INTRODUCTION

TOP ABSTRACT INTRODUCTION RESULTS DISCUSSION MATERIALS AND METHODS LITERATURE CITED

Plants have a remarkable ability to cope with highly variable environmental stresses, including cold, drought, and soils with changing salt and nutrient concentrations (i.e. abiotic stress). Nevertheless, these stresses together represent the primary cause of crop loss worldwide (Boyer, 1982), reducing average yields for most major crop plants by more than 50% (Bray et al., 2000). In contrast, the estimated yield loss caused by pathogens is typically around 10% to 20%.

Significant progress has been made to understand and manipulate abiotic stress responses (for reviews, see Shinozaki and Yamaguchi-Shinozaki, 1996; Bohnert and Sheveleva, 1998; Smirnoff, 1998; Blumwald, 2000; Bray et al., 2000; Cushman and Bohnert, 2000; Hasegawa et al., 2000; Knight, 2000; Schroeder et al., 2001; Serrano and Rodriguez-Navarro, 2001; Thomashow, 2001; Zhu, 2001b, 2001a). Three important themes have emerged.

First, the initiation of most stress treatments correlates with a cytosolic calcium release, in some cases with stimulus-specific patterns of oscillation (Allen et al., 2000; Knight, 2000; Posas et al., 2000). Second, stimulus-specific changes in gene expression are often observed alongside a set of shared stress responses. For example, in a survey of 1,300 Arabidopsis genes, the majority of cold and drought stress-regulated genes were observed as a shared stress response (Seki et al., 2001). Together, these observations support the hypothesis that a common set of signal transduction pathways are triggered during many stress responses.

A third important theme is that increased levels of stress tolerance can be engineered into plants by reprogramming the expression of endogenous genes. For example, overexpression of the transcription factor C-BOX BINDING FACTOR-1 (CBF1) resulted in plants with increased tolerance to cold stress (Jaglo-Ottosen et al., 1998). Inducible expression of the transcription factor DEHYDRATION-RESPONSIVE ELEMENT BINDING-1A (DREB1A) also resulted in improved tolerance to several stress conditions, including drought, salt, and cold (Kasuga et al., 1999). These two successes probably resulted from the overexpressed transcription factor altering the expression of many downstream genes. Overexpression of a cold- and drought-inducible calcium-dependent protein kinase gene in rice (Oryza sativa) also up-regulated the expression of several stress-regulated genes and increased drought tolerance in the transgenic rice plants (Saijo et al., 2000). However, in some cases, a stress response can be improved by changing the expression of a single downstream gene. For example, overexpression a Na⁺/H⁺-antiporter gene in tomato (Lycopersicon esculentum) and Arabidopsis provided a dramatic increase in NaCl resistance (Apse et al., 1999), presumably by preventing the build-up of toxic levels of cytosolic Na⁺.

Understanding a plant's response to a stress will require a comprehensive evaluation of stress-induced changes in gene expression. Using oligonucleotide and cDNA microarrays providing a partial coverage of the Arabidopsis genome, expression profiling studies have revealed a large number of changes associated with particular stages of plant development (Zhu et al., 2001), the circadian clock (Harmer et al., 2000), and various stresses such as wounding, cold, and pathogens (Maleck et al., 2000; Schenk et al., 2000; Bohnert et al., 2001; Seki et al., 2001). Expression profiling has also been used to study stress responses in other species, such as salt stress in rice (Kawasaki et al., 2001), barley (Hordeum vulgare; Ozturk et al., 2002), and yeast (Saccharomyces cerevisiae; Posas et al., 2000; Rep et al., 2000; Bohnert et al., 2001; Yale and Bohnert, 2001). A compiled list of genes connected to abiotic stress responses in Arabidopsis and other plants can be viewed at http://stress-genomics.org.

Here, we present mRNA expression profiles of leaves and roots from Arabidopsis subjected to salt (100 mM NaCl), hyperosmotic (200 mM mannitol), and cold (4°C) stress treatments. We used an Arabidopsis GeneChip microarray (Zhu and Wang, 2000), which provided probe sets for approximately 8,100 genes. We had two specific objectives. First, to identify mRNAs that change in a "stress-specific" fashion in response to cold, salt, or hyperosmotic stress. Second, to identify mRNAs that are coregulated during the acute phase (first 3 h) of a shared stress response. This is the first study to compare all three stresses using a global expression profiling strategy. Through a comparison of three different stress treatments, we reasoned that we could better identify both shared and stimulus-specific responses. Our results revealed changes in approximately 30% of the transcriptome present on the GeneChip, with most changes classified as stimulus specific. This global view illustrates the "fluid" nature of the transcriptome and the challenge we face in understanding the complexity of any given stress response.

	RESULTS

TOP ABSTRACT INTRODUCTION RESULTS DISCUSSION MATERIALS AND METHODS LITERATURE CITED

Using a GeneChip microarray, we identified 2,678 probe sets representing a combined total of 2,409 unique stress-regulated genes that displayed a greater than 2-fold change in expression compared with a fresh medium control. Expression profiles were made separately for roots and leaves isolated from plants exposed for 3 or 27 h to a 100 mM NaCl, 200 mM mannitol, or 4°C stress (Supplemental Table 1, which can be viewed at www.plantphysiol.org).

Non-Stress-Regulated Controls

A set of 10 representative control genes (non-stress regulated) were identified that did not show a significant change in expression under any of the stress treatments (Table I). These examples include commonly used loading controls, such as genes encoding polyubiquitin, eukaryotic initiation factor 4A1 and actin-2. A gene for a V-type H+-ATPase 16-kD subunit provides an example of a moderately expressed gene with similar expression levels in roots and leaves.

Identification of Shared and Stimulus-Specific Responses

Figure 1 illustrates the breakdown by stress for the 2,678 probe sets identifying a 2-fold or greater change in expression. It is important to note that our fresh medium control also resulted in 741 changes between 3 and 27 h, of which nearly one-half (407 probe sets) were also affected by stress treatments (Fig. 1).

View larger version (14K): [in this window] [in a new window]   Figure 1.   Venn diagrams showing an overview of stress-regulated changes (> 2-fold) for different treatment conditions. Stress-induced changes were identified in 2,678 probe sets (which approximately equals the number of genes). The number of changes and the percentage corresponding to up-regulation are shown. Numbers for fresh medium represent a comparison between 3 and 27 h. Numbers for the three stresses represent a comparison with the average of the 3- and 27-h control.

The Venn diagrams shown in Figure 2 illustrate one of the many ways in which this large data set can be sorted to reveal potential insights. These diagrams provide an important overview showing the distribution of changes into shared and stress-specific responses. To provide gene-specific information on the hundreds of shared and stress-specific responses, we have organized 15 supplemental tables (Supplemental Tables 2A-G, 3A-G, and 4; they can be viewed at www.plantphysiol.org), as identified in Figure 2. These tables can be accessed on-line and analyzed with computer software.

View larger version (22K): [in this window] [in a new window]   Figure 2.   Venn diagrams showing the distribution of stimulus-specific and shared stress responses (> 2-fold). Numbers in italics correspond to probe sets identified in a specific category. Most of these changes were only observed once as a transient change. Non-italicized numbers in parentheses correspond to individual genes that passed a more stringent criterion for reproducibility (see text). Further information on these selected changes can be found in Tables II through V (shown here) or as supplemental information in Supplemental Tables 2A-G and 3A-G (They can be viewed at www.plantphysiol.org).

To illustrate important themes and ways to view the data, four tables were selected for presentation alongside the text. A change was listed in these tables (Tables II-V; Supplemental Tables 2A-G, 3A-G; they can be viewed at www.plantphysiol.org) only if it met several conservative criteria. First, we required all changes to show reproducibility in at least two treatments (e.g. at 3- and 27-h time points, at the same time point in more than one tissue, or at the same time point with at least two stresses). Second, the direction of the change had to be the same for the gene to be labeled as coregulated by more than one stress (e.g. cold and mannitol regulated means induced or repressed in both cold and mannitol). Finally, we required all changes to show reproducibility of a 2-fold change in comparison with both the averaged 3- and 27-h controls as well as with their respective 3- or 27-h individual time-point controls. This was done to filter out changes that were likely attributable only to changes induced in the fresh medium control. The annotation listed for each gene was derived from the Institute for Genomic Research (or AGI) listing. In some cases we provided updated information. Functional annotation is expected to change for many of the genes as experimental information becomes available.

Acute Responses Shared by All Three Stresses

Tables II and III together list 118 genes that are up- or down-regulated by all three stresses during the acute phase (first 3 h) of each stress response. These tables are arranged in descending order of -fold change observed in NaCl stress. Of the 118 genes that show changes, the largest category (29%) was annotated as "unknown," 15% were predicted to be directly involved in regulating gene expression (e.g. transcription factors), 9% in membrane transport, and 8% in phosphoregulation. The 12 genes that were coregulated in both roots and leaves are identified in Table II. Among all of the shared 3 h responses, the transcript for LOW TEMPERATURE-INDUCED PROTEIN 78 (LTI/COR78, At5g52310) was observed as the most strongly induced (i.e. 98-fold with cold; 57-fold with NaCl). The actual level of induction was probably even greater, because the transcript was undetectable in controls (i.e. assigned a value of 24, which means that the signal was below the threshold for accurate detection).

NaCl-Specific Responses

Table IV (Supplemental Table 2E, which can be viewed at www.plantphysiol.org) contains 22 genes that are exclusively regulated by salt stress at both 3 and 27 h. This represents only 5% of the combined salt-specific changes observed at 3 and 27 h in the root. Although most of the remaining 440 changes are likely to represent salt-specific changes, these changes were only observed once as part of a transient response and were therefore not listed in a stress-specific response table. Of the 22 salt-specific genes persisting as a 3- and 27-h response, the largest category (50%) was related to oxidative stress enzymes (e.g. glutathione reductase and cytochrome P450), 23% were annotated as "unknown," and only one gene each was predicted to be directly involved in regulating gene expression, membrane transport, or phosphoregulation. The greatest -fold induction was observed for a putative "steroid sulfotransferase" At2g03760 (19-fold at 3 h). Of all salt-induced changes (shared or specific), this putative steroid sulfotransferase ranked as the fourth highest -fold change.

Cold-Specific Responses

Table V (Supplemental Table 4, which can be viewed at www.plantphysiol.org) contains 42 genes that are exclusively regulated by cold stress in both root and leaves at both 3 and 27 h. This represents only 2% of all the cold-induced changes. This set of changes represents the most reliable set of changes in this study, being detected as specifically cold induced in four different samples, whereas no change was detected in the 12 other non-cold-stressed samples. Of these 42 genes, the largest two categories (approximately 20%) were annotated as unknown or predicted to be directly involved in regulating gene expression. Only one gene was annotated as a membrane transporter, and none was directly related to phosphoregulation. The greatest -fold induction was observed for an EARLY LIGHT-INDUCED PROTEIN (ELIP; At4g14690; 232-fold at 27 h in leaves).

Top Three Changes

Table VI shows the three highest ranking -fold inductions for each of the three individual stresses. This table revealed that the three largest -fold changes were all induced by cold stress. Interestingly, the LTI/COR78 gene ranked first in all three stress treatments.

	DISCUSSION

TOP ABSTRACT INTRODUCTION RESULTS DISCUSSION MATERIALS AND METHODS LITERATURE CITED

The large number of stress-regulated transcriptome changes observed here underscores the difficulty of understanding the global context of a stress response. Using probe sets representing approximately 8,100 unique Arabidopsis genes, our expression profiling revealed a greater than 2-fold change for 2,409 genes in response to cold, salt, or osmotic stress (Supplemental Table 1, which can be viewed at www.plantphysiol.org). Extrapolating to the entire Arabidopsis transcriptome, the expression levels of more than 7,000 genes (approximately 30% of the genome) are potentially regulated by these common abiotic stresses.

Because all aspects of plant physiology are impacted by stress, we consider a large number of transcriptome changes to be reasonable. Although many stress-regulated genes have been identified previously (http://stress-genomics.org), our study provides the first global expression profile, to our knowledge, comparing three of the major abiotic stresses: salt, hyperosmotic, and cold. Our greatly expanded list of potential stress-regulated genes is consistent with our use of a GeneChip microarray strategy that allowed a sensitive and accurate quantification of a large number probe sets. With the observation of 2,409 stress-regulated changes, it is impractical to discuss the potential functions of individual changes. Instead we offer selected comments and observations to illustrate important themes.

The GeneChip Can Reliably Detect 2-Fold Changes

We expect that most of the 2,409 changes represent a biological response of the plant to its environment rather than a technical artifact of inconsistent hybridizations or probe labeling. A false change error of less than 0.25% changes is expected (i.e. around six genes) from control experiments conducted under identical conditions with the same detection threshold (i.e. expression levels >25; Zhu and Wang, 2000). To minimize the detection of random changes from a single sample, the GeneChip was probed with RNA pooled from two independent experiments. Although some of the observed changes may still be random or functionally unrelated to a stress treatment, our results did confirm many of the stress-regulated genes found in other surveys conducted on a smaller scale. For example, we detected all of the cold-regulated changes identified by Seki et al. (2001) from their analysis of 1,300 genes. From the set of 85 salt-regulated transcripts identified by Gong et al. (2001), 28 were represented by probe sets on the GeneChip microarray, of which more than 50% were observed to be stress regulated in this study. The incomplete overlap with salt-regulated genes observed by Gong et al. may be attributable to differences in growth conditions or detection methodologies.

Interpreting Transcriptome Changes Requires Caution

When considering the relative significance for each of the 2,409 changes, two important qualifications must be considered. First, expression profiling does not by itself define the critical genes required for any of stress responses. It is important to emphasize that changes in mRNA levels may not correlate with changes in protein or enzyme activity levels (e.g. Gygi et al., 1999). Nevertheless, expression profiles do provide useful starting points for more in depth analyses (e.g. Harmer et al., 2000). For instance, expression profiling can be used to create a candidate gene list to help prioritize the arduous task of using reverse genetics to assign functionality to genes. In the present study, stress-induced transcription factors represent good candidates for under/overexpression experiments.

Second, any interpretation of our results must include the realization that a plant is always changing and adapting to its environment. Thus, experimental changes are always being observed in a background of uncertain variation. Some of the changes observed here may be unique to our experimental conditions. We note two potentially important variables. First, tissue was harvested 3 h into the photoperiod. A 24-h interval was maintained between the two time points to avoid mistaking a circadian clock-controlled change for a stress-induced change. Nevertheless, some stress responses may be very different if observed in the dark versus the light. Second, all plants were stressed after exposure to fresh medium. Although the addition of fresh medium allowed a uniform and rapid initiation of parallel stress treatments, the fresh medium also induced a number of changes on its own. Thus, our experimental design certainly resulted in both hiding and revealing stress-induced changes. For example, fresh medium appeared to induce a greater than 10-fold increase in the expression levels of a nitrate transporter gene NRT2 (At1g08090) in the roots (Supplemental Table 2A, can be viewed at www.plantphysiol.org). However, this induction was almost completely blocked by all three stresses and therefore showed up as a common stress-induced change. This example emphasizes that the physiological status of the plant will impact how it responds to stress. Because plants are constantly adapting to a changing environment, there is no perfect "background" condition to reliably identify all stress-specific changes. In the future, additional expression profiling will be needed to classify stress-induced changes under different experimental conditions.

To provide a starting point for considering the importance of a given stress-induced change, we organized a set of tables to reveal the most consistent set of stimulus-specific or shared responses (Fig. 2; Tables II-VI; Supplemental Tables 2 and 3, which can be viewed at www.plantphysiol.org). We decided to only list genes that passed the more conservative criteria for reproducibility at different time points, tissues, or treatments. We did not include changes observed only once. However, these criteria for "reproducibility" excluded more than 1,000 potentially important changes that were only observed in a transient and stress-specific fashion.

Cold, NaCl, and Mannitol Trigger Primarily Stimulus-Specific Responses

Our results indicate that the majority of transcriptome changes are stimulus specific and not part of a general stress response common to cold, osmotic, and NaCl stress. During the acute phase of the stress responses (3 h), less than 5% of the changes were shared by all three stresses. By 27 h, the shared responses were reduced to less than 0.5%. This picture of predominately stress-specific responses is analogous to that observed in a comparison of drought- and NaCl-stressed barley, as revealed by an expression profile of 1,463 genes (Ozturk et al., 2002). Nevertheless, our results are in contrast to the observation made by Seki et al. (2001) from their expression profile analysis of 1,300 Arabidopsis genes. In their study, the majority of drought and cold stress-regulated changes appeared to be part of a shared response. A partial explanation for this difference may lie with the increased sensitivity of the GeneChip compared with the cDNA spotting technology used by Seki et al. (2001). Our GeneChip analysis allowed us to compare more genes, detect lower abundance mRNAs, and more accurately score changes close to a 2-fold threshold. Thus, although our analysis identified even more shared responses than previous studies (e.g. 142 from this study compared with 16 by Seki et al. [2001]), these shared responses only represented a minor fraction of the observable changes.

NaCl and Mannitol Induced Both Iso-Osmotic- and Stress-Specific Responses

We used a 200 mM mannitol treatment as an iso-osmotic control for the 100 mM NaCl stress. Not surprisingly, 174 shared "osmotic stress"-specific changes were observed at the 3-h post-stress time point. Nevertheless, the majority of NaCl or mannitol changes appeared to be stimulus specific. Thus, these two osmotic stress treatments clearly triggered very different responses within the first 3 h of stress.

Interestingly, about 40% (68) of the common 3-h iso-osmotic changes were observed in leaves. This is of special interest because the leaves were not in direct contact with the 100 mM NaCl or the 200 mM mannitol medium. These relatively rapid changes provide candidate markers for detecting the long-distance messengers that move from the root to the shoot. Potential long-distance signals include nutrients, hormones such as abscisic acid (ABA), or calcium-mediated action potentials (Dennison and Spalding, 2000; Forde, 2002; Knight, 2000; Schroeder et al., 2001) or a disturbed water potential gradient from roots to leaves (Nonami et al., 1997). However, our experiments were not conducted to exclude the possibility that, within 3 h, NaCl and mannitol had a direct effect on leaf tissues by translocation through the vascular system.

An important design feature of this study was the ability to compare multiple stresses and time points to more precisely identify potential stimulus-specific responses. For example, we identified 27 NaCl-specific responses (Table IV; Supplemental Tables 2E and 3E, which can be viewed at www.plantphysiol.org) and 46 mannitol-specific changes (Supplemental Tables 2F and 3F, which can be viewed at www.plantphysiol.org) that occurred at both 3 and 27 h. Of these, 16 were annotated as unknown genes. In these lists, the most highly induced genes for NaCl and mannitol stress, respectively, were a putative steroid sulfotransferase (At2g03760; 19-fold with NaCl), and a putative transcription factor (At5g47640; 12-fold with mannitol).

Numerous Cold-Specific Changes

Of the three stress treatments used here, cold induced nearly twice as many changes as either mannitol or NaCl (2,086 in total; Fig. 1). In the roots and leaves respectively, 173 and 188 cold-specific changes were observed as reproducible changes between 3 and 27 h (Supplemental Tables 2G and 3G, which can be viewed at www.plantphysiol.org), compared with 73 changes combined for NaCl and mannitol (Supplemental Tables 2E and F, 3E and F, which can be viewed at www.plantphysiol.org).

As mentioned above, an important design feature of this study was the ability to compare multiple stresses and time points to more precisely identify potential stimulus-specific responses. In the case of cold stress (Supplemental Tables 2G and 3G, which can be viewed at www.plantphysiol.org), we further explored the overlapping responses between root and leaves at 3 and 27 h to identify the most reliable set of mRNAs that are regulated specifically by cold, regardless of tissue or time. Forty-two genes were identified in this comparison (Table V, shown here), of which 10 were annotated as unknown. In this list, the most highly induced genes in leaves and roots respectively were ELIP (At4g14690; 231-fold induced) and COLD-REGULATED PROTEIN 15B (COR15B; At2g42530; 78-fold induced). Abiotic stress regulation of ELIP gene expression has been observed previously (Adamska and Kloppstech, 1994; Ouvrard et al., 1996; Shimosaka et al., 1999). Biochemical data indicate that ELIPs bind chlorophyll a and lutein (Adamska et al., 1999), but an exact role for their function in the plant's response to abiotic stress has not been reported.

Overlapping Responses to Salt, Osmotic, and Cold Stress

Although many stress-regulated genes have been identified previously, another important design feature of this study was the ability to compare multiple stresses at different time points to more precisely identify changes that are part of a "common" or "shared" response. At 3 h, we observed a total of 118 unique gene changes shared by all three stresses (Tables II and III). Of these, 30 (25%) were annotated as unknown. This group of stress-regulated genes is of potential interest in identifying targets of common stress-signaling pathways.

Most shared responses were specific for either roots or leaves, because only 12 of 118 showed coregulation in both tissues (Table II). In this list, LTI/COR78 (Atg52310) was the most highly induced in both leaves and roots (respectively, 40- and 98-fold for cold; 9- and 57-fold for NaCl; and 10- and 39-fold for mannitol). The biochemical function of LTI/COR78 is poorly understood, as are most of the highly induced genes observed for all three stress (Table VI).

In the roots and leaves, respectively, only two and eight genes were consistently observed as part of the shared response at both time points (Supplemental Tables 2A and 3A; they can be viewed at www.plantphysiol.org). One-half of these 10 genes were annotated as unknown. Interestingly, the expression of LTI/COR78, which was the most highly induced gene for both tissues at 3 h, was only observed as a consistent change in the leaves (i.e. 3 and 27 h). In 27-h roots, LTI/COR78 transcript levels returned to near normal for NaCl and mannitol stress treatments while more than doubling expression levels under cold stress. This example shows how even the most dramatic overlapping stress responses could be missed by an experiment that examined a limited number of time points or ignored tissue-specific differences.

Dynamic Changes Occur between 3 and 27 h of Stress

There were dramatic changes in both the numbers and identities of transcriptome changes between the 3 and 27 h stress time points for each stress treatment. We offer three examples to illustrate this point. First, in the transition from 3 to 27 h, the total number of shared changes (all three stresses) underwent a 4-fold reduction from 118 to 24. This dramatic reduction is consistent with the plant switching from shared stress responses to more stress-specific responses.

Two additional examples can be illustrated by examining the stress-specific responses. In the cold stress response, between 3 and 27 h both leaf and root tissues responded with a greater than 4-fold increase in the number of changes (Fig. 2). Interestingly, the opposite trend (4-fold reduction) was seen with the NaCl stress. This may reflect the success of the plant in mounting a NaCl-resistance response and the restoration of the transcriptome to a prestress program, or alternatively, the failure of the plant to establish an adaptive response. Given that Arabidopsis is classified as a salt-sensitive plant, the latter interpretation is worth considering. In contrast, the continuing increase in number of cold stress changes suggests a fundamentally different stress response. In this case, "cold stress" transcriptome of the plant appears to be moving toward a new and dramatically different "steady state," presumably better adapted to a cold environment.

The Root and Leaves Express Different Sets of Stress-Regulated Genes

Although roots and leaves contain different sets of specialized cells, it was not known to what extent the stress response programs would differ between these tissues. Our results support the hypothesis that roots and leaves have very different transcriptome responses to all three stresses. For example, 86% of the cold-induced changes are not shared between root and leaves (Fig. 2; Table V). Although similar root-leaf differences were also observed with NaCl and mannitol stress, these differences may partly reflect the fact that only the roots (and not leaves) were in direct contact with NaCl and mannitol treatments.

68% of the Circadian Controlled Genes Are Linked to Stress Regulation

Our results support a hypothesis that many of the circadian clock-controlled genes are also subject to stress regulation (Harmer et al., 2000). One important function of the circadian clock may be to regulate gene expression in "anticipation" of a predictable stress. For example, because the peak time for cold stress is normally predawn, a plant's ability to pre-activate a cold stress pathway may provide a higher level of resistance. Harmer et al. (2000) recently identified 18 known stress-regulated genes in an analysis of circadian controlled genes. Here, we identified 306 additional stress-regulated genes among the 453 known circadian controlled genes. Together our studies suggest that approximately 68% of the circadian controlled genes are linked to a stress response pathway, strengthening the argument that the circadian clock helps a plant adapt to daily environmental changes.

2,409 Steps toward Enhanced Annotation of the Genome

Our results revealed stress-regulated expression patterns of every conceivable pattern. For example, KIN2, which has been well characterized as part of the cold stress response (Kurkela and Borg-Franck, 1992), was also induced by NaCl and mannitol as part of an overlapping 3-h short-term response in both roots and leaves. However, after 27 h, roots and leaves displayed different patterns. In roots, the expression of KIN2 returned to normal for NaCl and mannitol stress, but remained high for cold stress (Supplemental Table 2G, which can be viewed at www.plantphysiol.org). In contrast, the expression of KIN2 in leaves remained high for all three stresses (Supplemental Table 3C, which can be viewed at www.plantphysiol.org). The variation in expression patterns emphasizes that each stress response examined in this study involves complex regulatory interactions.

Our study provides the first evidence for stress regulation of more than 370 genes with unknown functions. In addition, we extended the knowledge base for many already known stress-regulated genes, providing insights into their tissue specificity and regulation by multiple stresses. For example in leaves at 3 h post-stress, we observed an increase in transcript levels for protein phosphatase ABA Insensitive-2 and a transcriptional activator CBF1(At4g25490). These genes have been well studied in connection to stress responses (e.g. Leung et al., 1997; Jaglo-Ottosen et al., 1998; Sheen, 1998). Interestingly, ABA Insensitive-2 and CBF1 were not observed in roots as part of the 3-h acute response shared by all stresses. Instead, the shared response in roots included a different set of phosphatases and a related CBF1-like transcription factor. These examples illustrate two important points. First, the details of stress response may vary for different tissues and cell types. Second, some changes have the potential to make global changes in cellular functions, for example by altering phosphosignaling pathways or transcription.

The Role of Transcription and RNA Stability in Regulating the Transcriptome

Considerable effort has been made to identify stress-regulated promoter elements (Seki et al., 2001). One reason is to facilitate the engineering of plants with new or altered stress-regulated genes. The data set presented here was incorporated into a larger study of biotic and abiotic stress-regulated promoters (Chen et al., 2002). As expected, many of the cold-induced genes were found to have promoters with dehydration response element-like or ABA response element-like binding elements, providing the potential for regulation by DREB/CBF transcription factors and ABA. However, many of our 2,409 stress-regulated genes did not. This may reflect (a) the presence of undiscovered stress-regulated promoter elements, or (b) a large number of changes caused by differences in mRNA stability instead of gene transcription. More studies will be needed to understand the mechanisms by which the levels of each mRNA are regulated during stress responses.

The Role of Calcium Signals in Triggering General and Stress-Specific Responses

A central hypothesis supported by this study is that an abiotic stress initially triggers a set of common stress response pathways (e.g. output seen in Tables II and III) that are subsequently modified to be highly stimulus specific. Calcium signals have been observed as an early response to all three stresses used here (Knight, 2000). However, it is not clear whether these calcium signals are initiating a general stress response or communicating more specific information necessary to develop a stimulus-specific response. In addressing this question, one must consider that the same stress applied to different cell types can result in different calcium signatures (Kiegle et al., 2000). Thus, an important challenge for the future is to understand at the cellular level how cross-talk between calcium and other signal transduction pathways creates a stimulus-specific response.

	MATERIALS AND METHODS

TOP ABSTRACT INTRODUCTION RESULTS DISCUSSION MATERIALS AND METHODS LITERATURE CITED

Stress Treatments and Sample Preparations

Seven-day-old axenic seedlings of Arabidopsis (Columbia) were transferred to rafts floating on hydroponic medium in Magenta boxes (Sigma-Aldrich, St. Louis) and grown for 3 weeks with gentle agitation. Light (75 microeinsteins; a mixture of cool-white fluorescent and incandescent) was provided on a 12-h/12-h light/dark cycle. Medium was provided as 0.5× Murashige and Skoog salts, 0.5 g L¹ MES, 0.5× vitamins (Sigma-Aldrich), pH 5.7, and 0.5% (w/v) Suc. At the time of stress treatments, plants were still in a vegetative growth phase (i.e. pre-bolting). To initiate stress treatments, old medium were replaced with fresh medium that was prechilled to 4°C (cold stress) or supplemented with 100 mM NaCl (salt stress) or 200 mM mannitol (hyperosmotic stress). Cold stress was maintained for the duration of the experiment by placing Magenta boxes on ice. A fresh medium-only control was conducted in parallel. Stress treatments were initiated just after the lights came on, with samples taken at 3- and 27-h time points. Roots and leaves (entire shoot) were separated and frozen in liquid nitrogen. Each stress treatment and RNA extraction was replicated in two independent experiments. Total RNA was extracted from frozen and pulverized tissues using an RNeasy column (Qiagen USA, Valencia, CA). RNA samples for each replicate were pooled to obtain a single RNA sample for cDNA and cRNA probe preparation and expression profiling.

Expression Profiling

Detailed sample preparation and hybridization procedures were described previously (Zhu et al., 2001). Double-strand cDNAs were synthesized from 0.5 µg of total RNA using oligo(dT₍₂₄₎)primer containing 5'-T7 RNA polymerase promoter sequence SuperScript II (Invitrogen, Carlsbad, CA). Biotinylated complementary RNAs (cRNAs) were transcribed in vitro from synthesized cDNA by T7 RNA polymerase (ENZO Biochem, New York). Hybridizations with labeled cRNAs were conducted with Arabidopsis GeneChip Microarray (Affymetrix, Santa Clara, CA). This GeneChip contained probe sets for approximately 8,100 Arabidopsis genes (Zhu and Wang, 2000). Probe preparation (cRNA), hybridizations, and normalizations were conducted according to protocols optimized for sensitivity and reproducible comparisons (Harmer et al., 2000; Zhu and Wang, 2000). Expression data were normalized globally before data analysis. Genes with accurately detectable transcript levels were defined by probe sets showing averaged expression levels equal to or greater than 25, as described (Zhu and Wang, 2000). For probe sets showing weaker signals, expression values were adjusted up to 24 for further data comparisons. The -fold changes between stress treatments and fresh medium control were calculated by dividing stress-treated expression values by the averaged control expression values. The averaged control expression values were calculated by averaging the control value from the 3- and 27-h fresh medium samples. Using the above criteria for identifying genes displaying greater than 2-fold change, we expected less than 0.25% false changes resulting from inaccuracies of hybridization and detection.