (MAMPs) from Gram-negative bacteria has been shown to trigger innate immune responses in animals and plants [1,2]. While LPS perception in mammals is well-studied, little is known about the role(s) of this MAMP and/or its epitopes and defence signalling in plants. In addition, it is yet to be resolved if the unknown LPS receptor(s) require co-receptors or accessory proteins. Sanabria et al. [3] proposed the involvement of a LPS-responsive N. tabacum S-domain receptor-like kinase (RLK)

Multivariate Data Analysis
To further evaluate the effect of LPS Pst and LPS Xcc on the metabolomes of Arabidopsis WT and mutant (lbr2-2 and bak1-4) plants, as observed from the differential chromatographic profiles of the respective leaf extracts, and to extract valuable biological information from the multidimensional data, chemometric and multivariate data analysis (MVDA) tools were used. The data was processed using MarkerlynxXS TM and exported to 'Soft independent modelling of class analogy' (SIMCA) software for principal component analysis (PCA), hierarchical clustering analysis (HiCA), orthogonal projection to latent structures discriminant analysis (OPLS-DA) and OPLS-DA loading S-plots modelling. The unsupervised PCA modelling is vital for the analysis of multivariate data due to its ability to reduce the dimensionality of data, and reveal natural groupings and trends within and between the data sets [24]. The PCA models were validated using the cumulative modelled variation that explains the variation (R 2 ) and the models' predictive ability (Q 2 ). In addition, HiCA dendrograms were used to support treatment-related sample grouping observed in PCA models. Here, PCA scores plots and HiCA dendrograms revealed time-related grouping of samples treated with LPS Pst and LPS Xcc in the Arabidopsis WT, lbr2-2 and bak1-4 plants. Representative WT ESI (−) data is shown in Figure S2. The differential PCA and HiCA dendrograms sample clustering in WT and mutants supported evidence of metabolic reprogramming of the Arabidopsis metabolome subsequent to different LPS chemotype treatments.
To evaluate the cause of sample grouping observed in PCA, supervised modelling such as OPLS-DA score plots were subsequently used. Additionally, OPLS-DA loading S-plots were included to highlight and extract LPS-induced metabolite 'features' (ions with a specific Rt and m/z), contributing to the group separation observed in OPLS-DA score plots [25]. For example, the Arabidopsis WT OPLS-DA models showed clear separation of control vs. 0, 12, 18 or 24 h treated with LPS Pst (Figure 1A-D for ESI (−) and Figure S3A-D for ESI (+)), as well as the LPS Xcc (Figure 2A-D for ESI (−) and Figure S4A-D for ESI (+)). These models were validated, and an example of receiver operator characteristic (ROC) plots of control vs. 0, 12, 18, and 24 h post-LPS Pst -treated Arabidopsis WT are shown in Figure S5. ROC plots summarised the performance of OPLS-DA as an excellent binary classifier with 100% sensitivity and 100% specificity. Overall, differential separations observed in OPLS-DA models, when comparing control vs. treated samples in both WT and mutants (lbr2-2 and bak1-4), can be linked to changes in the metabolome that may arise from different LPS chemotype treatments. Subsequently, the LPS-induced significant variables of OPLS-DA loading S-plots that contributed to the observed class discrimination in both WT and mutants were annotated and putatively identified.
ESI (+)), as well as the LPSXcc (Figures 2A-D for ESI (-) and S4A-D for ESI (+)). These models were validated, and an example of receiver operator characteristic (ROC) plots of control vs. 0, 12, 18, and 24 h post-LPSPst-treated Arabidopsis WT are shown in Figure S5. ROC plots summarised the performance of OPLS-DA as an excellent binary classifier with 100% sensitivity and 100% specificity. Overall, differential separations observed in OPLS-DA models, when comparing control vs. treated samples in both WT and mutants (lbr2-2 and bak1-4), can be linked to changes in the metabolome that may arise from different LPS chemotype treatments. Subsequently, the LPS-induced significant variables of OPLS-DA loading S-plots that contributed to the observed class discrimination in both WT and mutants were annotated and putatively identified.

Metabolite Annotation and Functional Classification
Forty-five statistically significant LPS-induced metabolites were annotated and putatively identified according to level 2 of the Metabolomic Standard Initiative (MSI) [26], and listed (Table 1 and Table S1) for the metabolites and their respective diagnostic fragmentation patterns, as well as KEGG IDs. The identified metabolites were grouped according to their functional categories: glucosinolates, benzoic-and HCA derivatives, flavonoids, lignans, lipids, oxylipins, arabidopsides, phytohormones, and other metabolite classes (Table 1; Figure 3). Combined, LPS Pst induced accumulation of 35, 27 and 30 significant metabolites in WT, lbr2-2 and bak1-4, respectively, whereas LPS Xcc triggered the accumulation of 32, 23 and 29 significant metabolites in WT, lbr2-2 and bak1-4, respectively. These metabolites were either shared amongst the plants (common) or differentially detected in the WT, lbr2-2 or bak1-4 (Table 1). Similarly, all the metabolite classes identified from LPS Pst -or LPS Xcc -treated Arabidopsis WT, lbr2-2 or bak1-4 were presented as percentages using pie charts ( Figure 3). For instance, with both LPS Pst and LPS Xcc , the glucosinolate metabolite 8-(methylsulphinyl)octyl cyanide ( (only in LPS Xcc ) and 8-methylsulfinyloctyl glucosinolate (glucohirsutin) [#7] (in both LPS chemotypes treatment) were only detected in WT, but not in the two mutant lines. Overall, in WT, lbr2-2 and bak1-4 treated by LPS Pst , the percentage of glucosinolates identified were 14, 11 and 3%, respectively ( Figure 3A-C). In comparison, 22, 13 and 10% of glucosinolates were detected when the different plants were treated with LPS Xcc (Figure 3D-F). Moreover, the percentage of glucosinolates were dominant in WT treated with both LPS chemotypes followed by lbr2-2 and bak1-4, respectively.
Furthermore, Venn diagrams were constructed to visualise the number of LPS-induced metabolites that overlapped or were unique for Arabidopsis WT, lbr2-2 and bak1-4 ( Figure 4). LPS Pst induced seven unique metabolites in the WT compared to one and four observed in lbr2-2 and bak1-4, respectively ( Figure 4A). For the LPS Xcc -treated plants, nine, two, and four metabolites were unique in the WT, lbr2-2 and bak1-4, respectively ( Figure 4B). Eighteen and seventeen metabolites were overlapped in LPS Pst -and LPS Xcc -treated WT, lbr2-2 and bak1-4, respectively. Arabidopsis WT and bak1-4 shared five metabolites in both LPS Pst and LPS Xcc treatments. Similarly, three metabolites were overlapped between lb2-2 and bak1-4 in both LPS Pst and LPS Xcc treatments. Five metabolites were overlapped between WT and lb2-2 after LPS Pst treatment, as compared to one metabolite shared between them after LPS Xcc treatment.
From the classes of identified annotated metabolites, it is evident that the metabolomes of the Arabidopsis WT and mutants (lbr2-2 and bak1-4) were reprogrammed to a defence state upon LPS treatments. Furthermore, the presence and absence of certain metabolites as observed in the WT and mutants suggest that they responded differently to the two LPS chemotype treatments as a consequence of the functional LBR2 and BAK1 proteins in WT, but not in the respective mutants.

Metabolite Heat Map Visualization and Analysis
Significant metabolites from LPS-induced Arabidopsis WT, lbr2-2 and bak1-4 were submitted to the colour-coded heat map module in the MetaboAnalyst web-based tool, to visualise and analyse the time course of relative intensities across control, 0, 12, 18, and 24 h. Notably, the heat maps display the relative intensities of the annotated metabolites across all time points. In the constructed heat maps ( Figure S6 for ESI (+) data and Figure 5 for ESI (−) data), the rows represent the group of the identified metabolites, while the columns represent Arabidopsis plants with their respective LPS chemotype time-related treatments. The colour gradient of dark blue indicates the lowest concentration (relative peak intensity), while deep red indicates the highest concentration. In this study, the annotated metabolites showed a variation of metabolite accumulation as displayed by the average relative intensities (blue and red colour gradient). For the ESI (+) data, high intensities were observed within the lbr2-2 (toward the middle and lower sections for both LPS chemotypes), followed by the bak1-4 (towards the left for LPS Pst and towards the right for LPS Xcc ), and finally to the WT (right side of the heat map). The metabolites on the top half of WT showed higher intensity (mainly glucosinolates) than the bottom, while in the lbr2-2, the bottom half (mainly the LOA and other metabolites) were more intense than the top half (and clearly so compared to the WT), whereas in the bak1-4, intensities were relatively equal in the top and bottom half ( Figure S6). Metabolites identified from WT treated with LPS Pst appeared to have late increases in intensities compared to the plants treated with LPS Xcc that displayed early intensity increases ( Figure S6). A similar pattern was observed in the lbr2-2 and bak1-4 plants, but with higher increased intensities. For the ESI (−) data, similar trends were observed where metabolites on the top half (mainly glucosinolates, benzoic acid and HCA derivatives and flavonoids) of WT had higher intensities than the bottom half. In lbr2-2, the bottom half (mainly the LOA, phytohormones and other metabolites) was higher in intensity than the top half, and notably so compared to the WT. Finally, in the bak1-4, relatively similar intensities were observed on the top and bottom half of the heat map ( Figure 5). Furthermore, Venn diagrams were constructed to visualise the number of LPS-induced metabolites that overlapped or were unique for Arabidopsis WT, lbr2-2 and bak1-4 ( Figure 4). LPSPst induced seven unique metabolites in the WT compared to one and four observed in lbr2-2 and bak1-4, respectively ( Figure 4A). For the LPSXcc-treated plants, nine, two, and four metabolites were unique in the WT, lbr2-2 and bak1-4, respectively ( Figure  4B). Eighteen and seventeen metabolites were overlapped in LPSPstand LPSXcc-treated WT, lbr2-2 and bak1-4, respectively. Arabidopsis WT and bak1-4 shared five metabolites in both LPSPst and LPSXcc treatments. Similarly, three metabolites were overlapped between lb2-2 and bak1-4 in both LPSPst and LPSXcc treatments. Five metabolites were overlapped between WT and lb2-2 after LPSPst treatment, as compared to one metabolite shared between them after LPSXcc treatment.  From the classes of identified annotated metabolites, it is evident that the metabolomes of the Arabidopsis WT and mutants (lbr2-2 and bak1-4) were reprogrammed to a defence state upon LPS treatments. Furthermore, the presence and absence of certain metabolites as observed in the WT and mutants suggest that they responded differently to the two LPS chemotype treatments as a consequence of the functional LBR2 and BAK1 proteins in WT, but not in the respective mutants.

Metabolite Heat Map Visualization and Analysis
Significant metabolites from LPS-induced Arabidopsis WT, lbr2-2 and bak1-4 were submitted to the colour-coded heat map module in the MetaboAnalyst web-based tool, to visualise and analyse the time course of relative intensities across control, 0, 12, 18, and 24 h. Notably, the heat maps display the relative intensities of the annotated metabolites across all time points. In the constructed heat maps ( Figure S6 for ESI (+) data and Figure  5 for ESI (-) data), the rows represent the group of the identified metabolites, while the columns represent Arabidopsis plants with their respective LPS chemotype time-related treatments. The colour gradient of dark blue indicates the lowest concentration (relative  Overall, the observed variation in intensity from the heat map analysis further revealed the differential changes in the metabolome of the Arabidopsis WT and mutants (lbr2-2 and bak1-4) after different LPS chemotype treatments. More so, increased intensities observed in the LPS-treated mutants, especially the LOA metabolites group of lbr2-2, indicate an inducible response to the LPS chemotypes. The nature of these metabolomic responses seems different when compared to that of the WT as reflected by the greater peak intensities of defense-associated metabolites. This might suggest the presence of other LPS-interacting proteins/co-receptors other than LBR2/BAK1, or possibly the participation/complementation by close homologues such as LBP/BPI related-1 (LBR1)/BAK1-like 1 (BKK1).
ties observed in the LPS-treated mutants, especially the LOA metabolites group of lbr2-2, indicate an inducible response to the LPS chemotypes. The nature of these metabolomic responses seems different when compared to that of the WT as reflected by the greater peak intensities of defense-associated metabolites. This might suggest the presence of other LPS-interacting proteins/co-receptors other than LBR2/BAK1, or possibly the participation/complementation by close homologues such as LBP/BPI related-1 (LBR1)/BAK1like 1 (BKK1).  Table 1), while the columns represent Arabidopsis plants with their respective LPS chemotype time-related treatments. The colour gradient of dark blue indicates lowest intensity, while deep red indicates highest intensity. Metabolites are numbered as in Table 1. Abbreviations: GLS = glucosinolates; BHCAs = benzoic-and hydroxycinnamic acid derivatives; FL = flavonoids; LOA = lipids, oxylipins and arabidopsides; PhH = phytohormones.

Pathway Mapping and Quantification
To elucidate various pathways associated with the identified LPS-induced metabolites in WT, lbr2-2 and bak1-4 mutants, the Metabolomic Pathway Analysis (MetPA) tool, a component of the MetaboAnalyst suite, was used. Notably, MetPA mapped only metabolites with KEGG IDs and graphically presented the data as the metabolome view and a table that contains all the altered/perturbed pathways ( Figure 6A-C; Tables S2-S4).
In WT, lbr2-2 and bak1-4 treated with LPS Pst and LPS Xcc , alpha-linolenic acid metabolism (#1) was the most significantly altered pathway followed by the phenylpropanoid biosynthesis (#2). The flavone and flavanol biosynthesis pathway (#3) and glucosinolate biosynthesis (#4) were affected/induced in the LPS-treated WT, not in the lbr2-2 and bak1-4 mutants. The rest of the enriched pathways perturbed by LPS in both Arabidopsis WT and mutants are listed in Tables S2-S4. More so, the bar graphs ( Figures S7-S10) display the relative intensities of the mapped metabolites (and some unmapped metabolites) of affected pathways across all time points. These results, as elaborated below, suggest that Arabidopsis uses different pathways to reprogramme its metabolome into a defensive state when challenged by LPS MAMPs. The relative quantification of the average intensities of metabolites from the selected four pathways: alpha-linolenic acid metabolism (#1), phenylpropanoid biosynthesis (#2), flavone and flavanol biosynthesis pathway (#3), glucosinolate biosynthesis (#4), are shown ( Figures S7-S10).
To elucidate various pathways associated with the identified LPS-induced metabolites in WT, lbr2-2 and bak1-4 mutants, the Metabolomic Pathway Analysis (MetPA) tool, a component of the MetaboAnalyst suite, was used. Notably, MetPA mapped only metabolites with KEGG IDs and graphically presented the data as the metabolome view and a table that contains all the altered/perturbed pathways ( Figure 6A-C; Tables S2-S4).  Metabolite KEGG IDs from both LPS chemotype-treated WT, lbr2-2 or bak1-4 were mapped to pathways as displayed. (A-C) represents the metabolome view for the WT, lbr2-2 and bak1-4, respectively. The graphs were arranged such that the p-values from the pathway analysis are on the y-axis and the impact values from the pathway topology analysis on the x-axis. The colour gradient of the circles (pathways) from light yellow to dark red indicates lower to higher significance, respectively. Pathways are numbered #1: Alpha-linolenic acid metabolism, #2: Phenylpropanoid biosynthesis, #3: Flavone and flavanol biosynthesis and #4: Glucosinolate biosynthesis. Notably, pathways #3 and #4 were not mapped in the lbr2-2 and bak1-4 mutants. Figure S7A,B show the alpha-linolenic acid metabolism mapped metabolites, 13S(S)-HPOTrE and 12-OPDA with the relative intensities. The 13S(S)-HPOTrE had increased intensity in the WT treated with LPS Pst and LPS Xcc . There were increased intensities of 13S(S)-HPOTrE in lbr2-2 with reduced intensities at 12 h LPS Pst treatment ( Figure S7A). The 13S(S)-HPOTrE in bak1-4 decreased in intensity after 0 h, but picked up at 24 h when treated by LPS Pst . On the other hand, when treated with LPS Xcc , there was a steady increase in intensity from 12 to 24 h. In all, there was an overall greater intensity of 13S(S)-HPOTrE in WT than in the mutants, when comparing the control vs. treatments. For 12-OPDA identified in WT, there was an increase in intensities across treatments with LPS Pst and LPS Xcc , with the highest intensities observed at 24 and 18 h, respectively ( Figure S7B). In lbr2-2, there was an increase in 12-OPDA intensities at 18 to 24 h after treatment by LPS Pst . On the other hand, the LPS Xcc treatment showed a steady increase in intensities of 12-OPDA at early and late time points. The 12-OPDA in bak1-4 plants showed mainly late accumulation in both LPS Pst and LPS Xcc treatments. Overall, 12-OPDA identified in lbr2-2 appeared to have the highest intensities followed by WT and bak1-4, across all treatments. The second altered pathway was phenylpropanoid biosynthesis with the analysed intensities of the two mapped metabolites (sinapic acid and sinapoyl malate) ( Figure S8A,B). The intensity of another identified phenylpropanoid metabolite sinapoyl 1-O-sinapoyl-beta-D-glucose that was not mapped by MetPA was included in the analysis ( Figure S8C). Figure S8A shows that the relative intensities of sinapic acid identified from WT after LPS Pst treatment increased exponentially, but decreased when treated with LPS Xcc . A similar trend was observed in lbr2-2 plants but with reduced relative intensities. The relative intensities of metabolites from bak1-4 after LPS treatment appeared to have similar intensities across the time points ( Figure S8A). For sinapoyl malate ( Figure S8B), there was an increase in intensity in both early and late responses after LPS Pst treatment, but a reduced late response at 24 h LPS Xcc treatment. In lbr2-2, sinapoyl malate had both early and late increased intensities with a slight reduction when treated with LPS Xcc at 12 h. Sinapoyl malate from bak1-4 also had both early and late increased intensities with a slight reduction when treated with LPS Xcc at 12 h. Furthermore, the glucosinolate biosynthesis pathway mapped glucoerucin and glucobrassicin metabolites and their relative intensities are shown in Figure S10A,B, respectively. In addition, the intensities of MetPA unmapped glucosinolates such as glucohirsutin and glucosinolate degradation product 8-(methylsulphinyl)octylamine were analysed in Figure S10C,D, respectively. Glucoerucin identified in LPS Xcc -treated WT plants showed a slight decrease in intensities across time points when treated with LPS Pst , but with a slight increase at 12 h and 24 h with LPS Xcc . For lbr2-2, there was over a 2-fold decrease of intensities across time points in both LPS Pst and LPS Xcc -treated plants. Glucoerucin from bak1-4 showed decreased and similar intensities in both LPS chemotypes treatments. For glucobrassicin ( Figure S10B), there were decreased intensities in the WT treated with LPS Pst and LPS Xcc across time points, except at 24 h LPS Xcc treatment where there was, relatively, over a 1.5-fold increase in intensity. In lbr2-2, glucobrassicin showed over a 1.6-and 2.5-fold decrease in intensity when treated with LPS Pst and LPS Xcc , respectively. Glucobrassicin in bak1-4 treated with LPS Pst and LPS Xcc showed decreased and similar intensities when comparing the control vs. treatment. Glucohirsutin ( Figure S10C), was increased at 12 h and 24 h LPS Pst and LPS Xcc -treated WT, but decreased in both lbr2-2 and bak1-4. The glucosinolate degradation product 8-(methylsulfinyl)octylamine was accumulated in both WT, lbr2-2 and bak1-4 after both LPS chemotypes treatments, with lesser accumulation observed in LPS Pst -treated bak1-4 ( Figure S10D).
Overall, LPS chemotypes altered the glucosinolate biosynthesis pathway and led to more accumulation of glucosinolates and their products in the WT as compared to the mutants.

Discussion
LPSs found on the outer membrane of Gram-negative bacteria is one of the MAMPs associated with bacterial adhesion and induction of defence responses in animals and plants [28]. The structure of LPS is made up of lipid A, core oligosaccharides and an O-polysaccharide chain, which have been reported to induce defence responses in plants individually or collectively [4,29,30]. To date, the plant pattern recognition receptor (PRR) and/or co-receptor involved in LPS perception, as well as the molecular mechanism underlying the defence signalling, remains elusive. Sanabria et al. [3] have proposed the involvement of an LPS-responsive N. tabacum S-domain RLK (Nt-Sd-RLK) gene encoding conserved B-lectin, S-and PAN domains in LPS perception. Additionally, a recent study implicated Arabidopsis AtLBR-1 and AtLBR-2 in the LPS perception and defence responses [4].
In this study, untargeted metabolomics were used to profile the metabolite changes that might occur after treatment of Arabidopsis WT and mutant (lbr2-2 ad bak1-4) plants with different LPS chemotypes from Pseudomonas syringae pv. tomato DC3000 (Pst) and Xanthomonas campestris pv. campestris 8004 (Xcc). Advancement in liquid chromatographymass spectrometry (LC-MS) has improved the use of metabolomics in the study of MAMP-treated plants [11,12,31]. Here, chromatographic and spectrometric changes were observed in extracts from both Arabidopsis WT and the mutants after treatments with LPS Pst and LPS Xcc . Such perturbations can be linked to the metabolome changes caused by LPS chemotypes treatments. The 12-24 h period selected in this study has been shown to be sufficient to capture the metabolome changes that may occur and possibly return the plant system to homeostasis [12,31]. The chemometric modelling tools such as PCA and OPLS-DA play vital roles in the observed sample grouping and accompanying metabolite identification that reflects on the evidence of differential metabolic reprogramming that is related to the response of Arabidopsis plants to the LPS chemotypes. There were broad classes of annotated metabolites involved as significant response markers, including glucosinolates, benzoic acids and HCAs, flavonoids, lignan, lipids, oxylipins, arabidopsides and phytohormones, that unravelled the complexity of the plant metabolomes and the implication of LPS in the Arabidopsis defence responses.

Glucosinolates Accumulation Following LPSs Treatment
Glucosinolates are a sulphur-rich, structurally diverse class of secondary metabolites that are richly found in Brassicaceae such as Arabidopsis, with various roles that include response against plant pathogens [32]. In this regard, the glucosinolate-myrosinase defence system is activated upon pathogen or insect attack in plants. The glucosinolate breakdown activity of myrosinase leads to the production of various degradation products that are involved in the defence activities [33]. In this study, we annotated and putatively identified a number of glucosinolates and their products in both Arabidopsis WT and mutants (lbr2-2 and bak1-4) ( Table 1). However, WT treated with LPS Pst and LPS Xcc accumulated more glucosinolates compared to the lbr2-2 and bak1-4. Similar trends were observed when glucosinolates (e.g., glucoerucin and glucobrassicin) were mapped by MetPA. Consistently, there were higher qualitative accumulation of glucoerucin, glucobrassicin and glucohirsutin when comparing control vs. LPS treatments in the WT vs. lbr2-2 and bak1-4 ( Figure S10). While glucosinolates were identified only in the WT, glucosinolate degradation products were identified in both WT and mutants. For instance, glucosinolate degradation product 8-(methylsulfinyl)octylamine accumulated in both LPS Pst -and LPS Xcc -treated WT, lbr2-2 and bak1-4. Robin et al. [34] reported increased accumulations of both glucoerucin (e.g., of aliphatic glucosinolate) and glucobrassicin (e.g., of indole glucosinolate) in the resistant cabbage line BN4303 infected by Leptosphaeria maculans that causes blackleg disease in Brassica crops. Furthermore, aliphatic and aromatic glucosinolates enhanced resistance against necrotrophic bacteria Erwinia caratovora and hemi-biotrophic bacteria P. syringae, respectively, in Arabidopsis [35]. These studies highlight the significant role of glucosinolates in plant defence against pathogens. In our group, several glucosinolates including glucohirsutin, glucoerucin, glucobrassicin were putatively annotated from Arabidopsis treated by LPS from B. cepacia [11,12]. The results obtained in this study (both the annotated metabolites, heat map and MetPA analyses) suggested that different LPS chemotypes induced significant accumulation of glucosinolates in the WT compared to the mutants. Moreover, the variation in the accumulation of glucosinolates and their degradation products in the WT compared to the mutants may be caused by the knock out of the LBR2 and BAK1 proteins.

Hydroxycinnamic Acid Derivatives Accumulation Post-LPSs Treatment
HCAs and their derivatives are indispensable in the protection of plants against pathogens, ultraviolet B irradiation and central to the synthesis of secondary metabolites such as flavonoids, tannins and lignin [36,37]. The synthesis of HCAs and derivatives from phenylalanine through the phenylpropanoid pathway is well-documented in plants, including the Arabidopsis model system [38]. In this study, LPS Pst and LPS Xcc triggered the differential accumulation of metabolites that belong to the HCA derivatives such as 6,7-dimethoxycoumarin, sinapic acid and sinapoyl malate in WT and the mutants. For example, 1-O-sinapoyl-beta-D-glucose was detected in WT and bak1-4, but not in the lbr2-2, suggesting its differential accumulation that may be linked to the absence of LBR2 protein.
Sinapic acid may be found in free form or esterified like other HCAs. These esters can be sugar conjugates (glycosides) such as sinapoyl glucose (identified in this study), or conjugates of other compounds, such as sinapoyl malate (identified in this study) and sinapoyl choline or sinapine (a source of sinapic acid and choline in germinating mustard seed) [39][40][41][42]. Tinte et al. [11] also identified sinapic acid, sinapoyl glucose and sinapoyl malate in Arabidopsis after treatment with LPS from B. cepacia Furthermore, the antibacterial activity of sinapic acid against both Gram-positive and Gram-negative bacteria have been reported in mustard seeds [43,44]. Sinapoyl glucose was one of the infection markers identified in the apoplasts of Arabidopsis leaves infected by Verticillium longisporum [45]. An Arabidopsis mutant deficient in sinapoyl malate or sinapic acid affected the accumulation of sinapic esters and disrupted the phenylpropanoid pathway and eventual lignin biosynthesis [40,46]. Heat map ( Figure 5) and MetPA analysis ( Figure 6) showed that the phenylpropanoid biosynthesis pathway was altered with variations in the accumulation of sinapic acid, sinapoyl malate and sinapoyl 1-O-sinapoyl-beta-D-glucose when the control vs. LPS treatments were compared ( Figure S8). These results showed increased accumulation of HCAs and derivatives in WT compared to the mutants (lbr2-2 and bak1-4), further highlighting a possible influence of different LPS chemotypes and lack of functional LBR2-2 and BAK1-4 proteins in the perturbation observed in the Arabidopsis metabolome.

Flavonoids Accumulation Post-LPSs Treatment
Flavonoids are secondary metabolites produced in plants through the phenylpropanoid pathway with a variety of functions that include defence against environmental stresses, pigmentation, auxin transport, signalling, feeding deterrents and protection against UV light [47][48][49]. As shown in Table 1, more flavonoids were accumulated in WT followed by bak1-4, with none identified in the lbr2-2. Flavonoids accumulated mostly at 24 h post-LPS treatment in the WT suggesting a late flavonoid response in Arabidopsis. The absence of significant flavonoids in the lbr2-2 suggests a possible role of LBR in the LPS perception Arabidopsis. MetPA analysis (Figure 6) showed the mapping of kaempferol 3-O-rhamnoside-7-O-glucoside in the altered flavone and flavonol biosynthesis pathway. This LPS-induced kaempferol 3-O-rhamnoside-7-O-glucoside and unmapped flavonoids such as afzelin and kaempferitrin displayed distinct variations in quantitative intensities, suggesting differential metabolic accumulation in the WT and mutants. Kaempferol is involved in the defence response of Arabidopsis against the Cucumber mosaic virus containing satellite RNA (CMVsat) infection [50]. Kaempferitrin is reported as the key flavonoid that inhibits the polar auxin transport (PAT) and thereby regulates growth and development in Arabidopsis shoots [51]. A study of prenylated flavonoids showed high free radical scavenging activities that contribute to their use as potent anti-oxidants [52].

Lignan Accumulation Following LPSs Treatment
Lignans are synthesised from HCAs through the early phenylpropanoid pathway with precursors such as coniferyl alcohol, sinapyl alcohol and 4-hydroxycinnamyl alcohol [53]. Lignan biosynthesis is related to but different from that of phenylpropanoids such as lignins, norlignans and neolignans [54][55][56]. There is evidence of bacteria and fungi triggering accumulation of lignin and lignan, leading to the protection of plants against pathogen invasion [46,57]. Cinnamyl alcohol dehydrogenase (CAD) genes that code for the enzyme that catalyses the final step in the biosynthesis of monolignol have been shown to accelerate lignin and lignan biosynthesis and deposition are induced following pathogen infection [58]. Lignans such as G(8-O-4)G hexoside, lariciresinol hexoside and G(8-5)FA malate were identified in response to LPS treatment in this study ( Table 1).
The fungal pathogen V. longisporum induced accumulation of phenylpropanoid metabolites including lignans such as the syringaresinol glucoside (a sinapyl alcohol-derived lignan), secoisolaricoresinol, pinoresinol glucoside and lariciresinol glucoside (coniferyl alcohol-derived lignans) in Arabidopsis [46]. G(8-O-4)G hexoside, G(8-5)FA malate and lariciresinol hexoside identified in this study were reported to accumulate in Arabidopsis treated with silver nanoparticles [59]. In this study, the accumulation of lignans upon LPS treatment was higher in the WT followed by bak1-4 and decreased in lbr2-2 (from heat map analysis). This supports the above suggestions that different LPS chemotypes are able to trigger this metabolic pathway, as well as that the BAK1 and LBR2 proteins play a possible regulatory role in the accumulation of lignan. These findings further correlate with existing reports that link lignans in plant defence against biotic and abiotic stresses.

Lipids, Oxylipins and Arabidopsides Accumulation Post-LPSs Treatment
Lipids are involved in the establishment of membrane interfaces as well as in the regulation of intracellular signalling during plant-microbe communication [60,61]. Oxidation products of polyunsaturated fatty acids (PUFAs) known as oxylipins function as signal molecules in the regulation of plant growth, development, and biotic and abiotic responses [60,62]. Fatty acid derivatives such as 9,12,13-trihydroxy-10,15-octadecadienoic acid and 9,12,13-TriHOME detected in this study have been identified from the Arabidopsis root exudate profile [63] and the apoplast of Arabidopsis infected by V. longisporum [45]. Tinte et al. [11] reported accumulation of 13(S)-HPOTrE, 12-OPDA and Arabidopside A in Arabidopsis leaves treated by LPS from P. syringae. Arabidopsides are named based on the position at which OPDA is found esterified to the mono-or digalactosyl diacylglycerol (MGDG or DGDG), instead of the fatty acyl moiety. For instance, Arabidopside A (identified here) is a MGDG derivative containing OPDA esterified at position sn-1, while Arabidopside C is a DGDG containing dn-OPDA esterified at position sn-2 [64][65][66]. Oxylipin, 12-OPDA is a precursor for phytohormone jasmonic acid (JA) and related jasmonates such as methyl jasmonate (MeJA) and jasmonyl-L-isoleucine (JA-IIe), which are involved in plant defence responses and growth [64]. Arabidopside D (detected in this study) has previously been identified in the aerial parts of Arabidopsis with a characteristic inhibitory effect on cress root growth [67]. Additionally, V. longisporum induced accumulation of 12-OPDA and dinor-OPDA as signatory infection markers in the apoplast of Arabidopsis [45].
Our MetPA analysis implicated 13(S)-HPOTrE and 12-OPDA as mapped metabolites in the enriched alpha-linolenic acid metabolic pathway ( Figure 6 and Figure S7). Both metabolites are precursors of JA with evidence in the regulation of plant defence signalling against pathogens [66]. There were distinct relative intensity changes of both 13(S)-HPOTrE and 12-OPDA metabolites when comparing control vs. LPS Pst--and LPS Xcc -treated WT, lbr2-2 and bak1-4. Overall heat map analysis showed LPS-induced more LOA metabolite group in lbr2-2, followed by bak1-4 and WT. This alteration of metabolite intensities, especially LOA in mutants (mainly in lbr2-2), suggests that the knock-out of LBR2 and BAK1 appeared to enhance the accumulation of this group of metabolites upon LPS treatment. Overall, the result indicates that the alpha-linolenic acid metabolism is differentially altered in the Arabidopsis WT and mutants treated by different LPS chemotypes, with the WT exhibiting stricter metabolic control over LPS inducible responses/pathways.

Phytohormones Accumulation Following LPSs Treatment
Salicylic acid (SA) accumulates in both local and systemic defence responses against pathogens where it mediates disease resistance in both the infected and distal leaves [68,69]. SA can be synthesised through the isochorismic acid (IC) or phenylalanine ammonia lyase (PAL) pathways, using shikimic acid as a precursor in both routes [70]. Free SA level can be regulated through conversion into alternative forms by chemical modification such as methylation (e.g., methyl SA, MeSA), glycosylation (e.g., SA glucoside, SAG), and hydroxylation to form dihydroxybenzoic acids. SAG is stored in the vacuole and can be reconverted to SA when needed by the plant. SAG was identified as a disease infection marker in Arabidopsis treated with V. longisporum [45]. Furthermore, B. cepacia LPS induced accumulation of conjugated SA and systemic acquired resistance (SAR) against Pst in Arabidopsis [71].
In this study, SAG accumulated in both LPS chemotype-treated bak1-4, but not in the WT or lbr2-2. Two isomers of the benzoic acid precursors of SA such as 2,5-dihydroxybenzoic acid pentoside isomer I and II were identified in this study. While 2,5-dihydroxybenzoic acid pentoside isomer I accumulated in both LPS chemotypes-treated WT, lbr2-2 and bak1-4, it was detected earlier in bak1-4 only. The accumulation of both SAG and 2,5-dihydroxybenzoic acid pentoside isomer II in LPS-treated bak1-4 suggests that inactive SA was more abundant in bak1-4 mutant than WT and lbr2-2. While SA is involved in the defence response against biotrophic pathogens, JA regulates infection of necrotrophic pathogens or herbivorous insects, respectively, with some degree of antagonism reported between the two phytohormones [72,73].
Another phytohormone, abscisic acid (ABA), was detected in Arabidopsis treated by LPS in this study and in the study by Tinte et al. [11]. ABA accumulated in both LPS chemotypes-treated WT, lbr2-2 and bak1-4 with increased intensity observed in the mutants (lbr2-2 and bak1-4) compared to the WT. ABA is a sesquiterpene signalling molecule that regulates pathways involved in plant responses to abiotic stresses, plant growth and development [72]. An antagonistic interplay between ABA and the JA-ethylene signalling pathway has been shown to regulate expression of plant defence genes and resistance to diseases [74].
For genotyping, a two-step PCR genotyping assay, as reported in [75], with minor modifications was used to select mutant plants that were homozygous for the transferred (T)-DNA insert, and also according to the guideline tools of the Salk Institute Genome Analysis Laboratory (SIGnAL) (http://signal.salk.edu/tdnaprimers.2.html) accessed on 25 June 2018 (data not shown).

LPS Isolation and Characterisation
LPSs from Pst and Xcc were isolated using the hot phenol-water procedure [76], and characterised for suitable purity, as described in Tinte et al. [11] (data not shown).

Plant Treatment and Harvesting
LPS (dissolved to a final concentration of 100 µg/mL in sterile 2.5 mM magnesium chloride (MgCl 2 )) from Pst and Xcc was used to treat Arabidopsis WT and mutant (lbr2-2 and bak1-4) plants by gentle pressure-infiltration with a blunt syringe through the abaxial side of the leaf. Three different plants were used for one biological replicate out of a total of three, and the leaves were harvested after 0, 12, 18, and 24 h post-treatments, frozen in liquid nitrogen to quench metabolic activity, and stored at -80 • C until use. Negative control plants were treated with sterile 2.5 mM MgCl 2 to account for responses that may be generated by the MgCl 2 solution (though it has been shown not to trigger any secondary effects [11,12]) used to dissolve the LPS, as well as possible wounding upon LPS treatment.

Metabolite Extraction and Sample Preparation
A methanol (MeOH)-based extraction method that extracts semi-polar and some non-polar metabolites was used to extract metabolites from the leaf tissue. Hot MeOH (Romil SpS, Cambridge, UK) (65 • C) was used to quench myrosinase activity since this enzyme degrades glucosinolate secondary metabolites that play vital roles in plant defence against environmental stresses including pathogens [12]. Hot 80% MeOH (1:10 w/v) was added to 1 g frozen leaf material in 50 mL Falcon tubes and homogenised with an Ultraturrax homogeniser (IKA, Staufen, Germany) for 2 min. For extra lysis of cells, the homogenate fraction was sonicated with a sonication probe (Sonopuls, Berlin, Germany) set at 100% power, for 30 s. The mixture was centrifuged at 5 × 300 g for 20 min at 4 • C and the supernatants were carefully transferred into new Falcon tubes. The supernatants were evaporated to approximately 1 mL at 55 • C using a Büchi Rotavapor R-200, transferred to 2 mL Eppendorf tubes, and then further evaporated to complete dryness using a dry bath at 55 • C in a fume hood. Dried residue was re-suspended in 500 µL of 50% ultrahigh performance liquid-chromatography (UHPLC)-grade MeOH: milliQ water (1:1 v/v), filtered through 0.22 µm nylon syringe filters into HPLC glass vials fitted with 500 µL inserts and stored at 4 • C until ready for analysis. The quality control (QC) samples were prepared by pooling and mixing aliquots of equal volumes from all samples and included in the run order to monitor the stability of the samples and the instrument. Blank samples (containing 50% MeOH used to re-suspend the samples) were included in the run to gauge sample variance and monitor potential sample carryover.

Liquid Chromatography-Mass Spectrometry Analysis
Several LC-MS and hybrid analytical methods have shown improved sensitivity, speed and repeatability, and therefore have been used successfully in the analysis and detection of metabolome changes stemming from both biotic and abiotic stresses in plant systems [77]. In this study, samples were analysed using an Acquity UHPLC system coupled in tandem to a Synapt G1 high-definition (HD) quadrupole time-of-flight (qTOF) mass spectrometer (UHPLC-qTOF-MS) system (Waters Corporation, Manchester, UK) fitted with an Acquity HSS T3 C18 reverse-phase column (1.7 µm, 150 mm × 2.1 mm) (Waters Corporation, Milford, MA, USA). Four microliters of each sample were injected into a column that is housed in a column oven thermostatted at 60 • C. Analysis was conducted with a binary solvent system (solvent A: milliQ water and 0.1% formic acid (Sigma Aldrich, Munich, Germany); solvent B: acetonitrile (Romil SpS, Cambridge, UK) and 0.1% formic acid) at a constant flow rate of 0.4 mL/min for 30 min. The separation conditions were: 2% B over 0.00-2.00 min, increased to 90% B over 1.00-25.00 min, increased to 95% B over 25.00 min-25.10 min, held at 95% B over 25.10-27.00 min, decreased back to 2% B over 27.00-28.00 min, and held at 2% B over 28.00-30.00 min. Separated analytes were detected using a photodiode array (PDA) detector scanned from 200-500 nm with 1.2 nm resolution and a sampling rate of 20 spectra/s. The chromatographic effluent was further analysed and detected using the MS component of the SYNAPT G1 high definition qTOF-MS, used in V-optics and operated in the electrospray ionisation positive (ESI+) and negative (ESI-) modes. The MS conditions were also set up: a capillary voltage of 2.5 kV, sampling cone voltage at 30 V, trap collision 3 V and the extraction cone at 4 V, scan time of 0.1 sec covering the 50 to 1200 Da mass-to-charge (m/z) range, source temperature of 120 • C, desolvation temperature of 450 • C. N 2 gas was used as the nebulisation gas at a flow rate of 800 L/h. Leucine encephalin (50 pg/mL) was used as the reference mass calibrant to obtain typical mass accuracies between 1-3 mDa. The MS analysis file was set up in (MS E ) mode to perform unfragmented as well as five fragmenting experiments simultaneously by in-source collision energy from 3-40 eV.
To ensure experimental reproducibility, three independent biological replicates and three instrumental technical replicates were analysed for each time point in a randomised fashion. This generated n = 9, required for the multivariate data analysis of metabolomics data [26]. As mentioned, blank and QC samples were randomly included in the run in order to monitor instrument reproducibility.

Multivariate Data Analysis
UHPLC-qTOF-MS raw data was visualised and processed using MarkerLynx XS TM software (Waters, Manchester, UK). Parameters set to analyse ESI (+) mode data were: retention time (Rt) range of 1.1-24.1 min of the MS chromatogram, mass range of 100-1200 Da, Rt window of 0.3 min and mass tolerance of 0.05 Da. Those of ESI (−) mode data were: retention time (Rt) range of 0.9-24.0 min of the MS chromatogram, mass range of 100-1200 Da, Rt window of 0.3 min and mass tolerance of 0.05 Da. Peak detection and alignment were achieved using the MarkerLynx patented ApexPeakTrack algorithm. Normalisation was done using total ion intensities of each defined peak, while Savitzky-Golay smoothing and integration was done before computing intensities. Only data matrices with noise levels less than 50% were selected for downstream data analysis.
After data pre-processing and pre-treatment steps, the obtained data matrix of retention time (Rt)-m/z variable pairs, with m/z peak intensity for all samples, were exported to 'Soft independent modelling of class analogy' (SIMCA) software, version 15 (Sartorius, Umea, Sweden) for statistical modelling. The first statistical modelling used were unsupervised principal component analysis (PCA) and hierarchical clustering analysis (HiCA) dendrogram models that show trends, clusters, differences and similarities between samples. Thereafter, the statistical modelling used was a supervised model, orthogonal projection to latent structures discriminant analysis (OPLS-DA), which predefines the sample groups to observe the extent to which the variables affect the separation between groups. Selected discriminant ions for downstream metabolite identifications had a correlation [p(corr)] of ≥ 0.5 and covariance of (p1) ≥ 0.05. For all statistical modelling, Pareto-scaling was used to reduce the mask effect from the more abundant metabolites and, by doing so, enhance the model's predictive ability [78]. The different model validations used to gauge model quality included evaluation of the model goodness-of-fit and predicted variation (R 2 and Q 2 ), and receiver operator classifier (ROC) plots. The OPLS-DA models were also validated using analysis of variance of the cross-validation predictive residuals (CV-ANOVA), with a p-value of < 0.05 indicating a good model [79]. Only statistically valid models were selected and used in data mining for metabolite annotation.

Metabolite Annotation and Biological Interpretation
The statistically significant metabolite annotation was performed according to level 2 of the Metabolomics Standard Initiative (MSI) [26]. The accurate mass, fragmentation patterns, elemental composition, Rt and online database searches were used to facilitate assigning the correct chemical structure to each detected feature. Using Markerlynx software, fragment ion peaks were used to obtain elemental compositions which were searched on databases for metabolites annotation. Online databases used for metabolite identification were LIPID MAPS [80], MetaCyc [81] ChemSpider [82] and Dictionary of Natural Products [83]. Mass spectral information was also compared to available published literature and metabolites were accordingly annotated (putative identification).
MetaboAnalyst version 5.0 (https://www.metaboanalyst.ca/) accessed on 12 August 2021, a web-based tool for the analysis of metabolomics data sets, was used for further metabolic data analysis and functional interpretation for heat map visualisation and pathway mapping [84]. Peak intensities of metabolomic data were imported into MetaboAnalyst, transformed and Pareto-scaled to reduce systemic variance within the features. The average integrated peak intensities of the annotated metabolites were used to construct the heat maps with the statistical analysis software available on MetaboAnalyst (https: //www.metaboanalyst.ca/) accessed on 12 August 2021. The group averages were used to show metabolite intensities from WT and mutants (lbr2-2 and bak1-4) under different LPS chemotype treatments for all time points. Pathway analysis (PA) was performed by the MetPA (Metabolomics Pathway Analysis), a component of the MetaboAnalyst suite that uses the Kyoto Encyclopedia of Genes and Genomes (KEGG) metabolic pathways as the backend knowledge base. For the PA, KEGG IDs of the annotated metabolites were submitted to the MetaboAnalyst and significant mapped pathways were generated.

Conclusions
LPSs as a MAMP triggers defence responses in mammals and plants. The LPS perception and signalling in mammals are established, whereas in plants the PRR and/or co-receptor complex and the associated defence signalling are still unresolved. Understanding the metabolite profiles and perturbed metabolic pathways in WT Arabidopsis and mutants (lbr2-2 and bak1-4) treated with LPSs will provide useful insight into the mechanism underlying plant defence responses against pathogens. The aim of the study was to investigate how the lbr2-2 and bak1-4 Arabidopsis mutants respond to LPS treatment in comparison to the WT, using untargeted LC-MS-based metabolomics. In this regard, two LPS chemotypes, originating from P. syringae and X. campestris (LPS Pst and LPS Xcc ) that differ in the chemical 'molecular patterns' of the LPS sub-moieties, were applied. Both chemotypes were found to induce differential reprogramming of the metabolome in support of an enhanced defensive state in the WT, as well as in the lbr2-2 and bak1-4 mutants. Most of the annotated metabolites belong to the classes of 'glucosinolates', 'benzoic-and HCA derivatives', 'flavonoids', 'lignans', 'lipids, oxylipin and arabidopsides' and 'phytohormones'. The presence and absence of certain metabolites as observed in the WT and mutants suggest that they responded differently to the two LPS chemotype treatments as a consequence of the functional LBR2 and BAK1 proteins in WT, but not in the respective mutants.
Furthermore, the Metabolomic Pathway Analysis (MetPA) of significantly annotated metabolites from WT, lbr2-2 and bak1-4 showed distinct altered pathways that include flavone and flavanol biosynthesis, phenylpropanoid biosynthesis, alpha-linolenic acid metabolism, and glucosinolate biosynthesis. The dissimilar metabolite profiles and metabolic responses (as seen in the heat maps) observed in WT and mutant plants after treatment with the two LPS chemotypes may be attributed to the distinct structures of the tripartite LPS chemotypes. In addition, the involvement of functional LBR2 and BAK1 proteins in the WT, but not in the respective mutants, could have contributed to the different metabolic responses of WT vs. lbr2-2 vs. bak1-4. Moreover, the increased accumulation of defence-associated metabolites observed in the mutants indicates an inducible response to the LPS chemotypes (albeit different from that of the WT), indicating prior perception of the LPSs and subsequent signal processing. This might be attributable to a level of redundancy in LPS perception and point to the presence of other LPS-interacting/co-receptors that are participating/complementing the perception of intact LPS (or sub-moieties thereof) and resulting signalling in the mutants. Untargeted metabolomics was thus found to generate a wealth of information in monitoring plant biological systems such as profiling of differential metabolite accumulation as part of the adaptation of the host's 'defensome' in response to pathogenic stresses, and these metabolites may thus serve as biomarkers in LPS:plant interaction studies.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/metabo12050379/s1. Figure S1: Representative UHPLC-MS BPI chromatograms of methanolic extracts from WT Arabidopsis in ESI (−) data mode. Figure S2: Unsupervised chemometric modelling of LC-MS-analysed Arabidopsis WT leaf extracts (ESI (−) data). Figure S3: OPLS-DA modelling of Arabidopsis WT leaf extracts (ESI (+) data). Figure S4: OPLS-DA modelling of Arabidopsis WT leaf extracts (ESI (+) data). Figure S5: A representative receiver operator characteristic (ROC) plot summarising the performance of binary classifiers (OPLS-DA). Figure 6: Heatmap presentation of significant annotated metabolites in ESI (+) mode. Figure S7: Alpha-linolenic acid metabolism analysis by Metabolomic Pathway Analysis (MetPA). Figure S8: Phenylpropanoid biosynthesis analysis by Metabolomic Pathway Analysis (MetPA). Figure S9: Flavone and flavonol biosynthesis analysis by Metabolomic Pathway Analysis (MetPA). Figure  S10: Glucosinolate biosynthesis analysis by Metabolomic Pathway Analysis (MetPA). Table S1: Diagnostic fragments of annotated metabolites and KEGG IDs in Table 1. Table S2: List of significant metabolic pathways that were altered by the treatment of Arabidopsis WT with LPS Pst and LPS Xcc as generated by Metabolomic Pathway Analysis (MetPA). Table S3: List of significant metabolic pathways that were altered by the treatment of Arabidopsis lbr2-2 mutant with LPS Pst and LPS Xcc as generated by Metabolomic Pathway Analysis (MetPA). Table S4: List of significant metabolic pathways that were altered by the treatment of Arabidopsis bak1-4 mutant with LPS Pst and LPS Xcc as generated by Metabolomic Pathway Analysis (MetPA).