Metabolomic and transcriptomic analysis of the rice response to the bacterial blight pathogen Xanthomonas oryzae pv. oryzae
- First Online:
- Cite this article as:
- Sana, T.R., Fischer, S., Wohlgemuth, G. et al. Metabolomics (2010) 6: 451. doi:10.1007/s11306-010-0218-7
- 2.5k Downloads
Bacterial leaf blight (BLB), caused by Xanthomonas oryzae pv. oryzae (Xoo), gives rise to devastating crop losses in rice. Disease resistant rice cultivars are the most economical way to combat the disease. The TP309 cultivar is susceptible to infection by Xoo strain PXO99. A transgenic variety, TP309_Xa21, expresses the pattern recognition receptor Xa21, and is resistant. PXO99△raxST, a strain lacking the raxST gene, is able to overcome Xa21-mediated immunity. We used a single extraction solvent to demonstrate comprehensive metabolomics and transcriptomics profiling under sample limited conditions, and analyze the molecular responses of two rice lines challenged with either PXO99 or PXO99△raxST. LC–TOF raw data file filtering resulted in better within group reproducibility of replicate samples for statistical analyses. Accurate mass match compound identification with molecular formula generation (MFG) ranking of 355 masses was achieved with the METLIN database. GC–TOF analysis yielded an additional 441 compounds after BinBase database processing, of which 154 were structurally identified by retention index/MS library matching. Multivariate statistics revealed that the susceptible and resistant genotypes possess distinct profiles. Although few mRNA and metabolite differences were detected in PXO99 challenged TP309 compared to mock, many differential changes occurred in the Xa21-mediated response to PXO99 and PXO99△raxST. Acetophenone, xanthophylls, fatty acids, alkaloids, glutathione, carbohydrate and lipid biosynthetic pathways were affected. Significant transcriptional induction of several pathogenesis related genes in Xa21 challenged strains, as well as differential changes to GAD, PAL, ICL1 and Glutathione-S-transferase transcripts indicated limited correlation with metabolite changes under single time point global profiling conditions.
KeywordsMetabolomicsLC/MSGC/MSMicroarrayRiceBacterial leaf blightPathogenPattern recognition receptorPathogen associated molecular pattern
There’s a growing interest in studies that attempt to link gene expression with metabolite profiling (Brosche et al. 2005; Kaplan et al. 2007; Jumtee et al. 2008; Vijayendran et al. 2008). Many plant functional genomics studies have combined the generation of transgenic and mutant plants with the parallel analysis of mRNA expression, protein levels and plant histology studies (Fiehn et al. 2000; Finkelstein et al. 2002; Oliver et al. 2002; Henikoff and Comai 2003; Farrokhi et al. 2006; Steinberg et al. 2008). Extensive literature also exists for metabolic perturbation of host/plant interactions through parallel analysis of metabolomics experiments and microarray analysis (Urbanczyk-Wochniak et al. 2003; Dharmadi and Gonzalez 2004; Hirai et al. 2004; Verhoeckx et al. 2004; Matsuzaki et al. 2005; Frey et al. 2007; Cho et al. 2008; Hoefgen and Nikiforova 2008; Nam et al. 2009; Tan et al. 2009; Yang et al. 2009). Thus, workflows to integrate metabolic and genomic analysis will contribute to a comprehensive biological understanding for observed phenotypic differences between organisms that gene expression studies alone would otherwise be unable to achieve.
Rice provides food for over half the world’s human population, and the availability of its genome sequence (2005) is important for research and development of disease resistant cultivars. Bacterial leaf blight (BLB) is responsible for major rice production losses in irrigated and rain-fed lowland ecosystems. The use of resistant rice varieties continues to be the most economical and effective way to combat this disease. Cell surface pattern recognition receptors (PRRs) are key components of the innate immune response in animals and plants. The rice PRR, XA21 (Song et al. 1995), recognizes the pathogen associated molecular pattern (PAMP), Ax21 (Activator of XA21-mediated immunity) that is highly conserved in all tested species of the genus Xanthomonas as well as in Xylella species (Lee et al. 2006; Lee et al. 2009). Recently, mechanisms for XA21-mediated resistance have been elucidated (Park et al. 2008).
We used metabolomics and transcriptomics profiling workflows for discovery analysis to contrast the responses of two rice varieties to Xoo: (1) TP309, the parent rice genotype, which is susceptible to infection by the Philippine bacterial Xoo strain PXO99, expressing Ax21 and (2) TP309_Xa21, a transgenic variety of rice harboring the Xa21 PRR, which is resistant to Xoo strain PXO99. Production of Ax21 requires the presence of eight rax (required for activation of Xa21-mediated immunity) gene family members (Shen et al. 2002; Burdman et al. 2004; da Silva et al. 2004), including raxST. Strains carrying mutations in any of the known rax genes are no longer recognized by XA21-containing rice lines resulting in lesion development on XA21-lines. For example, PXO99 strains carrying a knockout in the raxST gene (PXO99△raxST) cause long lesions on TP309_Xa21 lines. In contrast, TP309 is susceptible to both PXO99 and PXO99△raxST strains.
We developed a robust, all-in-one rice leaf sample extraction method that uses a single organic solvent mixture of water/acetonitrile/isopropanol for LC/MS, GC/MS and microarray analyses. Importantly, contaminating waxes that can potentially foul the MS source were not extracted. Our MS data analysis workflow included post-processing data filtering to yield higher quality results that was required to shed further insights into metabolite and transcript perturbations of this plant/parasite interaction.
2 Materials and methods
2.1 Study design
Biological study design classes with susceptibility to bacterial infection
No treatment (NT)
2.2 Rice sample processing and metabolite extraction
The samples were processed according to Weckwerth et al. (2004) with the following changes. Rice leaves were collected under controlled conditions, flash frozen and stored at −80°C. The cutting of reproducible amounts of frozen rice leaf samples for metabolomics and gene expression analysis is critically important. First because the solvent volume and solvent ratios are optimized for samples in this weight range and second because similar reproducible measurements for all groups of samples allows inter-group comparisons of data. 18 ± 2.2 mg segments of rice leaves were cut and weighed. The rice leaves were placed in liquid nitrogen cooled, 2 ml Eppendorf tubes, each containing a 5 mm stainless steel ball bearing from Retsch (Newtown, PA). The tubes were transferred to MM301 Retsch mill adapter racks that had also been pre-cooled in liquid nitrogen. Rice leaves were ground in the homogenizer for 30 s at 25 Hz. Metabolites were extracted from membrane and cell wall components in 1 ml of a single phase, 2:3:3 v/v/v solvent mixture of water/acetonitrile/isopropanol, that had been pre-cooled at −20°C in a Digital One RTE-7 refrigerated circulating bath (Thermo Electron Corp., Newington, NH). This solvent mixture minimizes the extraction of waxes that are detrimental to the analytical instrumentation and enables analysis of metabolites by both LC–TOF and GC–TOF instruments. After the samples were incubated at −20°C in the refrigerated circulating bath for 20 min, the tubes were centrifuged to pellet mRNA and protein and were stored at −80°C for subsequent mRNA processing. The liquid phase supernatant extracts were divided equally and transferred under nitrogen gas to autosampler vials (Agilent, Palo Alto, California, USA) with Teflon/silicon septa, for LC–TOF or GC–TOF analyses.
2.3 Total RNA extraction, RNA quality analysis, amplification and labeling
Total RNA was extracted from the cell pellets using the Agilent Plant RNA Isolation Mini Kit (Agilent Technologies, Santa Clara, CA). This phenol-free spin column technique results in RNA with substantially reduced genomic DNA (gDNA) contamination without the need for DNase I treatment. Briefly, the pellet was extracted in 4 M guanidine hydrochloride, 25 mM Tris–HCl and β-mercaptoethanol solution, spun through a pre-filtration column and RNA was retained on a mini isolation column. The RNA was washed, dried and eluted off the column in 10–50 μl of nuclease-free water.
The NanoDrop® ND-1000 Spectrophotometer was first used to determine the concentration of purified RNA. The Agilent 2100 Bioanalyzer was used with the RNA 6000 Nano LabChip® Kit (Agilent Technologies, Santa Clara, CA), which uses fluorescent dyes to bind nucleic acids for tracking the electrophoretic separation of total RNA. One micro liter of each sample (25–500 ng/μl) was loaded per well to assess ribosomal RNA quality by an accurate reading of 5S, 18S and 28S peaks, along with any degradation. The Low RNA Input Linear Amplification Kit PLUS, Two-Color kit (Agilent Technologies, Santa Clara, CA) generated fluorescent complimentary RNA (cRNA) for use in hybridizing with oligonucleotide microarrays. Briefly, for each sample, 500 ng of extracted Total RNA was first mixed with two spike mixes from the Agilent Two-Color RNA Spike-In Kit (Agilent Technologies, Santa Clara, CA) according to manufacturer instructions. Each mix contains 10 in vitro synthesized, polyadenylated transcripts in predetermined ratios that in the subsequent hybridization process specifically binds to control probes on a microarray. The results are intended to provide expected versus observed log ratios for monitoring microarray hybridization linearity, sensitivity, and accuracy. After adding the two spike mixes, a primer containing poly dT and a T7 polymerase promoter was added to the mixture, and annealed to the poly A + RNA. The first and second strands of cDNA were synthesized by adding reverse transcriptase to the mix. Next, T7 RNA polymerase was added to synthesize cRNA from the double-stranded cDNA, and simultaneously incorporated cyanine 3- or cyanine 5-labeled CTP (Qiagen N.V., Venlo, Netherlands). RNeasy mini spin columns were used to purify the amplified cRNA samples. Fluorescently labeled cRNAs were analyzed on a ND-1000 Spectrometer (NanoDropTechnologies, Rockland, DE) for determining yield and specific activity. The yield for the rice RNA was in the range expected for plant products, with an average specific activity of cyanine 3- and cyanine 5-labeled cRNA targets of 14.6 (±3.3) pmol dye/μg cRNA.
2.4 Microarray analysis
The Agilent Gene Expression Hybridization Kit (Agilent Technologies, Santa Clara, CA) was used for hybridizing fluorescently-labeled cRNA to the Agilent 22 K Rice oligo microarrays according to the manufacturer instructions. This 22,575 probe microarray contains 60-mer probes, with broad gene coverage, representing an estimated 60% of the rice genome (japonica race). The probes were designed from the extensive cDNA library for rice, developed by Dr. Shoshi Kikuchi of Japan’s National Institute of Agrobiological Sciences (NIAS). Annotation for this microarray is presented on the NIAS website at (http://cdna01.dna.affrc.go.jp/cDNA/). The 1 × 3′′ glass slides were washed, dried and scanned using an Agilent laser scanner according to manufacturer instructions. Agilent Feature Extraction software was used for automated processing of microarray image files that included dye normalization using LOWESS, filtering of poor, flagged pixels and determination of feature intensities and ratios. The output files were imported into GeneSpring-GX data analysis software, where the data was normalized to the 50th percentile across the different slides. The replicate dye-swapped samples were grouped into one experiment and analyzed for statistically significant differential genes. To find differentially expressed genes between two different conditions, the gene lists were further filtered using Volcano plots, which permitted the visualization of the relationship between fold-change and statistical significance (which takes both fold change and P-value into consideration).
2.5 GC–TOF mass spectrometry
A mixture of internal retention index (RI) markers was prepared using fatty acid methyl esters of C8, C9, C10, C12, C14, C16, C18, C20, C22, C24, C26, C28 and C30 linear chain length, dissolved in chloroform at a concentration of 0.8 mg/ml (C8–C16) and 0.4 mg/ml (C18–C30). 1 μl of this RI mixture was added to the dried extracts. 10 μl of a solution of 20 mg/ml of 98% pure methoxyamine hydrochloride (Sigma, St.Louis, MO) in pyridine (silylation grade, Pierce, Rockford, IL) was added and shaken at 30°C for 90 min to protect aldehyde and ketone groups. 90 μl of MSTFA.1%TMCS (1 ml bottles, Pierce, Rockford, IL) was added for trimethylsilylation of acidic protons and shaken at 37°C for 30 min. The reaction mixture was transferred to a 2 ml clear glass autosampler vial with microinsert (Agilent, Santa Clara, CA). A Gerstel automatic liner exhange system was used in conjunction with a Gerstel CIS cold injection system (Gerstel, Muehlheim, Germany). An Agilent 6890 gas chromatograph (Santa Clara, CA) was controlled by the Leco ChromaTOF software versus 2.32 (St. Joseph, MI). A 30 m long, 0.25 mm i.d. rtx5Sil-MS column with 0.25 μm 5% diphenyl/95% dimethyl polysiloxane film and additional 10 m integrated guard column was used (Restek, Bellefonte, PA). Result *.txt files were exported to a data server with absolute spectra intensities and further processed by a filtering algorithm implemented in the metabolomics BinBase database (Fiehn et al. 2005). Quantification was reported as peak height using the unique ion as default, unless a different quantification ion was manually set in the BinBase administration software Bellerophon. A quantification report table was produced for all database entries that were positively detected in more than 50% of the samples of a study design class as defined in the SetupX database (Scholz 2007). A subsequent post-processing module was employed to automatically replace missing values from the *.cdf files using libraries obtained from the open access mzmine software. Replaced values were labeled as ‘low confidence’ by color coding.
Metabolites were unambiguously assigned by the BinBase identifier numbers, using retention index and mass spectrum as the two most important identification criteria. Additional confidence criteria were given by mass spectral metadata, using the combination of unique ions, apex ions, peak purity and signal/noise ratios as given in data preprocessing. All database entries in BinBase were matched against the Fiehn mass spectral library of 1,200 authentic metabolite spectra and the NIST05 commercial library (Kind et al. 2009). BinBase entries were named manually by both matching mass spectra and retention index and PubChem numbers and KEGG identifiers were added. In addition, all reported compounds (identified and unknown metabolites) are reported by the quantification ion and the full mass spectrum encoded as string. All raw and processed data are downloadable at http://fiehnlab.ucdavis.edu:8080/m1/main_public.jsp.
2.6 LC–TOF mass spectrometry
An Agilent 1100 LC equipped with a ZORBAX SB-Aq column 2.1 × 150 mm, 3.5 μm was used to separate the rice extracts. The LC parameters were as follows: autosampler temperature, 4°C; injection volume, 2 μl; column temperature, 20°C; and flow rate of 0.3 mL/min. The LC solvent and time table were: a 2–98% linear gradient of water/acetonitrile employed over 46 min followed by a solvent hold until 54.9 min, at which time data collection was stopped. 0.1% formic acid was used as a mobile phase modifier. Mobile phase: A = 0.1% formic acid in water, B = 0.1% formic acid in acetonitrile. Gradient: 2% B at 0 min., 98% B at 46 min., 98% B at 54.9 min., 2% B at 55 min. MS stop time: 54.9 min. LC stop time: 55 min.
Mass spectrometric analysis was performed on an Agilent 6210 Time-of-Flight LC/MS equipped with an electrospray (ESI) ion source. Nitrogen was used as the drying gas at 35 psi. The ESI source used a separate nebulizer for the continuous, low-level (10 μl/min) introduction of reference mass compounds: 121.050873, 922.009798 (positive ion mode) and 119.036320, 966.000725 (negative ion mode). ESI spray voltage was 4,000 V in positive mode and 3,500 V in negative ion mode and fragmentor at 170 V. Nitrogen was used as sheath gas at 30 psi. The accurate-mass LC–TOF chromatograms were acquired and deconvoluted into individual chemical peaks. The molecular features were identified using the Molecular Feature Extraction (MFE) algorithm in MassHunter™ software (Agilent Technologies, Santa Clara, CA, USA). The retention time/mass pairs generated by the MassHunter™ software were then exported for subsequent statistical analysis.
GeneSpring-MS 1.2 statistical analysis software was used for retention time and mass abundance adjustments. The multiple data sets were normalized to correct data for retention time and response drift, and aligned before they were compared. The workflow followed the following steps: (1) Alignment and normalization of features across all samples. (2) Data Transformation: measurements less than 0.01 were set to 0.01; (3) Normalization: “Per Mass” normalized to the median, (4) Filtering: the six independent biological replicate samples per group were analyzed as one group after the ion intensity for each molecular ion was averaged across the six biological replicates. The relative frequency of detection (number of samples a particular feature is found in one or more groups) was varied and compared. The highest quality features analyzed were those that were only present in all six replicates, in at least one or more groups.
The features were then analyzed by Volcano plots to reveal both fold changes and statistically significant (P < 0.05) differential abundance profiles. The resulting mass lists were then queried against the METLIN Personal Metabolite database to find matches based on neutral mass, using a 5 ppm window. This database currently contains over 23,000 compounds, and includes links to KEGG identifiers, CAS numbers, HMDB and LipidMaps. Because retention time information was not used as an orthogonal criterion for identification, a Molecular Formula Generator (MFG) algorithm was used to calculate an independent best fit empirical formula for each feature using the mass spectral metadata: monoisotopic mass, isotope ratio and abundance information. The MFG score helped in determining whether the calculated empirical formula agreed with the formula for the compound in the database match. Isomers and isobaric compound remained unresolved. Statistical evaluation was performed using univariate analyses, including the Student t-test for independent pairs of groups, and one way ANOVA for multiple groups. The effect of the treatment (example, mock-treated and PXO99-infected) and rice lines (TP309 and TP309_Xa21) were compared. Because PXO99△raxST was an unpaired condition that was only tested against the TP309_Xa21 line, it was independently compared to mock-treatment in the TP309_Xa21 line using a t-test. Values of P < 0.05 were considered statistically significant and the Benjamini and Hochberg (Benjamini and Hochberg 1995) False Discovery Rate of 5% was used for multiple testing correction. We then investigated which of the mean(s) in the ANOVA were different using the Tukey (Tukey 1949) multiple comparison test. Multivariate statistical analysis was performed by unsupervised principal component analysis (PCA) with mean centering and scaling to visually demonstrate the variance of the metabolic phenotypes from the groups based on the discriminating features from the ANOVA analysis. Hierarchical cluster analysis of metabolites was performed to reveal associations between replicate biological samples within a group based on the similarity of their mass abundance profiles.
For GC–TOF data, statistical evaluations were performed by Statistica DataMiner versus 7.0 (StatSoft, Tulsa, OK). For univariate statistics, data were log(10)-transformed to reduce weight of outliers and to render the data sets more normal-distributed. Two-tailed Student t-tests were performed with significance thresholds of P < 0.05. Ratios were calculated from antiloged class averages. For PCA and Partial Least Square (PLS) analyses, variables were scaled to unit standard deviations and components were calculated with the Non Linear Iterative PLS algorithm using 50 iterations and a 0.0001 convergence criterion with a 7-fold cross-validation for supervised PLS. PCA was carried out on unit variance scaled data in unsupervised mode for dimension reduction purposes.
3 Results and discussion
The susceptible cultivar TP309 was compared to the TP309_Xa21 transgenic variety in the absence of infection to see whether the transgenic event itself triggered metabolic differences leading to resistance. We further investigated metabolomic data to study if the TP309_Xa21 plants show specific metabolic responses after infection that might contribute to the resistance. As a positive control we used PXO99△raxST bacteria that are virulent on TP309_Xa21 plants in order to discriminate metabolic differences due to the transgenic host and perturbations that are caused by the infection.
3.1 The number of detected peaks or features increases linearly with class frequency threshold reduction
Application of different stringency filters to the processing of raw mass spectrometry data files significantly affects interpretation of global metabolite profiling analyses
Class frequency threshold (%)
LC–TOF positive ESI
LC–TOF negative ESI
In total, 514 unfiltered features were detected in LC–TOF runs in positive ion mode and 474 in negative ion mode. Upon filtering, the features were reduced to 355 (50% threshold) or 186 (100% threshold). Secondly, it was apparent that after comprehensive peak picking and constraining result files, a similar total number of peaks for GC were calculated for GC–TOF files, suggesting the complementary nature of these technologies. Furthermore, a similar number of peaks were obtained from negative ionization and positive ionization LC–TOF analyses, indicating that both modes are necessary for discovery metabolomics. In principle, it should be possible to combine negative/positive LC–TOF data files to a non-redundant list of mass features using the retention time/accurate mass information concomitant with lists of ion adducts in both modes. Unfortunately, the software did not support this strategy at the time of analysis. GC–TOF data can be classified to identified and structurally annotated metabolites (via retention index and mass spectral matching to reference libraries), in comparison to structurally uncharacterized unknown metabolites. In a similar manner to the total number of metabolic signals, the number of identified metabolites linearly increased from 104 to 154 non-redundant metabolites with decreasing of relative frequency threshold levels. This finding confirmed that setting the positive relative frequency threshold at 50% was reasonable for comprehensive metabolomics. The fact that only around one-third of all peaks were structurally assigned is typical for unbiased metabolomics screening, suggesting that functional annotation of plant genes would greatly benefit from improved algorithms and libraries used for compound identifications.
3.2 Rice cultivar TP309 and its Xa21 transgenic variant differ in metabolic phenotypes
3.3 Unsupervised and supervised classification tools clearly discriminate TP309_Xa21 challenged bacterial strains
The most interesting question in this study was whether PX099 challenge induces different metabolic responses in susceptible TP309 versus the resistant TP309_Xa21 genotype, and secondly, if TP309_Xa21 samples respond differently to PXO99△raxST versus PX099.
Although GC–TOF results showed that untreated (NT) TP309 and TP309_Xa21 metabolic phenotypes exerted a number of differences at t-test significance levels of P < 0.05, for example in sugar alcohols, trehalose, raffinose, pipecolic acid, fumaric acid and maleic acid (Supplement S2), none of these compounds were found to be relevant when exploring metabolic responses in the TP309_Xa21 towards PX099 (resistant) and PXO99△raxST (susceptible) bacterial challenges, for which clear metabolic phenotypes were found using both unsupervised and supervised multivariate statistics.
For GC–TOF analysis, where we benefited from the BinBase database of known, structurally identified compounds, supervised partial least squares classification was used to distinguish metabolic responses between bacterial challenged and the corresponding mock-treated controls. In both cases, distinct metabolic phenotypes were clearly discriminated (Fig. 2b), showing that responses in TP309_Xa21 were indeed different depending on the strain of bacteria used to infect the plant. Interestingly, the results for both LC–TOF and GC–TOF platforms more of the variance was explained by the general difference of mock-treated and PXO99 challenged plants (vector 1) than between PXO99 versus PXO99△raxST inoculations, discriminated by vector 2. This finding indicates that Xoo infection results in metabolic adaptations even if the strain used does not cause long lesions, and is consistent with observations that Xoo can grow to fairly high levels in the early stages of infection even in resistant Xa21 cultivars (Song et al. 1995; Lee et al. 2009).
Univariate ANOVA statistics on bacterial challenge comparisons for TP309 and TP309_Xa21 rice plants
Class frequency threshold (%)
TP309_PXO99 versus TP309_Mock
Xa21_PXO99 versus Xa21_Mock
Xa21_PXO99ΔraxST versus Xa21_Mock
LC/TOF (+) 100
LC/TOF (+) 75
LC/TOF (+) 50
LC/TOF (−) 100
LC/TOF (−) 75
LC/TOF (−) 50
3.4 Important precursors for alkaloid metabolism are significantly higher in the resistant TP309_Xa21 genotype challenged with the PXO99 strain
Phenylalanine and tyrosine had been found before to be inducible in response to biotic stress, for example by fungal attacks on plant leaves (Khan et al. 2003). Both compounds are precursors for alkaloid biosynthesis or, alternatively, for the production of phenylpropanoids which are furthered into lignins. Therefore, up-regulation of phenylalanine and tyrosine levels may help the resistant TP309_Xa21 transgenic rice to defend against bacterial infection by alkaloids and partial cell wall lignifications. The induction of phenylalanine- and phenylalanine/tyrosine ammonia lyases (PAL and PTAL, Os04g0518400 and Os02g0626100) is a hallmark of plant responses to bacterial infection and frequently used as indicator or even elicitor of cellular response mechanisms in rice (Bera and Purkayastha 1999; Vidhyasekaran et al. 2001; Jayaraj et al. 2004). Ornithine, citrulline and arginine are all member of the urea cycle that salvage ammonia, which is released by the phenylalanine/tyrosine ammonia lyase reaction, but are also precursors in alkaloid biosynthesis via putrescine or cadaverine intermediates. Hence, it can be concluded that several important precursors for alkaloid metabolism were found to be significantly up-regulated in the resistant PXO99-challenged TP309_Xa21 genotype in comparison to all other test cases including the PXO99△raxST-challenged TP309_Xa21 genotype. A range of further metabolites were significantly differentially regulated when comparing the PXO99△raxST-challenged TP309_Xa21 to both the PXO99-challenge and the mock-treated TP309_Xa21 plants. Some metabolites such as glutamine (Fig. 3) and phenol, ala-ala, 5-hydroxynorvaline and hydroxylamine (Supplement S2) were more than 2-fold higher in the PXO99-challenged Xa21 plants than in the susceptible PXO99△raxST challenge. Concomitantly, in the PXO99△raxST treatment, a range of metabolic differences were apparent that did not occur in any of the other treatments. Figure 3 details a down-regulation of glutamate levels that was found concurrent with up-regulation in two glutamate products, oxoproline and N-acetylglutamate. Oxoproline is formed from glutamyl-cysteine using the cyclizing (5-l-Glutamyl)-l-amino-acid 5-glutamyltransferase (EC 184.108.40.206), yielding cysteine and the cyclized amino acid product. Interestingly, a second cyclized product was observed, butyrolactam, for which yet no enzyme has been identified in rice but which might be produced from GABA (4-aminobutyric acid) using the same enzyme but the hypothetical GABA-cysteine as substrate. Indeed, cysteine was found more than 2-fold up-regulated in the PXO99△raxST treated Xa21 plants (susceptible) compared to both mock-treated and PXO99-treated plants, albeit less significant (P = 0.1). This finding supports the assumed cyclization reaction via EC 220.127.116.11 with cysteine as by-product. The intermediate Glu-Cys is also an important precursor in the biosynthesis of glutathione (Glu-Cys-Gly) which requires glycine as further substrate, and indeed, 3-fold higher levels of glycine were found at P < 0.05 in the PXO99-challenged Xa21 plants (resistant) than in the PXO99△raxST treated or the mock-treated Xa21 plants (Supplement S2). Taken together, these data indicate that the resistance phenotype in transgenic Xa21 plants upon PXO99-challenge is manifested in up-regulation of precursors for alkaloid biosynthesis as well a potential higher flux into glutathione biosynthesis than into the alternative reaction towards cyclization into oxoproline.
3.5 The virulence signal acetophenone is absent in TP309_Xa21-infected rice plants but abundant in mock-treated controls
Upon bacterial challenges, a range of further compounds were found to be differentially regulated in TP309_Xa21 plants (Supplement S2) which do not seem to be directly involved in the resistance phenotype itself. For example, the phytosterols stigmasterol and beta-sitosterol were found at 30% lower levels in PXO99 challenged Xa21 plants compared to PXO99△raxST challenged Xa21 plants or the corresponding mock-treated Xa21 plants, but the same difference was found in susceptible PXO99-challenged TP309 plants, lacking Xa21. In the transgenic Xa21-plants, a 2-fold reduction in the cell wall component sinapinic acid (but not ferulic acid) was found upon treatment with either PXO99△raxST or PXO99, compared to the mock control.
When investigating specifically which metabolites occur only in the XA21 mediated response to PX099, i.e. signals that were specific to XA21/PXO99 but are not found in any of the other combinations, we found an unknown with the BinBase database identifier number 203611 (quantification m/z 128, FAME-based retention index 223571). The closest annotation was cyclohexylamine, but the identity was not further validated. In follow-up investigations, the hypothesis of an involvement of this tentatively annotated compound in the Xa21-mediated defense response needs to be examined.
3.6 Microarray analysis
3.7 Correlating the results of transcriptomics and metabolomics provides several clues towards explaining which pathways are responsible for Xa21 resistance to PXO99
Correlations between microarray and associated metabolites for selected differential genes
Oryza sativa isocitrate lyase (ICL1)
Glutamate decarboxylase (GAD1)
GABA (4-amino butyric acid)
Phenylalanine ammonia-lyase (PAL)
Lipid transfer protein b1
UDP-glucose:salicylic acid glucosyltransferase (SA-GTase)
Uridine 5′ diphosphate
Other over-expressed genes included β-1,3-glucanase and chitinase (Table 4) in PXO99 compared to mock infected Xa21. Conversely, these genes were suppressed in the PXO99 challenged susceptible strain versus TP309 mock. Both Chitinases and β-1,3-glucanases catalyze a broad series of substrates and products. Transgenic rice plants that contain one or more fungal disease-resistant gene(s) have previously been reported to control fungal diseases of rice (O’Toole 2001). Chitinases and Beta-1,3-glucanases are fungal cell wall degrading enzymes and can be found in plants and microbes. Genes encoding different chitinases and β-1,3-glucanases have previously been cloned and used to transform tobacco, resulting in increased survival relative to control plants in the presence of pathogenic fungi (Benhamou et al. 1993; Collinge et al. 1993). Indeed, β-1,3-Glucanases are commonly found throughout the plant kingdom and there is abundant evidence that they are involved in the defense mechanism of plants against pathogen infection, and have been proposed to be involved in physiological processes such as the stress response (Simmons 1994). Apparently these genes, in addition to being induced by fungi, are also induced in TP309_Xa21 in response to PXO99 but not in Xa21-PXO99△raxST.
Perhaps the most compelling observation was the comparison of transcriptomics data and metabolite data for the putative transcript for Glutamate decarboxylase (GAD), which codes for an enzyme that catalyzes the decarboxylation of glutamate to γ-aminobutyric acid (GABA) and CO2. The decrease of glutamate was confirmed by GC–TOF data, in accordance to the induction of GAD in both strains that were challenged with PXO99 and PXO99△raxST. GAD was particularly elevated in the Xa21 line (Table 4). LC–TOF analysis also revealed elevated levels for C4H9NO2. This empirical formula includes annotation as GABA in the METLIN database. An overlay of the EICs for the corresponding m/z revealed a clear increase in the abundance for Xa21_PXO99 samples relative to Xa21_mock (data not shown). Although the role of GABA in plants remains controversial (Bouche et al. 2004; Bouche and Fromm 2004; Bown et al. 2006), environmental perturbations of oxygen, water, and temperature, is reported to prompt the accumulation of GABA in plant cells (Shelp et al. 1999). GC–TOF data indicated that GABA might be furthered into cyclization to butyrolactam as outlined above, but the significance of this process for pathogen response is not established in literature.
Phenylalanine metabolism was significantly affected in PXO99 challenged XA21 rice variety, but not in the wild-type variety. Phenylalanine ammonia lyase (PAL), a key defense-related enzyme of the phenyl propanoid pathway, has recently been associated with disease resistance in rice plants via synthesis of isoflavones, phytoalexins, and lignin, compounds (Lin et al. 2008). Indeed rapid accumulation of PAL mRNA following inoculation with a pathogen has been described for maize (Cui et al. 2002). Moreover, rice leaves inoculated with Magnaporthe grisea significantly increased the enzyme activity of PAL (Cai et al. 2008). The PAL transcript in our study was significantly induced only in Xa21_PXO99 compared to Xa21_mock (Table 4; Supplement S5). Taken together, these results are consistent with our GC–TOF data for Phenylalanine where it was elevated.
A large number of glutathione and putative glutathione S-transferases (GST), as well as a glutathione transport protein were either detected or significantly induced in PX099 challenged relative to mock Xa21. In TP309, the induction/repression response to PXO99 was dependent on the GST––or putative GST––and on average, both in the magnitude and number of GSTs cancelled out any differences between PXO99 challenged and mock. The Xa21 line challenged with PXO99△raxST showed many GSTs to be significantly induced relative to mock. Interestingly, however, although several GSTs were detected, no significant differences were observed between Xa21_PXO99 compared to Xa21_Mock. GSTs detoxify endogenous compounds such as peroxidised lipids, they bind to toxins, or function as transport proteins, all of which might indicate a general response to stress. Oxidized glutathione levels were significantly different by LC–TOF analysis in the PXO99△raxST Xa21 line, detected but not altered in PXO99 challenged Xa21, and reduced in the wild-type TP309. This is an interesting result suggesting that PXO99 challenged Xa21 is not stressed and that the oxidized form of glutathione in response to PXO99 may be depleted in TP309 wild-type cells. Such hypotheses would need to be validated by target analyses of GSH/GSSG ratios.
Lipid Transfer Proteins (Ltp) reversibly bind and transport hydrophobic molecules in vitro and are likely to play a role in key processes of plant physiology. The induction of Ltp gene expression has been previously characterized in a fungus, and coincides with a transient increase in the expression levels of phenylalanine ammonia-lyase (PAL) gene, as well as salicylic acid (SA) gene in mycorrhizal roots (Blilou et al. 2000). Although we were able to detect induction of Ltp b1 transcripts in TP309 cells, we did not find it to correlate with co-expression or differential expression with either PAL or phenylalanine and salicylic acid levels in PXO99 challenged TP309 condition (Table 4). However, these metabolites were detected in PX099_Xa21 challenged plants instead and can be due to factors including that our system was non-fungal, limitations of assay sensitivity, as well as offsets between maximal expression of transcripts and metabolites.
An optimized sample processing protocol, starting from the same leaf sample of average weight of 15–20 mg FW, yielded sufficient material for parallel analysis by LC–TOF, GC–TOF and microarrays. This protocol is useful for investigation of complex processes such as plant infection and immunity, in leaf samples at different distances from the location of the infection. Unbiased, discovery-based metabolomics and transcriptomics analyses yielded novel insights into the rice response to Xoo. Our results reveal global metabolic and transcriptomic changes in leaf tissues of the XA21 transgenic variant challenged with PX099 but not to PXO99△raxST. While central carbon catabolism is reduced in correlation between gene and metabolite expression in both Xa21 and TP309 rice genotypes, genes and metabolic precursors for alkaloid biosynthesis are increased specifically in the XA21-mediated response to PX099 but not to PXO99△raxST. The outcome of ‘omics’ studies such as these will aid in a better understanding of complex response to pathogen infection (Tarpley et al. 2005). Although our analyses shows the robustness of our system, enabling differentiation of metabolite and transcript responses of two different rice genotypes and bacterial strains, coherent interpretation of transcriptomics and metabolomics results is very challenging, requiring carefully designed experiments, a large number of biological replicates and inclusion of time course studies to provide the appropriate context for further biological conclusions.
This work was financially supported by NSF grant MCB-0820823 to OF and NIH grant GM55962 to PCR and DOE grant # DE-AC02-05CH11231 to PCR. Technical assistance by Yun Lu for GC–TOF data acquisition is appreciated. The assistance of Sean Coughlan in supplying the Agilent rice microarrays and Kshama Mehta who helped perform some of the microarray hybridizations is appreciated.
TRS carried out microarray and LC–TOF analyses, data interpretation and writing the manuscript. SF performed LC–TOF and data analysis. AK assisted with sample extraction and preparation. KHJ performed the bacterial challenge tests on the various rice genotypes. SC provided the rice arrays, KM assisted with microarray hybridization. GW altered BinBase algorithms to retain only high quality GC–TOF spectra for rice metabolites. OF performed GC–TOF data analysis, data interpretation and writing the manuscript. PR helped plan the experiments and assisted with interpretation of results and writing of the MS.
The author(s) declare that they have no competing interests.
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.