A transcriptomics approach uncovers novel roles for poly(ADP-ribosyl)ation in the basal defense response in Arabidopsis thaliana

Pharmacological inhibition of poly(ADP-ribose) polymerase (PARP) or loss of Arabidopsis thaliana PARG1 (poly(ADP-ribose) glycohydrolase) disrupt a subset of plant defenses. In the present study we examined the impact of altered poly(ADP-ribosyl)ation on early gene expression induced by the microbe-associate molecular patterns (MAMPs) flagellin (flg22) and EF-Tu (elf18). Stringent statistical analyses and filtering identified 178 genes having MAMP-induced mRNA abundance patterns that were altered by either PARP inhibitor 3-aminobenzamide (3AB) or PARG1 knockout. From the identified set of 178 genes, over fifty Arabidopsis T-DNA insertion lines were chosen and screened for altered basal defense responses. Subtle alterations in callose deposition and/or seedling growth in response to those MAMPs were observed in knockouts of At3g55630 (FPGS3, a cytosolic folylpolyglutamate synthetase), At5g15660 (containing an F-box domain), At1g47370 (a TIR-X (Toll-Interleukin Receptor domain)), and At5g64060 (a predicted pectin methylesterase inhibitor). Over-represented GO terms for the gene expression study included "innate immune response" for elf18/parg1, highlighting a subset of elf18-activated defense-associated genes whose expression is altered in parg1 plants. The study also allowed a tightly controlled comparison of early mRNA abundance responses to flg22 and elf18 in wild-type Arabidopsis, which revealed many differences. The PARP inhibitor 3-methoxybenzamide (3MB) was also used in the gene expression profiling, but pleiotropic impacts of this inhibitor were observed. This transcriptomics study revealed targets for further dissection of MAMP-induced plant immune responses, impacts of PARP inhibitors, and the molecular mechanisms by which poly(ADP-ribosyl)ation regulates plant responses to MAMPs.


Introduction
The plant immune system is composed of at least three basic components: pre-formed defenses, infection-induced basal defenses, and R-gene mediated defenses [1][2][3]. Once preformed structural barriers are breached, the basal immune response is activated by pattern
One additional plate each of Col-0 and parg1-2 was used to confirm the presence of transgene and reduced PARG1 mRNA in the parg1-2 plants compared to Col-0. The three biological replicates using this method were performed on separate dates.

RNA isolation and cDNA synthesis
Total RNAs were isolated from one pool per treatment per biological replicate using QIAGEN RNeasy mini kit and treated with DNAse using the Qiagen RNase-Free DNAse set (Qiagen, Valencia, CA, USA). RNA quality and concentration was preliminarily verified by spectrophotometry and gel electrophoresis, and then sent to the Gene Expression Center (University of Wisconsin-Madison Biotechnology Center, Madison, WI, USA) for RNA integrity checks, double-stranded cDNA synthesis (NimbleGen Roche, Madison, WI, USA) and cDNA purification. cDNA was then submitted to NimbleGen Roche.
Array hybridization and initial data analysis cDNA was labeled with Cy3 and hybridized to a NimbleGen Arabidopsis thaliana 60-mer 4-plex 4x72k expression array (30,361 genes, 2 probes per target). Two technical replicates (two locations on one array) were performed. To generate expression data, quantile normalization [57] and robust multichip average (RMA) analysis [58] were performed (NimbleGen Roche, Madison, WI, USA). Hierarchical clustering was performed on the entire data set, using the average for technical replicates but keeping biological replicates separate, to allow visual inspection of reproducibility across replicates and to identify relative similarity of overall mRNA abundance profiles after the different treatments (Fig 2). For each gene, standardized transcript abundances were calculated with z scores to achieve mean = 0, standard deviation = 1.
To identify genes differentially regulated by MAMP treatment compared with no treatment, whose mRNA abundances after MAMP treatment were then altered by 3AB treatment or parg1-2 mutation, a number of criteria were applied to the entire data set (Fig 3). For After initial analysis (see methods), lists of genes differentially regulated between flg22 and flg22 + 3AB (A-D) or between Col-0 elf18 and parg1-2 elf18 (E-H) were assembled. Upregulated (red) (A, C, E, F) and downregulated (blue) (B, D, G, H) genes were determined separately. "Broken" genes were defined for this study as those genes that displayed statistically significant differences in mRNA abundance after treatment with MAMP (flg22 or elf18) (stringent cutoff = FDR <0.05, fold-change>1.3 or <-1.3) versus untreated, but for which MAMP treatment in the presence of 3AB or parg1-2 did not cause a statistically significant difference (even at the relatively permissive cutoff of p<0.05, fold-change>1.0 or <-1.0). "Misregulated" genes were defined as those genes upon which MAMPs had no statistically significant impact on mRNA abundance in wild-type plants, even using non-stringent cutoff values (p<0.05 and fold-change>1.0 or <-1.0), but for which MAMP treatment did cause significant differences (at stringent cutoff values) in the presence of either 3AB or parg1-2. To further reduce the occurrence of Type I false positives, the highlighted gene lists were then filtered once more for flg22 v flg22 + 3AB (p<0.05, fold-change>1.3 or <-1.3) or Col-0 elf18 v parg1-2 elf18 (p<0.05, fold-change>1.3 or <-1.3).

Hierarchical clustering of significantly regulated genes
Hierarchical clusters were generated from combined gene lists (Fig 4: 3AB broken genes + 3AB misregulated genes, parg1-broken genes + parg1-misregulated genes; Fig 5: elf18-regulated genes + flg22-regulated genes). The entire microarray dataset was filtered as described above to obtain the desired lists, and the resulting gene abundances were standardized by calculating z scores (mean = 0, standard deviation = 1). Row and column Euclidean distances were calculated, and row and column clusters were calculated using average linkage. Column clusters were colored based on a user-determined threshold based on visual analysis of the calculated clusters, and these were used to visualize patterns of gene expression across multiple treatments and/or genotypes. For further visualization, average gene intensities were then calculated for each gene for each relevant treatment/genotype within any one cluster.

Gene ontology enrichment analysis
The GO (gene ontology) analysis software BinGO [63] was used to analyze gene lists for GO term enrichment (Tables 1-4). ANOVA p-values for significantly overrepresented terms were calculated and larger gene lists are required to yield statistically significant output. Therefore, the statistical criteria used for generating the gene lists of Figs 4 and 5 and S1 Table were relaxed. For example, to generate a larger list of flg22-upregulated genes broken by 3AB, the untreated versus flg22 comparison criteria were relaxed from (FDR < 0.05, fold-change > 1.3 or < -1.3) to (FDR < 0.1, fold-change > 1.2 or < -1.2), while the subtraction criteria (such as for 3AB v. untreated and 3AB + flg22 v. 3AB) were relaxed from (p < 0.05, fold-change > 1.0 or < -1.0) to (FDR < 0.1, fold-change > 1.0 or < -1.0). This statistical "loosening" increased this example gene list size from 180 to 538 genes. GO terms that were overrepresented in those gene lists were then identified.

Knockout lines
Homozygous SALK T-DNA knockout lines for each of the selected genes were identified as described [64] (ABRC, Columbus, OH, USA). The parg1-2 mutant allele corresponds to SALK_116088; all other knockout lines are shown in S4 Table. Those knockout lines that showed altered MAMP-response phenotypes were then confirmed by RT-PCR to exhibit absent or reduced mRNA abundance for the designated knockout gene. Hierarchical clustering of gene products whose regulation by MAMPs are broken or misregulated by PARP inhibitor or parg1-2 knockout. A, C, Hierarchical clustering of transcript abundance for MAMP-regulated genes identified (see Fig 3) as exhibiting broken or misregulated expression due to 3AB treatment (A, n = 102) or due to parg1-2 mutation (C, n = 78). Each column represents a single gene and each row a single treatment (all treatments replicated three times); gene abundances standardized to mean = 0, standard deviation = 1.0; red = more abundant, blue = less abundant. Clustering was performed by calculating Euclidean distances on columns, using average linkage scores. B, D, Average transcript abundance (y-axis) for each treatment (x-axis), for each gene within the designated color-coded clusters shown at the top of A and C. Red lines denote overall mean mRNA abundance for all genes within the cluster, and grayscale lines represent each individual gene within the cluster.  , lists of genes differentially regulated between flg22 and elf18 were assembled. Upregulated (red) and downregulated (blue) genes were determined separately. Flg22-regulated genes not differentially regulated by elf18 in these experiments were defined as those that displayed statistically significant differences in mRNA abundance after treatment with flg22 versus untreated (FDR < 0.05, fold-change > 1.3 or < -1.3), which did not display statistically significant difference in mRNA abundance after treatment with elf18 versus untreated even using more permissive cutoff values (p < 0.05, fold-change > 1.0 or <-1.0). The reciprocal calculations were used to determine elf18-regulated genes not differentially regulated by flg22 in these experiments. B, Hierarchical clustering of transcript abundance for MAMP-regulated genes identified in A as exhibiting differential regulation by only one of the two studied MAMPs (A, n = 1897). Each column represents a single gene and each row a single treatment (all treatments replicated three times); gene abundances standardized to mean = 0, standard deviation = 1.0; red = more abundant, blue = less abundant. Clustering was performed by calculating Euclidean distances on columns, using average linkage scores. C, Average transcript abundance (y-axis) for each treatment (x-axis), for each gene within the designated color-coded clusters shown at the top of A. Red lines denote overall mean mRNA abundance for all genes within the cluster, grayscale lines represent each individual gene within the cluster. https://doi.org/10.1371/journal.pone.0190268.g005

Callose deposition
Twelve Arabidopsis seedlings per treatment were grown on 0.5× MS, 1.5% (w/v) sucrose, and 1× Gamborg's vitamins media for five days and then transferred to 24-well plates (one seedling per well) containing 400 μL of liquid 0.5× MS salts, 1.5% (w/v) sucrose, and 1× Gamborg's vitamins media. Seedlings were then treated with a final concentration of 1.0 μM flg22. After an additional 24h, seedlings were fixed overnight in formaldehyde/acetic acid/ethanol (FAA), cleared with 100% ethanol, and stained in 0.01% aniline blue (67mM K 2 HPO 4 pH12). Callose was visualized using ultraviolet epifluorescence microscopy [4]. Independent experiments were performed three times with similar results. In high-throughput screens, callose deposits Transcriptomics analysis of basal defense responses Table 2. Gene ontology over-representation in genes whose regulation by elf18 is broken or misregulated by parg1-2 knockout. were qualitatively scored using a scale of zero to five: 0, no callose deposits; 1, single isolated callose deposits across entire cotyledon; 2, very few deposits scattered across entire cotyledon; 3, moderate coverage of callose deposits across otherwise unstained regions; 4, many callose deposits covering entire surface of cotyledon but with some unstained space remaining; 5, uniform callose deposits covering entire surface of cotyledon. Callose deposits were also quantified by calculating area of callose deposits based on the difference in hue compared with the background hue of each leaf (ImageJ, NIH, Rockville, MD).  Number of genes in differential regulation gene list (see Fig 4) with the specified GO term relative to total number of genes in that gene list for Biological

MAMP
Process, Molecular Function, Cellular Component, or GoSlim Plant terms c Number of GO-annotated genes on array with specified GO term relative to total number of GO-annotated genes represented on array for Biological

Data deposition
Raw data and normalized calls from the microarray presented here have been deposited in GEO (http://www.ncbi.nlm.nih.gov/geo)), #GSE100205. A summary table with abundance values for each treatment replicate for all 60,770 gene probes, as well as average normalized call values and fold-change versus untreated Col-0, is also available as an Excel file (S5 Table).

Results
Expression profiling of plant response to MAMPs when poly(ADPribosyl)ation is altered An expression profiling experiment was designed primarily to characterize two phenomena: first, the effect of PARP inhibition on the transcriptional response to flg22; and second, the effect of knocking out PARG1 expression on the transcriptional response to elf18 (Fig 1A and  1B). Because the treatments for both experiments were carried out simultaneously for any given biological replicate, additional comparisons could be made to assess two additional comparisons: first, the differences between basal mRNA levels (without MAMP treatment) between wild-type, parg1-2 knockouts, and plants treated with PARP inhibitors; and second, the differences between wild-type Arabidopsis responses to flg22 and elf18. The experimental design included three replicates of nine genotype/treatment conditions (Col-0 treated with 3AB, 3MB, flg22, elf18, 3AB + flg22, or 3MB + flg22, or untreated; and parg1-2 untreated or treated with elf18) for a total of 27 samples. Principal component analysis was used to identify sources of variation in the data set; mean F ratios for batch (replicate), label (treatment/genotype), and error were 1.71, 17.20, and 1.00, respectively, indicating that batch contributed very little to variation, while genotype and treatment were responsible for most of the variation present. These F ratios indicate high reproducibility between biological replicates.

Impact of PARP inhibitors and parg1 knockout on untreated plants
The effect of PARP inhibitor treatment and PARG1 knockout on otherwise untreated or wildtype plants was first investigated. When all 30,387 predicted genes represented on the arrays were hierarchically clustered, 3AB treatment clustered closely with untreated plants, indicating very few pleiotropic effects (Fig 2). Only 228 genes displayed statistically significant alterations in mRNA abundances after 3AB treatment (contrast, for example, to flg22 treatment, in which 3402 genes displayed a significant difference in abundance using the same criteria). This result is consistent with earlier observations that 3AB-treated plants phenotypically resemble untreated plants [12], and that 3AB treatment does not induce visually evident generalized stress in plants in the absence of other stressors. In contrast, 3MB treatment elicited significant differences in the gene expression of 3935 genes relative to untreated plants. These data indicate that 3MB substantially impacts the physiological status of otherwise untreated plants grown in favorable conditions, and so subsequent analyses for PARP inhibition focused on data acquired using 3AB rather than 3MB. parg1-2 and wild-type Col-0 untreated plants also clustered closely in the above hierarchical cluster, with only 128 genes showing altered mRNA abundance between untreated Col-0 and parg1 plants, again indicating few pleiotropic effects in low-stress conditions (Fig 2).

Transcription profile of flg22 response in PARP inhibitor-treated plants
Within one hour of exposure, flg22 induces a wide range of transcriptional responses in Arabidopsis [67,68]. It was previously observed that 3AB treatment did not block flg22-induced ROS production or elevated expression of the signature flg22-induced genes WRKY29 and FRK1, but did disrupt a number of downstream events in the plant response to flg22 [7]. A primary purpose of the present study was to identify genes whose mRNA abundances are differentially regulated by flg22, but that are no longer differentially regulated by flg22 in the presence of 3AB (referred to here as "broken" genes), as well as those genes up or downregulated by flg22 only in the presence of 3AB (referred to here as "misregulated" genes). The specific tests used for gene selection, and the number of genes in each sub-group, are presented in Fig 3A-3D, and the complete gene lists are provided in S2 Table. The 102 genes present in the final gene lists generated as described above were then visualized by hierarchical clustering (Fig 4A). They grouped in four self-organizing clusters C1- C4   Fig 7. Callose deposition assay. A. 10-day-old Arabidopsis seedlings were treated with distilled, deionized water (H 2 O) or 1 μm flg22, fixed 24 h after flg22 elicitation, and visualized for callose deposition by aniline blue staining and epifluorescence microscopy. Degree of callose deposition was categorized using a scale of 0 to 5, 0 = no callose deposits, 5 = dense callose deposits over entire field of view. Twelve cotyledons per genotype were examined and compared to wild-type (Col-0) responses per biological replicate (n). pmei-2 (At5g64640) n = 1; rba-2 (At1g47370) n = 5; f-box-1 (At5g15660) n = 4; fpgs3-1 and fpgs3-2 (At3g55630) n = 4, n = 1, respectively. Asterisks summarize ANOVA results across all experiments for tests of similarity of means between the mutant genotype and wild-type plants treated with flg22 (Tukey's simultaneous test: †P<0.15; † †P<0.1; *P<0.05; **P<0.005; no asterisk, P > 0.05). B. For selected lines, callose deposits were quantified as average percent area covered by white pixels within the viewfield, corresponding to flg22-induced callose, +/-standard error. Ten cotyledons per genotype were examined for each of four biological replicates. Asterisks summarize ANOVA results across all experiments for tests of similarity of means between the mutant genotype and wild-type plants treated with flg22 (Tukey's simultaneous test: *P < 0.055; **P < 0.005; no asterisk, P > 0.05). Representative flg22-treated cotyledons for each genotype in B are shown in C. https://doi.org/10.1371/journal.pone.0190268.g007 Transcriptomics analysis of basal defense responses (Fig 4B). C1 and C2 contain those genes normally not under flg22 regulation that are either flg22 up-or downregulated, respectively, in the presence of 3AB (3AB-misregulated genes). C3 and C4 contain those gene products up-or downregulated by flg22 that are no longer differentially regulated by flg22 in the presence of 3AB (3AB-broken genes).
Transcription profile of elf18 response in parg1 knockout plants Two independent parg1 mutant alleles have been previously shown to display a number of defense phenotypes, including hyper-responsiveness to elf18 and greater susceptibility to necrotroph infection [7]. For the present study, experiments were designed to detect early transcriptional events after elf18 treatment that are disrupted or overactivated by loss of PARG1. Genes were identified whose mRNA abundances were affected by elf18 in wild-type plants but not in parg1 mutant plants (parg1-broken genes), as well as those regulated by elf18 only in parg1 knockout plants (parg1-misregulated genes). Logic similar to that used for the PARP inhibitor experiments was used (Fig 3), generating a stringent list of those genes whose response to elf18 is broken or misregulated in parg1-2 plants (S1 Table).

GO-term analysis of flg22-and elf18-regulated transcriptional responses that are altered by disruption of poly(ADP-ribosyl)ation
To identify specific cellular pathways that may be regulated by poly(ADP-ribosyl)ation during MAMP-elicited immunity, we analyzed the above lists (Fig 3) of broken and misregulated genes for overrepresentation of Gene Ontology (GO) terms. Among genes classified as broken by 3AB, overrepresented GO terms include nucleolus, chloroplast (membrane and thylakoid membrane), receptor activity, and phospholipase C activity. Overrepresented GO terms for 3AB-misregulated genes included chloroplast (photosystem and thylakoid membrane genes), lipid metabolism, and chromatin silencing related genes ( Table 2).
Some of the GO terms overrepresented among those genes misregulated during MAMP response in a parg1-2 mutant background include COPII vesicle coat, GPI anchor biosynthesis, and xanthophyll biosynthesis genes. GO terms that were overrepresented among those genes whose expression was broken during MAMP response in a parg1-2 mutant background include protein and RNA processing, mitochondrial translocation, and, of particular interest, response to other organisms/innate immune response genes (Table 3). Table 4 lists those genes associated with the GO term "innate immune response" that are normally upregulated by elf18 and are then no longer activated by elf18 in parg1-2 seedlings. These genes do not fall into a single pathway or branch of any known defense response pathway, but are instead spread across a number of aspects of the innate immune response. Regulation of expression of a number of processes associated with plant defense are altered by the absence of the PARG1 gene product, including calcium influx (AtCNGC12), NPR1-regulated transcription (TGA3), trans-acting anti-viral dicer-like activity (DCL4), recognition of pathogen effector molecules (PBS1), defense-activated tryptophan and chorismate production (ASA1), oxidative stressinduced flavonoid glycosylation (UGT73B2), the defense gene PR5, and the R gene RPP13.

The MAMPs flg22 and elf18 elicit separate, yet overlapping transcriptional responses
The hypothesis that flg22 and elf18 elicit the same set of basal defense responses was also examined. To identify genes in our experiment that were regulated only by flg22 or elf18 but not both, a modified version of the Venn diagram method described above (see Fig 3 legend) was used (Fig 5A), and genes up or downregulated only by on or the other MAMP were identified (S3 Table). These gene lists were then combined to perform hierarchical clustering (Fig 5B) and self-organizing cluster analysis (Fig 5C). It is noteworthy that elf18 regulated most of the same genes as flg22, but flg22 elicited a set of additional responses above and beyond those shared by both MAMPs. Approximately 2000 genes showed altered mRNA abundances with only one of the two studied MAMPs. Up to 20% of flg22-regulated gene products were not regulated by elf18, whereas only 3-4% of elf18-regulated gene products were not also regulated flg22. Table 5 outlines the overrepresented GO terms present in these clusters of gene products regulated by only one of the two MAMPs studied. Of particular interest are a number of defense genes only induced one hour after treatment by flg22 and not by elf18. These genes include a number of TIR-NBS-LRR class genes, as well as PAL1, Respiratory Burst Oxidase Homolog A (RBOHA), WRKY70 and CDPK-Related Kinase 4 (CRK4). In addition, a number of auxin-responsive gene products showed reduced abundance only after flg22 treatment, as well as other auxin-responsive genes upregulated only by elf18.

Gene knockout studies for candidate genes
Disrupting poly(ADP-ribosyl)ation mechanisms disrupts a subset of plant innate immune responses [7,12], including callose deposition and seedling growth inhibition. We therefore hypothesized that knockouts of some of the genes identified in the above transcriptional profiles (ie., those genes downstream of PARP or PARG activation) will alter some plant defense responses. Homozygous Arabidopsis T-DNA knockout plant lines for genes selected from a subset of the categories in Figs 3 and 4, the gene products whose regulation by MAMP is then broken or misregulated by PARP inhibitor treatment or parg1-2 knockout. Of the 178 genes represented by lists C1-C4, 54 homozygous T-DNA knockouts were available and initially screened (S4 Table). We did not investigate genes with previously confirmed roles in plant defense. Plants were first screened for altered MAMP-induced seedling growth inhibition and callose deposition, two downstream indicators of detection and response to MAMPs. Of the 54 plant lines screened, one (At3g55630) showed attenuated seedling growth inhibition and 3 (At1g47370, At3g55630, and At5g15660) displayed reduced callose deposition phenotypes, including one knockout (At3g55630) with both seedling growth inhibition (Fig 6) and callose phenotypes (Fig 7). Table 1 provides a summary of these experiments, including tests with second knockout alleles that are discussed below.

A folylpolyglutamate synthetase is required for wild-type responses to MAMPs
At3g55630 was identified from the microarray as a parg1-2 misregulated gene whose expression is only downregulated by elf18 in parg1 knockout plants, but not in wild-type plants (-1.4-fold downregulated compared to efl18-treated Col-0). This gene encodes a cytosolic FPGS (FPGS3) that adds polyglutamate tails to folate and folate derivatives. A T-DNA line knocking out expression of this gene (Salk_038762C) is less responsive to a nonsaturating concentration of flg22, as measured by seedling growth inhibition. This fpgs3 knockout line is also deficient in flg22-induced callose deposition. These plants also displayed fewer stained callose deposits than wild-type seedlings (Fig 7).
T-DNA lines knocking out the other FPGS genes encoded by the Arabidopsis genome were also studied. Two separate alleles of both fpgs1 and fpgs3 knockout plants produced statistically significantly more callose in response to flg22 than Col-0 or fpgs2 knockout plants (Fig 7C). Additionally, the fpgs1-2 (SALK_032544C) and fpgs2-2 (SALK_821517C) alleles displayed statistically significantly less seedling growth inhibition than wild-type, fpgs1-1, and fpgs2-1 plants (Fig 6E).
Knocking out expression of an F-box gene, a TIR-domain containing gene, and a pectin methylesterase inhibitor disrupts flg22-induced callose deposition At5g15660 (Salk_063563c, Salk_047400c) was identified from the microarray as a gene whose flg22-induced upregulation is broken by 3AB treatment (-3.5-fold downregulated compared to flg22-treated Col-0 plants). These knockout plants had fewer callose depositions than wildtype plants (Fig 7).
The gene At5g64640 encodes a pectin methylesterase inhibitor protein. This gene was identified through our microarray analysis as a parg1-2 misregulated gene whose expression is only downregulated by elf18 in parg1 knockout plants, but not in wild-type plants (-1.3-fold downregulated compared to elf18-treated Col-0 plants). A single At5g64640 knockout allele (Salk_063593c) shows attenuated responses to flg22 in the form of both reduced seedling growth inhibition (Fig 6) and fewer callose depositions (Fig 7). This seedling growth inhibition phenotype was not repeatable in a second allele (SALK_121275C), or at higher, saturating flg22 concentrations. Salk_063593c plants also had fewer callose depositions than wildtype (Fig 7).
The gene At1g47370 encodes a TIR (Toll-Interleukin-like Receptor) domain-containing protein. This gene was identified through our microarray as being upregulated by elf18 only in the parg1 knockout background (2.2-fold upregulated compared to elf18-treated Col-0 plants). At1g47370 knockout plants (Salk_037091 and Salk_024124), show wild-type seedling growth inhibition (Fig 6), but are deficient in flg22-induced callose deposition (Fig 7). In preliminary studies all of the above-described knockout plants showed wild-type ROS burst and defense gene induction in response to flg22.

Discussion
Gene expression profiling studies provide a number of opportunities for further characterization of molecular pathways of interest. For example, microarray data can be used to identify genes and acquire knockouts for a targeted mutant screen, to examine cis-regulatory element overrepresentation or Gene Ontology term overrepresentation, to conduct other pathway/biological process analyses, to compare with other published microarrays for common trends, and to manually examine expression patterns displayed by particular genes of interest [69][70][71].
The present transcriptomics study has opened up a number of research opportunities, some new and some previously implicated, for further study of the molecular mechanisms of both poly(ADP-ribosyl)ation and plant innate immunity and the links between them. We previously hypothesized putative roles for poly(ADP-ribosyl)ation in a subset of innate immune responses, including MAMP-regulated cell wall modifications, phenylalanine ammonia lyase activity, and genotoxic stress responses (Adams-Phillips 2008; Adams-Phillips 2010). This study was designed to investigate these hypotheses and provide a finer focus on how PARP, PARG and poly(ADP-ribose) may affect basal defense responses in plants. The following sections of this discussion address these hypotheses, as well as other new insights gained.

PARP and PARG activity in the absence of MAMP exposure
A general examination of mRNA abundances across all treatments provided insights into the roles of PARP and PARG in untreated, unstressed plants. In this study, hierarchical clustering placed the overall transcriptome of 3AB-treated plants closest to untreated parg1 mutants and wild-type plants (Fig 2), and in our hands 3AB-treated plants have appeared healthy (present study and Adams-Phillips 2008). However, we did identify genes that exhibit differential mRNA abundance between untreated and 3AB-treated Col-0 plants. Future study of those genes may offer insights into roles of PARP activity in relatively unstressed plants. The data from this study are available for other inquiries (S5 Table And GEO (http://www.ncbi.nlm.nih. gov/geo) #GSE100205). For example, although representing potentially problematic differences in mode/timescale of pathway inhibition, comparison of 3AB-treated plants with parg1 plants might provide some insight into the transcriptional effect of depletion vs. accumulation of polyADP-ribose. The genome-wide hierarchical clustering presented in Fig 2 also demonstrates why further analysis of 3MB treatment data was not pursued -3MB treatment caused a large number of mRNA abundance changes, suggesting pleiotropy and off-target effects of that PARP inhibitor. 3AB is likely to also have off-target impacts beyond its demonstrated inhibition of PARP, but apparently those impacts are much more limited.
Hierarchical clustering placed the overall transcriptome of parg1 mutants closest to 3ABtreated plants (Fig 2), which itself is noteworthy. However, close inspection of their respective lists of genes differentially expressed relative to wild-type plants revealed an intriguing defense-related observation. Ten defense-related NB-LRR genes exhibited increased rather than decreased mRNA abundance in the parg1 mutant. The de-activation of these genes in PARP inhibitor-treated plants and their reciprocal activation in parg1 knockout plants suggests further possible impacts of plant poly(ADP-ribosyl)ation on plant immune responses. This type of observation also demonstrated the need for the subtraction scheme used above (Fig 3) to generate lists of genes whose mRNA abundances were truly affected by both MAMP treatment and either 3AB or parg1 knockout.

Use of GO term overrepresentation analysis as a discovery tool
Unlike many other transcriptomics studies [11] GO term overrepresentation analysis yielded few results that motivated us to further investigation in the current study. The fact that a "loosening" of statistical criteria was required in order to perform a such an analysis may partially explain this-most of the genes in the lists were only 1.2-fold up or downregulated, with q < 0.1. We therefore focused more on targeted knockouts and manual examination of gene lists. However, over-represented GO terms for this gene expression study included "chloroplast membrane", "chloroplast thylakoid membrane" and "photosystem" for flg22/3AB plants, suggesting possible stress points that arise when MAMP responses are disrupted by PARP inhibition. The "innate immune response" GO term was significant for elf18/parg1, bringing attention to specific elf18-activated defense-associated genes whose expression is altered in parg1 plants. As with all microarray studies, only a limited set of conditions could be analyzed in the current study. We chose to focus our analysis on the plant response to flg22 and elf18 peptides at one hour after treatment, but it is likely that other time points and/or other pathogen-associated stimuli could uncover additional relevant pathways. Other users with different viewpoints may also be able to exploit not only the specific genes on the gene lists of this study, but also the GO analyses, to help shape future investigations.

Poly(ADP-ribosyl)ation regulates a subset of innate immune responses
We previously reported that ROS synthesis and induction of some early defense genes remained intact in the presence of 3AB while callose and lignin deposition and PAL activity were inhibited [7]. A wide array of gene products display altered mRNA abundances upon MAMP treatment (up to a quarter of the genome, in fact), but only a small subset of those MAMP-regulated transcriptional responses is altered by PARP inhibitor or parg1 knockout. This observation supports our previous findings (Adams-Phillips 2010) that only some-not all-MAMP-regulated basal defense responses are altered by disrupting poly(ADP-ribosyl) ation.
PARP inhibition disrupted a number of MAMP-regulated transcriptional responses that involve cell wall-related genes, further demonstrating a role for poly(ADP-ribosyl)ation in defense-related cell wall reinforcement. In the presence of flg22, 3AB downregulated a synbindin ER to golgi transport gene (At5g02280). PARP inhibition also disrupted the flg22-regulated accumulation of an alpha-glucosidase (oligosaccharide metabolism) transcript (At1g24320) and, perhaps most interestingly, a glyoxal oxidase-related gene product necessary for the production H 2 O 2 for ligninolytic peroxidases (i.e., lignin biosynthesis) (At3g57620). 3AB treatment also disrupted the flg22-induced downregulation of a cellulose synthase gene (CESA10, At2g25540). And while no defense-related cell wall phenotypes have yet been reported for PARG1 knockout plants, PARG1 knockout did disrupt elf18-regulated upregulation of a pectate lyase transcript (At2g02720) while downregulating a pectin methylesterase inhibitor gene (At5g64640).
Poly(ADP-ribosyl)ation may also regulate the innate immune response via FPGS enzymes, specifically FPGS3 (At3g55630). FPGS3 attaches glutamate residues to folate and folate derivatives, especially THF-10 [72]. This isoform of FPGS is found in the cytosol, the location in which purine nucleotide synthesis primarily occurs [73]. Purine nucleotide synthesis is necessary for poly(ADP-ribosyl)ation, as the ADP-ribose units contain adenine. Further connections of FPGS3 to the innate immune response are discussed below.
Our previous report of 3AB blocking MAMP-induced PAL activity is also supported by the current microarray study. Expression of an AMP-synthetase family gene involved in phenylpropanoid metabolism (At1g20490) was downregulated when 3AB was added to flg22 treatment. This gene could provide a target for further examination into the mechanism(s) by which poly(ADP-ribosyl)ation regulates defense-related PAL activity and phenylpropanoid pigment production.

Poly(ADP-ribosyl)ation at an intersection between plant defenses and DNA repair
A role for poly(ADP-ribosyl)ation in plant genotoxic stress responses has been demonstrated [10,29,42,52,53]. Early induction of host DNA double-strand breaks after infection by plant pathogens has been demonstrated [27]. The reactive oxygen species (ROS) burst activated by plant immune systems in response to pathogens has also been proposed to induce DNA damage that the plant would have to protect against or repair. Therefore it is not surprising that some DNA repair and reduction-oxidation homeostasis-related genes were either broken or misregulated by 3AB treatment or PARG1 knockout. PARP inhibition broke the flg22-induced upregulation of oxidoreductase (At3g13610), glutathione-S-transferase transcripts (At1g02940 and At1g78340), and the flg22-regulated downregulation of a glutaredoxin family gene involved in protein disulfide oxidoreductase activity (At5g39865). In the presence of elf18, parg1 plants had reduced expression of RAD51B (a double-stranded DNA repair gene) (At2g28560) and SWI1 (involved in sister chromatid cohesion and chromosome organization) (At5g51330). The previously reported observation that PARP inhibition leads plants toward unproductive, toxic outcomes after exposure to MAMP [12] may also reflect a need for intact poly(ADP-ribosyl)ation machinery to protect the host from its induced defenses. Future studies that quantify reduction-oxidation and energy homeostasis inside plant cells undergoing defense responses (with and without disruption of poly(ADP-ribosyl)ation) could provide further insight into this hypothesis.

Knocking out PARG1 expression disrupts defense gene expression
The expression levels of a number of defense genes were also altered by PARG1 knockout. The elf18-induced downregulation of a gamma interferon responsive lysosomal thiol reductase (which, in animals, recruits immunity-related MHC complexes to the plasma membrane) (At4g12890) was broken in parg1 plants. PARG1 knockout also broke elf18 induction of a number of other genes ( Table 3) that are not common to any one signaling pathway, suggesting that PARG activity may be required not for any one specific pathogen-stimulated pathway, but rather at many junctures. This result is consistent with our previously-published results that PARG2 gene expression is strongly upregulated by a wide variety of defense-associated stimuli, including MAMPs, biotrophic bacteria, necrotrophic fungi, and in constitutive defense mutants (Adams-Phillips 2010), despite the fact that parg2-1 knockout plants remain visibly healthy. Such generalized responses may indicate roles for PARG and poly(ADP-ribosyl)ation in the plant host's protection mechanisms of protection from its own induced defenses.

Different MAMPs activate separate, yet overlapping, basal defenses
For many years, flg22 and elf18 peptides have often been used interchangeably as elicitors of basal plant defenses. Recognized by the separate PRRs FLS2 and EFR, respectively, they have been suggested to elicit very similar sets of downstream responses. And while many microarray studies over the years have examined either flg22 or elf18 transcriptional responses [74][75][76][77], the present study examined responses to both peptides within the same microarray experiment. Our analysis indicates that in our experiment system, 1690 genes were regulated by just flg22, while only 207 genes were differentially regulated by elf18 alone. From this result it is clear that the two receptors activate some of the same pathways, but with very different individual behaviors. Another clue that such differences in MAMP-elicited defenses exist could also be seen in our previously reported studies [7,12], which guided the experimental setup for this transcriptomics investigation (some effects of 3AB and/or PARG1 knockout on basal defenses were only observed for flg22 or elf18 and not the other). This was a major reason why two different MAMPs were used in the current investigation. More recent investigations with EFR and FLS2 have demonstrated specific elements of signaling that are more prominent for one of the two receptors [77][78][79]. FLS2 is an evolutionarily older protein than EFR that is much more broadly distributed across plant families [80]. This may account for the larger set of gene products regulated only by FLS2, if FLS2 has had more time to gain new and different functions in addition to the core set of basal defenses that it activates in common with EFR.

Dissecting roles of poly(ADP-ribosyl)ation in basal defense responses
Screens of T-DNA knockouts chosen based on the gene lists generated in this study identified five candidate genes that may play a role in poly(ADP-ribosyl)ation's impacts on basal defense responses. These knockouts were identified based on altered seedling growth inhibition and/ or callose deposition phenotypes in response to flg22. We screened 135 Arabidopsis T-DNA lines and identified one showing altered seedling growth inhibition and four showing altered callose deposition phenotypes. The reduced growth inhibition and callose deposition suggest that the plants may not be properly detecting the presence of MAMP, or that their ability to respond to MAMP is compromised.
At3g55630 was identified from the microarray as a parg1-2 misregulated gene whose expression is only downregulated by elf18 in parg1 knockout plants, but not in wild-type plants. This observation suggests that PARG1 protein may be necessary for de-repression of FPGS expression in response to MAMP elicitation. This gene encodes a cytosolic folylpolyglutamate synthetase (AtFPGS3) that adds polyglutamate tails to folate co-enzymes, enhancing their stability and affinity [72]. A T-DNA line knocking out expression of this gene (Salk_038762c) is less responsive to 0.05 μM flg22, as measured by seedling growth inhibition, and callose deposition. Additionally, a T-DNA line knocking out expression of the FPGS1 isoform of FPGS (Salk_032544C) is also less responsive to flg22 than wild-type. These altered responses suggest that at least one folate-dependent enzyme-many of which are involved in methionine biosynthesis and photorespiration [72]-may be necessary for wild-type levels of basal defense response activation.
An intriguing link exists between folylpolyglutamylation and another pathway implicated by our knockout screens: pectin methylesterification. fpgs3 single mutant plants have reduced levels of methionine and methylesterified pectins [72]. At5g64640 encodes a pectin methylesterase inhibitor that was downregulated by elf18 only in parg1 knockout plants, displaying a similar expression pattern to FPGS3, and further implicating poly(ADP-ribosyl)ation in the regulation of MAMP-induced cell wall reinforcement. At5g64640 knockout plants show a reduced response to flg22 peptide, with fewer callose deposits compared to wild-type plants when treated with flg22. Pectinesterases are a group of cell-wall localized proteins that catalyze the de-esterification of methyl-esterified D-galactosiduronic acid units in cell wall pectins, creating acidic pectins that can lower cell wall pH, allowing cell expansion and growth, and leading to either cell wall stiffening (via production of pectate gels in the apoplasm) or cell wall loosening (via proton-stimulated activity of cell wall hydrolases) [81][82][83]. Pectinesterases are known to be involved such physiological responses as pollen tube growth and abscission. A pectin methylesterase inhibitor protein was demonstrated to be necessary for basal disease resistance and antifungal activity in pepper [84,85], but this enzyme's role in plant defenses remains unclear. In this current study, the absence of PARG1 led to downregulation of a pectin methylesterase inhibitor gene. Further studies will be required to characterize both the role that PARG1 plays in this pectin methylesterase inhibitor's gene expression, as well as the function of this cell wall-associated enzyme in MAMP-induced basal defenses.
Beyond containing conserved F-box, there is little known about the other two genes identified in our basal defense screen, At5g15660 and At1g47370. F-box protein domains regulate protein-protein interactions, creating links between target proteins and ubiquitin-conjugating enzymes [86,87]. TIR-X proteins resemble TIR-NB-LRR resistance proteins, but lack the nucleotide binding and leucine-rich-repeat domains. When activated by pathogen effectors, resistance (R) proteins activate a wide range of defense responses, and TIR-X proteins have been hypothesized to modulate the activity of some TIR-NB-LRR proteins [88,89]. RBA1 (At1g47370), the TIR-X protein identified in our screen, has recently been shown to induce cell death in response to the pathogen effector HopBA1.
In summary, disruption of poly(ADP-ribosyl)ation processes alters the expression of a number of plant genes during the early stages of the response to the MAMPs flagellin (flg22) and EF-Tu (elf18). Knowledge of the specific genes that are impacted can contribute to the generation of hypotheses for future research regarding poly(ADP-ribosyl)ation and/or more general aspects of plant responses to pathogens.
Supporting information S1