The 197 genes in cluster N12 (Figure 2) are induced by a broad range of diverse stress conditions: cold, osmotic, salinity, wounding, and biotic stresses (including treatments with elicitors). The 'limma' analysis indicated that approximately 80% of these genes were significantly regulated under all treatment conditions, whereas the rest of the included genes were marginally regulated in one (mostly the wounding treatment) but significantly regulated in all other conditions (P < 0.01; Table 1; Additional data file 5). They appear to represent a common or universal stress response transcriptome because most of these genes are conserved among plants, animals and fungi, and are stress regulated in all organisms, with the inclusion of a few genes related to the plant-specific hormones ABA and JA (Figure 3 and Table 1). Several Gene Ontology (GO) categories were enriched among these genes: GO:0009611 (response to wounding), GO:0009613 (response to pest, pathogen, or parasite), GO:0006970 (response to osmotic stress), GO:0009737 (response to ABA stimulus), GO:0009651 (response to salt stress), GO:0009723 (response to ethylene stimulus), GO:0009751 (response to salicylic acid stimulus), GO:0009753 (response to JA stimulus), GO:0050832 (defense response to fungi), GO:0006839 (mitochondrial transport), and GO:008270 (zinc ion binding). Signaling pathways related to mitogen-activated protein kinase (MAPK), calcium, reactive oxygen species (ROS), phospholipids, apoptosis, and protein degradation were induced. Equally, part of this cluster of genes that generally are upregulated by stress is functionally related to vesicle transport and mitochondrial functions. N12 included induced genes that had previously been identified as related to or specific for biotic stresses, but these were also induced by abiotic stresses, and vice versa. Past restrictions in the scope of analyses, which typically focused on single treatment conditions, and the resulting problem of annotation stringency did not compromise the fuzzy k-means clustering analysis. We discuss these universal stress response genes by organizing them into different pathways (Figure 3).

Several MAPK pathways, organized into signaling cascades, are conserved in eukaryotic organism 2324. In Saccharomyces cerevisiae, for example, the high osmolarity glycerol (HOG) signaling pathway is responsible for osmotic stress sensing 2526. The Arabidopsis AtHK1, MEK1, MPK4, and MPK6 can complement yeast deletion mutants of the HOG pathways. Other examples of plant MAPKs are alfalfa stress-induced MAPK (SIMK), tobacco salicylic acid-induced protein kinase (SIPK), wound-induced protein kinase (WIPK), and Nicotiana Fus-3-like kinase6 (Ntf6).

Among common genes that are upregulated by stress, several MAPK components were identified: MPK5, MKK9, and MAPKKK14. The MAPK pathway has been suggested to be involved in ethylene signaling 272829. Included among ubiquitous stress-regulated genes is also ACS6, encoding the rate-limiting enzyme of ethylene biosynthesis and a substrate for MPK6 30, together with six ERF/AP2 transcription factors (AtERF). This implicates the ethylene signaling-mediated engagement of a subset of the MAPK family as a component of the common stress response.

However, the ethylene response transcriptome is not strictly clustered in the stress transcriptome, notwithstanding its importance in developmental processes such as fruit ripening. Incorporating the results from a study that measured transcript changes in Arabidopsis Col-0 wild-type 31 into the cluster structure obtained by fuzzy k-means, the significantly ethylene-regulated genes identified in the study were located in a large number of different clusters.

The yeast Snf1 protein kinase and the mammalian AMP-activated protein kinase act as metabolic sensors that monitor cellular AMP and ATP levels. Activation increases the ATP:AMP ratio. Snf4 is part of the Snf1 protein kinase complex. In higher plants, they are involved in response to environmental or nutritional stress. Related common stress-induced genes were CIPK11 (encoding a Snf1-related protein kinase that is similar to SOS2, a protein kinase that is involved in plant salinity stress responses) 32, SKIP2 (a conserved SCF ubiquitin ligase subunit that interacts with SnRKs), and AZF2 and ZAT10 (C2H2 zinc finger proteins) 33. Both AZF2 and ZAT10 suppressed the Snf4 deficiency in yeast and function as transcription repressors in Arabidopsis 3334. ZAT10 can activate salt stress tolerance, controlled in yeast by MSN2 and MSN4 factors, and ZAT10 can repress the expression of the plant stress gene RD29A 35. Several Snf1-related genes appeared in stress-induced clusters other than N12 as well, suggesting functions that are specific for particular stress conditions (data not shown).

The Bax-inhibitor 1 (BI-1) is an endoplasmic reticulum (ER) protein that suppresses cell death induced by ER stress in both animal and plant cells. It can inhibit the activation of Bax and its translocation to mitochondria, and suppresses the activation of caspase, and functions in reducing calcium release from the ER 36. In Arabidopsis, Bax over-expression causes ROS accumulation and cell death, and BI-1 attenuates the cell death effect without affecting production of ROS 3738. It alleviated cell death caused by biotic and abiotic stresses 39. BI-1 (At5g47120), one of three genes in Arabidopsis with this sequence signature, was induced by several other stresses in a specific manner as well, and appears to represent a signature gene and protein of the common stress response cluster.

Although mechanisms of vesicle transport have been studied extensively, little is known about regulation in response to stress. A plant vesicle-related protein, AtVAMP, when ectopically expressed, can suppress Bax-induced apoptosis in yeast, possibly by improving membrane repair 40. The over-expression of AtRab7, a gene that is involved in regulation of vesicle trafficking, increased endocytosis in roots, as well as salt and osmotic stress tolerance 41. This indicates the importance of regulated vesicle trafficking for acquisition of stress tolerance.

Several genes related to trafficking from endosomes to central vacuoles were placed into N12. They are SYP21, Vps28-related, Tsg101-related, SRC2, Ras-related GTPase, and genes for two Snf7 family proteins. In roots, the Tsg101-related and Vps28-related genes, as well as SYP21 and one gene encoding a Snf7-like protein are specifically expressed in the endodermis of the root hair zone.

A multitude of signaling molecules is generated from membrane phospholipids. Their involvement in osmotic stress responses has been demonstrated. Several related genes are induced, such as encoding inositol polyphosphate 5-phosphatase II, FYVE domain-containing phosphatidylinositol-4-phosphate 5-kinase (PI4P5K), and lipase class 3 family proteins. PI4P5K leads to the synthesis of PI4,5P2. Mutations in the offsetting phosphatase gene, SAC9, lead to over-accumulation of PI4,5P2 and constitutive expression of stress-response pathways 4243. The product of the BAP1 gene, which is also upregulated, interacts with BON1, a protein with two C2 domains that binds to phospholipids. Together, BAP1 and BON1 control plant growth homeostasis 44.

ROS have been associated with stress sensing and signaling, but have emerged more recently as important, general signals 454647. Irrespective of their ubiquitous presence, ROS that derive from different stimuli appear to be recognized as specific, indicating that a number of different signal mediators must exist. We suggest that cluster 12 identifies the evolutionarily conserved set of these genes. SRO5 is a gene that controls ROS in plants, which is upregulated by various stresses. SRO5 transcript expression overlaps partially with that of P5CDH mRNA. The induction of SRO5 leads to production of a 24-nucleotide nat-siRNA that guides cleavage of P5CDH mRNA, resulting in regulated proline levels 48. Additionally, ZAT12, and possibly ZAT10 of the Snf1 pathway, also participate in ROS signaling transduction 46.

Multiple calcium-related functions are induced by stresses. Among them is a SOS2-like protein kinase, namely CIPK11. However, little is known about the other genes in this group, including two calmodulins, three calcium-binding proteins, and three calcium-dependent kinases. These calcium-related genes cannot be organized into a pathway-like structure, in part because of the lack of detailed experimental evidence and also based on the multiplicity of functions that are channeled through calcium-binding proteins.

CCR4 and CCR4-associated factor 1 (CAF1) are critical for mRNA turnover in yeast 49. Pcf11 is an mRNA 3'-end processing factor and binds the carboxyl-terminal domain of the largest subunit of RNA polymerase II 50. Both CAF1 and Pcf11 have their Arabidopsis homologs upregulated by different stresses, indicating a role for control over mRNA processing and degradation. Another upregulated gene is the eukaryotic translation initiation factor SUI1. Other examples are AZF2 and ZAT10, which encode transcription repressors.

Stress-related transcription factors were also among the common stress response genes, including five WRKY family members, four Myb, three HSF, three NAM and two AP2, and the transcription factor SCL13. Included are WRKY18 and WRKY40, which physically interact with both overlapping and antagonistic roles in pathogen responses 51. WRKY25 and WRKY33 are substrates of MKS1, which itself is a substrate of MPK4 and regulates plant defense reactions 52. WRKY33 is also required for resistance to necrotrophic fungal pathogens 53. WRKY11 interacts with calmodulin and acts as a negative regulator of basal resistance in Arabidopsis 54. SCL13 has been shown to function in light signaling 55. These WRKYs function in resistance to necrotrophic but not biotrophic pathogens, whereas necrotrophic damage is more closely related to the physical damage caused by abiotic stresses, as also reflected in the cluster structures. Little information is available for other transcription factors in cluster N12, although several isoforms of Myb, NAM, HSF, and AP2 not included in N12 have been associated before with stress response pathways.

Among the genes upregulated by many stress treatments, several are localized to mitochondria. They are three BCS1-like ATPases (which could function as chaperones, whose yeast homologs are required for cytochrome bc[1] complex assembly), two DIC1-like, one ANT1-like, one MTM1-like, and one other mitochondrial substrate carrier family protein. Furthermore, a ferrochelatase I gene, an NADH dehydrogenase-related gene, and a PP2C are part of this group. Also upregulated here was the Bax-inhibitor 1 gene. To appreciate their precise functions in plants, more studies are required.

Among the common stress response genes were two ABA-related genes, RPK1 and CYP707A3. RPK1 encodes a leucine-rich repeat receptor-like kinase 1, a membrane-bound regulator of ABA early signaling 56. The rpk1 mutant exhibited decreased sensitivity to ABA, and over-expression resulted in hypersensitivity. CYP707A3 encodes a cytochrome P450 protein catalyzing ABA 8'-hydroxylation and catabolism. Its knockout mutant exhibited exaggerated ABA-inducible gene expression and enhanced drought tolerance, whereas over-expression was associated with growth retardation by ABA and increased transpiration 57.

ADC genes are essential for polyamine (PA) production. Over-expression of ADC2 led to GA-deficient plants and accumulation of putrescine, a phenotype reversed by GA3 58. The null mutant adc2-1 was sensitive to salt stress, but could be rescued by external putrescine 59. ADC2 is among the common stress response genes.

The stringent response is crucial for stress adaptation in bacteria, mediated by the production of the nucleotide guanosine-3',5'-(bis-)pyrophosphate (ppGpp). RelA and SpoT encode bacterial enzymes for ppGpp synthesis. RSH is the higher plant homolog of this RelA/SpoT protein 6061.

These genes function in resistance to the bacterial pathogen Pseudomonas syringae pv. tomato DC3000 carrying avirulence proteins 516263, and they - as identified here - were also induced by various abiotic stresses. Interestingly, NHL3 over-expression in Arabidopsis enhances resistance to the virulent Pseudomonas syringae pv. tomato DC3000, without an increase in PR gene expression or H₂O₂ accumulation 64. PBS1 and RPS5 are required for avrPphB mediated Pseudomonas syringae resistance in Arabidopsis. AvrPphB can proteolytically cleave PBS1, which is required for RPS5-mediated resistance 65. PUB17 is a U-box ARMADILLO repeat E3-ligase, which regulates cell death and defense 66. Another disease resistance family protein, similar to Cf-2.1 (At2g34930), is also upregulated by various stresses. Its null mutant was particularly susceptible to fungus attack 67. The inclusion of these genes in cluster N12 suggests their function in common mechanisms that counter both abiotic and biotic stresses.

An additional 120 genes are included in the common stress response cluster (ST3). In part, their functions are known by specific activities (for example, trehalose-6-phosphate phosphatase), whereas most are identified only by domain identifiers (for example, protease-associated or thioredoxin family-related), or their functions are not clear or completely unknown. The group included transcripts for 19 zinc-finger family proteins, five protein kinases, four protein phosphatases, a number of glycosyl hydrolases, thioredoxins, cytochromes P450, and hormone-responsive functions, mostly annotated according to similarity criteria, and 40 expressed proteins without any annotation. Among the genes that lack annotation, the majority is most strongly induced by conditions that affect redox homeostasis and ROS responses, in particular treatments with ozone, H₃BO₃, H₂O₂, AgNO₃, hypoxia, and triiodobenzoic acid (an inhibitor of polar auxin transport; Genevestigator dataset 8).

The high correlation of genes in cluster N12 with experimentally verified or alleged functions in a wide variety of stress conditions in species across all kingdoms suggests that the functions identified by this cluster categorize the basic stress response transcriptome (Figure 3). By their nature, these functions appear to identify ubiquitous cellular stress defense programs in all organisms, whereas pathways that integrate stress responses at the organ or organism levels may be based on programs that diverged during evolution. Conceivably, reverse genetics will determine the functions of little understood and completely unknown genes in N12, and provide a clear separation of these genes from pathways that are specific to individual stress conditions. The common stress response genes epitomize components of crosstalk between biotic and abiotic stress response mechanisms by identifying genes such as WRKY transcription factors, NHL3, and PUB17. Indeed, the Arabidopsis mutant bos1 exhibited compromised resistance to the pathogen Botrytis cinerea and reduced tolerance to drought, high salinity, and oxidative stress 68.

The WB-BOX motif TTTGACT was identified in clusters N0, N11, and N19. Genes in clusters N0 and N19 were generally induced by abiotic stresses, whereas genes in cluster N11 were upregulated markedly in roots by salt treatment. The WB-BOX represents a binding site for WRKY transcription factors 70, which have 12, 4, and 5 members in clusters N0, N11, and N19, respectively. Among other clusters, only N12 included a number of WRKY factors (five in total). It seems that WRKYs correlate well with pathogen response activity. Genes in cluster N0 were also induced by osmotic and ionic stresses, and the additional HSF motif (heat shock factor binding) (A|G)GAANNTTC was over-represented in this cluster (with N representing any nucleotide). Also, two HSF transcription factors are included in this cluster.

Genes in clusters N1 and N8 were responsive to light treatment. The ABRE motif ACGTG(G|T)C was identified in both clusters, together with an unknown motif, namely (A|G)ACCACA(A|G). ACGTG(G|T)C is similar to the G-Box motif that mediates light signaling 12. Also identified in cluster N1 was the I-Box motif GATAAG(A|G).

Clusters N3, N9, N10, N12, N13, and N20 were induced by ABA treatment to variable degrees. The ABRE motif was over-represented in these clusters, but only cluster N3 contained the ABRE-binding proteins AREB1 and AREB2 13. The DRE motif (A|G)CCGAC was identified in the clusters N3, N9, and N12, which is in agreement with the strong induction by cold stress of genes in these clusters 13. Within these three clusters, four, two, and one DRE-binding (DREB) transcription factors were included, respectively. Although cluster N11 included seven DREB genes, the DRE motif was not over-represented in this cluster. Cluster N11 also contains an additional nine ERF/AP2 transcription factors. Clusters N3 and N9 additionally exhibited the EVENINGAT motif (AAAATATCT), which functions in the circadian control of gene expression 71. Further identified were the P1BS motif (GNATATNC) and an unknown, hypothetical motif (A|G)(C|T)TAA(A|T)NNNTGA(C|T) in cluster N10, and the 2S-SEED-PROTBANAPA motif (CAAACAC) in cluster N13.

Over-representation of the well known ABRE motif in multiple clusters of genes that respond to either light or ABA treatment points toward the existence of additional motifs 13. These could be the I-Box and DRE motifs that are over-represented in these clusters, and other putatively cis-acting motifs are suggested by the analysis. More likely, however, is the presence of transcriptional control mechanisms that act on cis-element binding proteins rather than on the promoter elements.

Motifs of prevalence similar to those in upregulated genes appear to be largely absent from the stress-repressed genes in clusters N4, N7, and N17. For cluster N15 genes, strongly downregulated by osmotic and high salinity stresses in roots, the MYCATERD1 motif (CATGTG) and the SORLRP3AT motif (TGTATATAT) were identified. Clusters N2, N5, N14, and N18 included many genes related to general gene expression functions, protein synthesis, cell organization, and metabolism. Several known motifs were enriched in these clusters. The UP1ATMSD motif GGCCCA(A|T)(A|T)(A|T), which is related to axillary bud growth 72, was over-represented in all four clusters. Additionally, over-represented in genes in cluster N2 were the I-Box motif GATAAG(A|G) and the SORLIP5AT motif GAGTGAG 73, which appear to be connected to the expression of genes in metabolic functions. Cluster N5 showed the E2F1OSPCNA motif GCGGGAAA, the E2FANTRNR motif TTTCCCGC, and the E2FCONSENSUS motif (A|T)TT(G|C)C(G|C)(G|C). These motifs are typically associated with genes that are involved in cell cycle progression and cell division 747576. At lower frequency, clusters 14 and 18 exhibited similar motif combinations (Table 2).

In general, fuzzy k-means clustering applied to 5'-regulatory sequences confirmed known motifs in the major stress-responding clusters, whereas different clusters shared subsets of these motifs. Additional, secondary motifs between and within large clusters are suggested (Figure 4b, Table 2), but attempts to distinguish between clusters that shared similar expression patterns through motif analyses alone proved inconclusive. Different approaches will be necessary to reveal how combinations of motifs may control gene expression. Methods for identifying such combinations are emerging 77.