Plant Core Environmental Stress Response Genes Are Systemically Coordinated during Abiotic Stresses

Studying plant stress responses is an important issue in a world threatened by global warming. Unfortunately, comparative analyses are hampered by varying experimental setups. In contrast, the AtGenExpress abiotic stress experiment displays intercomparability. Importantly, six of the nine stresses (wounding, genotoxic, oxidative, UV-B light, osmotic and salt) can be examined for their capacity to generate systemic signals between the shoot and root, which might be essential to regain homeostasis in Arabidopsis thaliana. We classified the systemic responses into two groups: genes that are regulated in the non-treated tissue only are defined as type I responsive and, accordingly, genes that react in both tissues are termed type II responsive. Analysis of type I and II systemic responses suggest distinct functionalities, but also significant overlap between different stresses. Comparison with salicylic acid (SA) and methyl-jasmonate (MeJA) responsive genes implies that MeJA is involved in the systemic stress response. Certain genes are predominantly responding in only one of the categories, e.g., WRKY genes respond mainly non-systemically. Instead, genes of the plant core environmental stress response (PCESR), e.g., ZAT10, ZAT12, ERD9 or MES9, are part of different response types. Moreover, several PCESR genes switch between the categories in a stress-specific manner.

While it still remains unknown what kind of roles these genes might play, it seems evident that a common battery of responses can be triggered by various stresses.
In analogy to the common environmental stress responses in yeast, we identified sets of genes that were commonly responsive in several stresses of the AtGenExpress abiotic stress experiment [3,4,34]. A considerable portion of the immediate early responsive genes is differentially expressed in more than only one stress treatment and might represent a plant core environmental stress response (PCESR). Importantly, this PCESR is not restricted to Arabidopsis thaliana, but appears to be conserved between different plant species, such as rice, barley or wheat [3,[35][36][37]. These early-induced, common genes are not related to those responsible for the CESR in yeast cells, which suggests that distinct stress response pathways are required during plant evolution [4,32,33].
Here, we make use of the intercomparability of the nine stresses of the AtGenExpress abiotic stress experiment [4,34]. Six of these stresses were applied to either the shoots or the roots and, hence, allow the comparative investigation of the systemic environmental stress response. We found that gene expression trajectories can be classified into three distinct categories: non-systemic (regulated in the stressed tissue only), systemic type I (regulated in the non-treated tissue only) or systemic type II (regulated in both tissues). Although these categories display distinct functional groups, they still partially overlap between the different stresses and response types. Moreover, meta-analysis suggests that several of the stress responsive gene loci are also SA or JA responsive. We show that type II systemic expression responses are more stress specific than others. In contrast, PCESR genes are common to several abiotic stresses, but are rather promiscuous for the response types. Most PCESR genes have paralogs that transduce presumably redundant information of the incoming stresses, which reveals a possible backup function.

Gene Expression Responses Follow a Highly Diverse Pattern during Abiotic Stresses
The analysis of the AtGenExpress abiotic stress experiment disclosed a plethora of gene expression responses also in those tissues that were not directly exposed to the stress treatment [4,34]. We observed both non-systemic and systemic responses to characterize a specific abiotic stress response in Arabidopsis thaliana. To dissect these responses and to provide a detailed description of our observations, we need to extend the basic terminology into three principal response types of gene expression ( Figure 1). Genes that are differentially expressed in the treated tissue only display a non-systemic response ( Figure 1A). We next divided the systemic responses into two categories. A type I systemic response displays gene expression changes exclusively in the non-treated tissue, while, conversely, a type II systemic response is characterized by gene expression changes in both the treated and non-treated tissue ( Figure 1A). We exemplified these three principal response types for a subset of genes that were responsive after UV-B light treatment ( Figure 1B).

Systemic and Non-Systemic Gene Expression Responses
We applied these definitions to our previously described multidimensional AtGenExpress abiotic stress microarray dataset, which currently provides the unique opportunity of intercomparability between nine different environmental conditions or between the shoot and the root tissue [3,4,34]. Most important for our present analyses are six stresses, i.e., genotoxic, osmotic, oxidative, salt, UV-B and wounding stress, that were applied either to the shoot or the root exclusively. Nevertheless, the three remaining stresses, i.e., drought, cold, and heat, serve as our controls for data validation, because these stresses were simultaneously applied to both tissues and, hence, we can expect overlapping local and systemic responses throughout the entire organism.
To unambiguously categorize differentially expressed genes into the three response types, we applied a chronology-dependency filter to the previously described sets of differentially expressed genes [4]. We require that only those genes are classified, which exhibit differential expression in two consecutive samples with the same trend in both of the replicates. This approach effectively removes outliers and fluctuating gene expression pattern from time series datasets. On the one hand, we are aware that relevant stress-specific information is lost in this process, especially in comparison with the many genes previously identified [4,34]. On the other hand, plain and unequivocal expression response patterns of high informative value are disclosed. These non-systemic, type I and type II systemic responses provide precious information, which are otherwise hidden within complex and partially overlapping expression trajectories.
Indeed, each stress experiment uncovered genes that could be placed in one of the three categories, which indicates the elicitation of systemically activated gene expression after any treatment. In most stress treatments, the highest number of regulated genes is observed locally in the non-systemic organ directly exposed to the stimulus (Table 1; Table S1). In general, the salt, osmotic and UV-B stress treatments affect more genes, compared with genotoxic and wounding stress. There are more systemic type I responsive genes found during the osmotic and oxidative stresses than during the non-systemic response, which underlines a rapid alarm signaling mechanism between the treated and the systemic organs. Table 1. Number of genes involved in a systemic or non-systemic stress response. Genes were classified as regulated under high-stringency conditions, as was described in the text. Identification of significantly regulated genes was conducted for shoot and root datasets independently. Detailed lists of genes are given in Table S1. Genotoxic 257  220  99  158  37  62  Osmotic 2,687 1,185 2,066  621  1,502  564  Oxidative  28  11  18  10  17  1  Salt  1,934 1,664 456  1,478  270  186  UV-B  1,166 146 1,100  1,020  66  80  Wounding 453  19  440  434  13  6 Most noteworthy, there was only a single gene of yet unknown function (At3g20340) that was type II responsive during oxidative stress (Table 1). In addition, more type I than type II responsive genes are regulated during the osmotic, salt, wounding and oxidative stress treatments.

Functional Categorization Uncovers Specificity in the Different Response Patterns
A subset of gene ontology (GO) terms was used to classify genes with respect to their putative function ( Figure 2). Remarkably, the comparison between all differentially expressed genes and the non-systemic, type I and type II responsive genes revealed significant differences in the GO terms. There is one exception, however, nearly all conditions and categories are enriched for the GO term "stress response," which is consistent with the types of experiments.
As all of our distinct stress response types contain genes that are linked with "stress response", we propose that several of those genes are non-specifically responding to many different stress stimuli. This observation can be explained by gene sets that are shared between the GO categories of different treatments, as was proposed before [4], or by genes that switch between our response types in a stress-dependent manner. Percentages of genes categorized into the seven well-annotated GO terms "developmental processes", "protein metabolism", "responses to stress", "transcription", "transport", "chloroplast", or "mitochondria" are given for each of the six stress responses. Number of genes, observed GO counts and p-values can be found in Table S2.
Besides the ubiquitous "stress response", it is noteworthy that some responses involve the categories "developmental processes", "transcription", or "organelles". We noted already that these functional categories are common to almost all stress responses and are shared between different plant species [3]. In addition, significant GO terms differ between the response types and did not correlate with the number of genes [osmotic vs. salt vs. UV-B] or [genotoxic vs. wounding] and, hence, indicate an overall difference in stress response pattern.

Overlap in Non-Systemic, Type I and Type II Systemic Stress Responses
We have described that considerable overlap between stress responses exists [34]. Especially diverse and seemingly unrelated stresses such as cold, drought and UV-B light stress treatments share a set of common genes, in addition to a generally independent transcriptional response pattern [4]. To gain a more detailed insight into the specificity of systemic or local stress responses, we next examined the responsive gene overlap between our non-systemic, type I systemic and type II systemic datasets amongst the six different stresses.
A hierarchical graph of the stress responsive gene sets was generated ( Figure 3): nodes (gene sets) and directed edges (connecting arrows) are shown proportionally to the total number of genes. As the number of genes inside the sets varies from very few to many regulated genes (Table 1), we choose 10% as the minimal shared rate to equally represent all datasets. The graph separates into two major clusters in the non-systemic expression responses with overlap in the salt and osmotic or the UV-B light and wounding stress responses. This split represents the major differences between the root or the shoot [4]. We can conclude that the principal component of the tissue, to which the stress was applied, dominates over the response type ( Figure 3). This tissue-dependent expression is also found in the type I systemic gene sets, even though they are triggered by treatments, which are applied in the non-systemic tissue.
In contrast, the oxidative stress treatment does not activate a unique set of genes and, instead, its genes are found responsive within the other stresses. This is consistent with previous reports about the general involvement of reactive oxygen species (ROS) in many different stresses [18,19,21,34]. Similarly, the genotoxic treatment induces sets of gene expression changes that overlap with many other stresses, but exhibits also specific non-overlapping expression trajectories.
Systemic type II genes have a more complex patterning when compared to the non-systemic or type I systemic gene sets. Interestingly, the type II systemic responses display only little overlap between the type II responses of the other stresses. Instead, significant overlap with the response types of the other stresses exists. Thus, type II systemic response genes are more stress specific than non-systemic or type I systemic genes. In contrast, distinct sets of non-systemic or type I systemic genes respond independently in either the shoot or the root.

The Systemic Stress Response Utilizes Jasmonic Acid
We have shown that the successful stress response requires both the coordinated systemic and non-systemic expression responses. In addition, there is significant overlap between the stress responses in both the treated and the non-treated tissues, which requires yet unknown systemic signaling modules for the rapid coordination of gene expression throughout the entire plant.
To address whether already known signaling molecules such as salicylic acid (SA) or methyl-jasmonate (MeJA) might at least partially be involved in the systemic expression responses, we compared our gene lists with already published microarray data on SA and MeJA signaling ( Figure 4).
We find SA and MeJA to be involved in the systemic stress responses, as has been described before [14][15][16]19,38,39]. Interestingly, there is considerable overlap between SA responsive and the non-systemic responsive genes irrespective of the kind of stress. A similar observation is found for MeJA and, thus, both hormones are important for the local non-systemic response. In contrast, type I and type II systemic responses predominantly share genes with the MeJA responses, while the overlap with SA is negligible in the systemic tissue. A portion of genes appears to be under the control of both phytohormones in the non-systemic tissues.
These data suggest that MeJA is a good candidate for the coordination of gene expression in the systemic stress response, whereas the concerted action of SA and MeJA is implicated in the local non-systemic responses.

WRKY Genes Are Mainly Involved in the Non-Systemic Stress Response
We have shown that there is a considerable amount of overlap between different stress responses that involves MeJA for its systemic coordination. Gene expression is controlled by transcription factors, and the WRKY transcription factor genes are well known to be responsive during biotic and abiotic stresses [40][41][42][43]. To investigate whether these transcription factors might be involved in the coordination of either the type I or type II stress responses, we compared our datasets with all 61 WRKY genes present on the microarray. The majority of the WRKY genes is not involved in the systemic stress response, but implies a substantial function almost exclusively in the treated non-systemic tissue (Table 2). Consistently, we analyzed the occurrence of cis-regulatory elements in each of the response types and found that the cognate W-box binding motif of WRKY transcription factors is exclusively enriched in the non-systemic responsive genes (Table S3).

Plant Core Environmental Stress Response (PCESR) Genes
Besides the stress specific gene expression responses, it was noted that a group of genes exists that is differentially expressed in almost any biotic or abiotic stress condition. Therefore, we mined the AtGenExpress abiotic stress experiment for putative Plant Core Environmental Stress Response (PCESR) genes that are involved in most of the nine stresses, as was suggested also in previous publications [3,4]. One example of such a gene that meets this definition is ZAT12, which is regulated by light, ROS, oxidative, heat, UV-B, cold and drought stress [4,44,45]. Hence, we used ZAT12, which encodes a putative zinc-finger transcription factor protein, as a suitable PCESR marker.
First, we focused on the overlap between the osmotic, salt and UV-B stress treatments. Only 209 genes are shared between the three stresses, including ZAT12 ( Figure 5). Os_S_UV_W (51) Os_S_UV_W_G (5) Oxidative (all) There are 56 of the 209 genes also differentially expressed after wounding of the leaves. Only five of those 56 are also regulated by genotoxic stress, while four of these are also responsive during the oxidative stress treatment. We, therefore, restrict our set of PCESR genes to these 56, as this comprises the best overlap between as many experiments as possible ( Figure 5; Table 3). Interestingly, our PCESR marker ZAT12 is not contained in all of the datasets. The reasons are its rapidly changing transcript abundance between the shoot, the root and between the conditions, as well as its temporal expression fluctuations, which has led to its exclusion from the oxidative and UV-B stress dataset. We performed a similar comparison also for the osmotic, salt and UV-B stress treatments and we surprisingly found only nine systemically responsive ( Figure S1). This finding also underlines that the systemic stress responses are rather specific to each of the stresses, which we have previously noted (Figure 3). Table 3. The 56 genes of the plant core environmental stress response (PCESR). All genes are regulated during UV-B, osmotic, salt and wounding stress. CDU: genes that were previously proposed as PCESR genes [4] during cold, drought and UV-B (CDU) treatment. TR: gene products that are involved in the regulation of transcription (TR). H3K27me3: genes that are known targets for histone methylation in the shoot [46,47]; Paralog: genes with paralogs contained in the list (x), genes that have paralogs in the genome (p), probes that identify paralogous genes (o).  Most of the previously proposed PCESR genes responsive during the seemingly unrelated cold, drought and UV-B light (CDU) stresses are expressed only in the shoots and failed to show up in this analysis due to our stringent filtering criteria (Table 3) [4]. Nevertheless, the vast majority of these CDU responsive genes are also responsive in our systemic datasets examined in this paper and are contained in the 209 genes that overlap between UV-B light, osmotic and salt stress ( Figure 5; Table S1) Consistent with the subset of PCESR genes identified previously [4], our 56 PCESR genes are enriched for genes encoding transcriptional regulators (34%; Table 3). Of the nine genes that are shared between our PCESR gene set and the one reported by Kilian et al. [4], six are transcription factors, including ZAT12. Interestingly, we found nine of our 56 PCESR genes to be target gene loci for the H3K27 trimethylation mark that coincides with active repression and silencing of the respective genes in the shoot tissue (Table 3) [46,47]. Instead, we observed a rapid activation of gene expression within a few minutes for several of the presumably silenced PCESR genes after application of the stress stimulus. The gene loci of ZAT12 and MES9 are rapidly induced in the shoot tissue, although both are also amongst those genes that are H3K27me3 decorated in the shoot. This finding suggests the involvement of rapid chromatin remodeling after the onset of the stress stimulus to mark the PCESR genes that are transcriptionally active. During the process of PCESR gene identification we found that 30 out of the 56 (54%) genes possess one or more paralogs of presumably redundant function (Table 3) [48]. In the cases of the membrane-associated transcription factors Anac036 and Anac062 [49,50] or the ribonucleases CAF1a and CAF1b [51], the paralogs of the genes are contained in our PCESR gene list. This proportionally high number of paralogous PCESR genes might be indicative of a putative backup functionality that transduces redundant information of the incoming stresses [48].

PCESR Genes Are Systemically Coordinated during Abiotic Stresses
From our observation with ZAT12, we have noted that some of the PCESR genes might not uniformly follow only one type of expression response or might even be non-responsive under certain conditions. Therefore, our PCESR genes are analyzed for their involvement in the non-systemic, type I or type II systemic gene expression responses for each of the six stresses (Table 4).  Intriguingly, the 56 PCESR genes exhibit a much more diverse response pattern than what we have previously assumed on the basis of the simple overlap of gene expression responses ( Figure 5). In fact, there were only two genes that displayed a coherent response pattern: both the DA1-related gene DAR3 and the WRKY18 transcription factor exclusively respond in the non-systemic tissue, which is characteristic at least of the majority of the entire WRKY transcription factor family ( Table 2). Most of our PCESR genes, however, respond in two categories, non-systemically, and in either type I or type II systemic responses, which underlines the probable importance of PCESR genes in mediating a specific stress response and to regain plant's homeostasis. During non-systemic or type I systemic responses, the dirigent gene DIR5, which is involved modifying the cell wall, is the only gene that was differentially expressed in all of the six datasets. Interestingly, the two ribonucleases CAF1a and CAF1b follow predominantly a non-systemic response category. During osmotic stress, however, CAF1a displays a type I response, while CAF1b is type II responsive.
We recognized that the systemic response pattern is especially prominent during the related osmotic and salt stresses. Almost all of our PCESR genes (53/56; 95%) follow a systemic expression response during these two stresses. Moreover, several of the PCESR genes switched their response category between these stresses. For example, the heavy-metal-transport superfamily gene At5g52750 exhibits mainly a type II systemic response, except for the osmotic stress, where it is type I responsive. Likewise, the MATE-transporter SID1 shows a non-systemic response pattern, while its gene expression is type I responsive during the osmotic stress treatment.
Some PCESR genes are well characterized for their involvement in the stress responses in Arabidopsis; however, it has not been discovered that their expression response pattern are specifically shifted between the different stress responses: ZAT10 is also a zinc-finger transcription factor gene like ZAT12. However, the expression responses of ZAT10 are either type I systemic (oxidative stress), type II systemic (osmotic and salt stress) or non-systemic (UV-B light and wounding stress). Likewise, the heavy-metal-transport superfamily gene At5g52750, ERD9, methyl-esterase MES9 or a member of the GNAT-family At2g32030 exhibit a distinct and specific expression pattern in all three response categories and in a stress-dependent manner. Hence, the majority of PCESR genes are likely to respond to various environmental conditions, but with diverse expression trajectories that are stress specific.

Stress-Induced Systemic Signaling Uses Multiple Pathways
We have shown that each of the abiotic stresses did not specifically invoke only a discrete set of genes, but also many genes that are responding specifically to the different stress conditions. We have focused on the analysis of the six abiotic stresses that can be mined for a non-systemic and systemic response pattern. Indeed, most of the genes that respond to the six systemic stresses are also regulated during cold, heat and drought, which were applied to the whole plant and which affected both tissue types ( Figure 5). To illustrate that there is considerable overlap of differentially expressed genes between the stresses, we chose to display the three response types for the salt stress response. The majority of non-systemic as well as type I and type II systemic response genes of the one stress stimulus is also differentially expressed in response to other stresses. Furthermore, systemic responsiveness of a given gene to a certain stress condition did not imply a general role as a systemic response gene ( Figure 5), which we have already shown for the PCESR genes. Instead, the diverse expression signatures suggest an important, underlying role in homeostatic balance and, thus, an activation by local or systemic signals as needed.

Discussion
Our detailed analysis of the AtGenExpress abiotic stress data set allowed us to identify genes, which are either non-systemic or of two different types of systemic responses (type I, type II). This provokes the idea of the existence of specific, yet unknown, systemic signaling mechanisms, which must rapidly integrate the stress perception for stress-specific signaling from the treated to the non-treated organ. Based on what is currently known in the literature, the systemic signals could be ssRNA, siRNA, peptides, proteins, JA or electrical signals. The earliest time point in the AtGenExpress dataset is 15 min, which is sufficient time for such a systemic signal to move. Electrical signals take seconds to minutes to be propagated [25]. JA signal transmission (not necessarily JA itself) has been shown to occur within minutes [52][53][54]. The difficulty in defining stress-specific transcriptional changes is likely due to components that overlap with other stress. For example, mechanical wounding is characterized by the physical rupture of tissues, which affects both ionic fluxes (presumably similar to salt stress) and cellular osmotic potential (osmotic stress treatments). This is in agreement with the observation that wounding elicits responses that overlap with those of biotic and abiotic stresses [55]. Likewise, UV-B treatments activate many genes that are not just specific to UV radiation treatments, but instead had been attributed to pathogen response [56][57][58], further supporting a connection between abiotic stresses and pathogen defense. The transcriptional wounding response has been shown to overlap with that of JA [15,16,38], which is in turn required for the plant defense against pathogen and herbivore attack [52].
We also observed that genes could be activated either directly or systemically depending on the type of stress, which generates an alarm signal to coordinate responses throughout the entire organism. This has already been proposed in earlier experiments that investigated the systemic activation of transcription by wounding [59], pathogen attack, or the systemic detection of electrical signals [60]. In these publications, it is suggested that an alarm state was raised first followed by the treatment-specific response. Comparable to the systemic leaf-to-leaf signaling described in these publications, in our analysis the establishment of an alarm state is also sent from shoot-to-root or root-to-shoot.
A major outcome of our work is that type II systemic responses appear to be more stress specific than non-systemic or type I systemic responses. This finding might be indicative of the fact that the type II responsive genes have an important function in specifying the stress response, downstream of a mobile systemic signal. In contrast, non-systemic or type I systemic responses displayed considerable overlap also between seemingly unrelated types of stresses or between different stress categories. In fact, Chen et al. [33] showed that it was difficult to identify transcriptional responses that could easily be attributed to an individual stress. Similarly, also the intercomparable AtGenExpress abiotic stress experiment suggests the involvement of different molecular pathways that contain the Plant Core Environmental Stress Response (PCESR) genes [4].
The PCESR genes identified in this work are by no means exhaustive, and as we have shown, many PCESR genes will probably be identified from different stress treatments. It is important to note that it is unclear how interlaced the effects on transcription are during different stresses [39]. Nevertheless, the PCESR genes identified in this study have several properties that are consistent with a fundamental role in mediating an alarm response. First, they are enriched in transcriptional regulators that are needed for transcriptional re-programming. Second, most of them systemically coordinated and expressed in both the shoot and root. Third, some of the PCESR genes are functionally conserved throughout plant evolution, which is a requirement for an effective definition of a plant core environmental stress response, as was proposed for yeast [32,33]. Support that our 56 PCESR genes are indeed part of a common alarm system in plants can be taken from the high number of already known genes that have an assigned function in environmental stress responses: ZAT10 has already been shown to respond to multiple stresses [61,62] and its gene product is known to be involved in diverse developmental pathways. More convincing is the identification of AtCOR413-TM1 and AtCOR413-PM1 amongst the PCESR genes. AtCOR413-TM1 and -PM1 genes are activated by water stress, ABA, light, freezing tolerance and these responses are conserved between wheat and Arabidopsis [36]. An evolutionary conserved function is one of the major arguments that core environmental stress responses exist. AtCOR413-TM1/PM1 are proposed to play a pivotal role in environmental stress signaling and a structural role by stabilizing the plasma membrane lipid bilayer [36]. It is tempting to speculate that gene families evolved and gene duplication occurred to compose a robust system of stress responses. Temperature and UV-B stress have been shown to affect genome stability epigenetically, indicating that abiotic stresses can have heritable effects at least for a few generations [63]. The recent findings that PCESR gene expression is increased in Polycomb mutants hints to active repression of some of these genes by H3K27me3 in the non-stressed (shoot) tissue [64]. These findings are supported from a previous publication that describes the dynamic removal of H3K27me3 marks at the PCESR gene locus of COR15A during cold stress treatment [65]. Similarly, pathogen attack has been shown to cause an increase in the frequency of somatic DNA recombination [66]. It is, therefore, apparent that responses to stress have short and long-term consequences influencing plant evolution. Not all of our 56 PCESR genes responded under all stress stimuli. Especially genotoxic and oxidative stresses revealed little overlap between otherwise commonly involved PCESR genes. One explanation is that most of the PCESR genes are downstream from a highly specific and well-defined ROS signal [18]. Consistently, the genotoxic and oxidative stresses are presumably affected in ROS production and signaling more generally. A concerted activation of PCESR genes under these circumstances might, thus, be impaired.

Microarray Data Processing
We used our publicly available AtGenExpress microarray experiment [4], which consists of control arrays (included individually within every stress set at TAIR The arrays were adjusted for the background of optical noise with the GC-RMA package in the R statistical environment of bioconductor [67] and normalized using quantile normalization [4,35,41,68]. Gene expression was normalized to the controls (36 chips) and the 118 conditions (236 chips) present in the entire dataset to identify regulated genes as in Kilian et al. [4]. Genes were considered differentially expressed, if they were regulated at least at two consecutive time points (each stress time point was compared to the respective control). Very noisy genes contaminating type II and non-systemic gene sets were removed by hand. The primary normalization was used to generate the principal component analysis. The secondary normalization was used to identify up-or downregulated genes. Centroids were formed for each classification (non-systemic, type I and type II) with k-means set to 2 clusters on the respective tissue. The PCA, regulated genes, and centroids were calculated using GeneSpring GX v7.3.1 (Agilent Technology, Böblingen, German).

Gene List Analysis
A functional categorization using GO-terms was performed with the TAIR web-interface [69]. Significant over-or under-representation (p ≤ 0.01) of a particular GO category was assessed by applying a hypergeometric distribution calculation restricted to the universe of all 22,747 genes on the ATH1 array. Calculations were performed in Excel; Results were plotted with plot 0.997 [70]. Cis-regulatory element analysis was performed with Athena [71]. Updated using Toronto "_at to AGI converter" 2009-7-29 release [72]. Transcription Factors were found using The Plant Transcription Factor Database (PlnTFDB) [73]. Genes that were hormone responsive were taken from literature [74]. Gene loci that are targeted by histone methylation were derived from publications [46,47].

Network Graph Analysis
The arrays were adjusted for the background of optical noise with the GC-RMA package in the R statistical environment [67,68] and normalized using quantile normalization. Gene expression was normalized to the controls (36 chips) and the 118 conditions (236 Gene lists were compared using the List Distance function from Motif Mapper v5.2.4.01 [75,76]. The values for direct overlap were used as input for Cytoscape v.2.6 [77] and arranged using the yFiles hierarchal clustering algorithm (Table S4). Node size was continually scaled to list size; edge thicknesses were continually scaled to overlap-percentage.

Conclusions
To conclude, the different stress treatments share an initial alarm state even before the plant can determine exactly what type of stress is actually there. We propose that unknown systemic signaling modules must be responsible for this coordination. However, our categorization of the gene expression responses into the different non-systemic type I and type II systemic stress responses allowed for the first time the in-depth analysis and characterization of six abiotic stresses. Moreover, PCESR genes have been identified and have been shown to display a highly stress-specific response pattern. Some of the PCESR genes switch between non-systemic, type I and type II systemic responses in a stress-specific manner. In addition, the type II stress responsive genes were found to be more stress specific than the non-systemic or the type I systemic responsive genes. The identification of the mobile signals and the analysis of how a specific response type is enforced at distinct gene loci will be a challenging task for future research.