In stark contrast to Arabidopsis, a related species, Thellungiella halophila (Thellungiella salsuginea; salt cress), displays extreme tolerance to high salinity, low humidity and freezing. High nucleotide sequence identity permits the use of tools developed for Arabidopsis for Thellungiella transcript profiling, for which a microarray platform with >25 000 DNA elements (70-mer oligonucleotides) was used. Microarray transcript profiling and intensity analysis, quantitative RT-PCR, and metabolite profiles define genes and pathways that showed shared and divergent responses to salinity stress in the two species. Shared responses are exemplified by 40% of the regulated genes functioning in confining ribosomal functions, photosynthesis and cell growth, as well as activating osmolyte production, transport activities and abscisic acid-dependent pathways. An additional 60% of regulated genes distinguished Thellungiella from Arabidopsis. Analysis of the differences showed that Arabidopsis exhibited a global defense strategy that required bulk protein synthesis, while Thellungiella induced genes functioning in protein folding, post-translational modification and protein redistribution. At 150 mm NaCl, Thellungiella maintained unimpeded growth. Transcript intensity analyses and metabolite profiles supported the microarray results, pointing towards a stress-anticipatory preparedness in Thellungiella.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Evolutionarily, T. halophila occupies an intermediary position between the Brassica and Arabidopsis clades, separated from Arabidopsis by 15–30 Myr at a time during which crucifers experienced polyploidization events (Blanc et al., 2003; Koch et al., 2000). Conceivably, protective measures to ionic or osmotic imbalances that distinguish species could have evolved at several levels, including degrees of protein activity and structure; the behaviour of the transcription machinery at different control points; or altered responsiveness to external factors or internal, hormonal signals. Apart from adaptive evolutionary changes in coding regions, an expansion of the gene complement leading to paralogy, for example through gene or (partial) genome duplications, could equally have been at the basis of Thellungiella's adaptation to extreme habitats. The appearance and establishment of paralogs may be at the root of how species in many plant families acquired the ability to thrive in habitats from which most other family members became excluded (Hwang et al., 2002; Lespinet et al., 2002; Madlung and Comai, 2004; Tognolli et al., 2002). Considering the vast variability in the capacity to adapt to stressful conditions shown by individual species in many plant families, or even ecotypes within a species, contrasting the two crucifers should provide clues about mechanisms that reveal their distinctive stress-response capacities.
We describe transcript profiles and metabolite levels in Arabidopsis and Thellungiella grown under controlled conditions in the absence and presence of salt stress at 150 mm NaCl with Arabidopsis, which represents a break point for this species, and at 150 and 250 mm NaCl, conditions that represent a minor irritation or a very moderate stress, respectively, for Thellungiella. The use of the Arabidopsis long-oligonucleotide array platform is justified by a high nucleotide sequence identity between the species in the range of 92–95% for most genes (Inan et al., 2004; Wang et al., 2004; Wong et al., 2005). We define at least three different Thellungiella response categories that are not activated or are not represented in Arabidopsis. In addition to the shared salt-stress responses, Thellungiella activated additional pathways different from the intense activation of disparate pathways in Arabidopsis. The response of Thellungiella also included an increase in metabolites that fail to increase in Arabidopsis. Second, pre-stress intensities of genes in certain pathways were significantly higher in Thellungiella, especially in categories that highlight Thellungiella-specific, salt stress-inducible pathways. Third, deduced from intensity profiles and metabolites, Thellungiella appears to possess novel stress-relevant genes whose homologs in Arabidopsis are not responsive to stress.
Microarrays, normalization, statistics and data clustering
For the microarray study, 4-week-old Arabidopsis (At) and 6-week-old Thellungiella (Th), similar in size and before bolting (Figure 1), were treated with 150 mm NaCl (At-150, Th-150), and 250 mm NaCl (Th-250) for 3 and 24 h, respectively. The gene-expression levels were compared with those of untreated controls (see Experimental procedures). Each experimental condition was represented by six slides from two biological repeats, including dye swaps. In addition to controls, the microarray platform included 25 425 probes spotted as 70-mer oligonucleotides. After background subtraction, normalization and filtering (TIGR-TM4-MIDAS), approximately 80% of the signals remained on each slide (Table 1). Only probes with at least five out of six intensity values were considered further, resulting in the acceptance of 9708 ± 865 probes for each of the six experimental conditions. The accepted transcripts were processed by one-way anova (P = 0.01; TIGR-TM4-MEV), then filtered for high degrees of freedom (d.f. ≥ 27), resulting in the acceptance of 2090 transcripts considered as highly significantly regulated across all experimental conditions.
Table 1. Regulated genes identified by one-class t-test
Total genes (d.f. ≥4)
Significant genes (d.f. ≥4)
Upregulated > twofold
Downregulated > twofold
t-test parameters: P = 0.01; permutation = 64 (all possible combinations from six slides); adjusted Bonferroni correction for alpha. Experimental conditions identify species, NaCl concentration (mm) and time of stress (h).
Hierarchical clustering (Figure 2) showed that the Thellungiella 3-h experiments grouped closely, followed by the Arabidopsis 3-h data set, then Thellungiella 24-h (150 mm), then Thellungiella 24-h (250 mm), and finally the Arabidopsis 24-h (150 mm) data. Furthermore, with respect to the apparently most strongly regulated genes, At-150 mm and Th-250 mm were more similar than Th-150 mm and Th-250 mm. The latter observation was consistent with the one-class t-test results generated from each data set (Table 1). Both observations suggested that Thellungiella and Arabidopsis dealt with salt stress by regulating both similar and different genes and pathways, and that Thellungiella might respond in a more determined way.
The fuzzy k-means clustering method with principle components analysis (PCA) was introduced as a novel way of clustering the microarray results (Gasch and Eisen, 2002; Figure 3). By this process, centroids were generated according to the assigned parameter, k, combined with the PCA result. Hence the method captures and assembles the most typical gene-expression patterns. Each gene is linked to each centroid with different membership values, with a total membership value of 1.0 apportioned to all centroids. A higher membership value for a centroid corresponds to an expression pattern more similar to genes associated with this centroid. Finally, clusters based on centroids were generated (see Experimental procedures). Using this method, we obtained 12 clusters that properly described the behaviour of 80% of the 2090 transcripts. These included five large clusters (C0–C4) that included 75% of all data, and seven small clusters (C5–C11) with 5% of the data (Table S4). The remaining 20% of the transcripts could not be placed with a certainty higher than 0.50 and were therefore not analysed (Figure 4).
An overview of these clusters showed that both species had a balanced number of up- and downregulated genes. Two of the 12 clusters (C0 and C4), containing 40% of all regulated genes, showed consistent upregulation (C4) or downregulation (C0) among the three sets of data (At-150, Th-250, Th-150), and 3 and 24-h time points, respectively. In each species, the 24-h time points represented an amplification of the 3-h response. One small cluster (C9) showed upregulation at the 3-h time point and downregulation at the 24-h time point in all three sets of data. The remaining nine clusters, which included >50% of the data, revealed different time-specific and, in particular, species-specific regulation. Cluster C1 showed moderate downregulation in At-150 at both 3 and 24 h; slight downregulation at 24 h in Th-250; and slight to moderate upregulation in Th-150 at the 24-h time point. C2 showed upregulation in Th-150 and Th-250, and downregulation in At-150. C3 showed upregulation in At-150 but practically no change in Th-150 and Th-250.
Similarity between the two species: downregulated genes (C0)
The 411 genes in C0 were consistently downregulated in all six experimental conditions, with the decline in At-150 more dramatic than in Th-250 or Th-150. They divided into four groups. The first and largest group contained cytosolic and chloroplast ribosomal subunit genes. Consistent with this, four FKBP genes, including FKBP13 and TWD1/UCU2, appeared as downregulated. Histone genes and genes functioning in DNA synthesis, replication, DNA damage repair, amino acid synthesis (except for proline synthesis), protein synthesis (including two tRNA synthetase genes, elongation factor genes), and cell division were also included. The second group consisted of genes that function in photosynthesis, including genes encoding photosystem I and II subunits, ferredoxins and ferredoxin-NADP reductases, magnesium-chelatase subunits GUN5, and CHLI2; protochlorophyllide reductases PORA, PORB and PORC; coproporphyrinogen III oxidase LIN2, oxygen-evolving enhancer PsbQs, RuBisCo activase, Calvin cycle genes, and other chloroplast-located gene products for chloroplast protein import, chloroplast growth and division. As the third group, a considerable number of genes specify their functions in cell-wall degradation and modification, including cell elongation (EXGT-A1 and XTH9), expansions (EXP6 and EXP8), and cell structure (tubulins, TUA6 and TUB1). Similarly downregulated were genes encoding proteins in fatty acid synthesis and elongation genes, including ACP1, ACP4, AAE14, MOD1, LTA2/PLE2, lipid-transfer protein genes, and GDSL-motif lipase/hydrolase genes. The latter group combined auxin-responsive SAURs and gibberellin-responsive genes.
Similarity between the two species: upregulated genes (C4)
The 272 genes in C4 were consistently upregulated in all six experimental conditions, with regulation in At-150 slightly more dramatic than in Th-250 and Th-150. Although functionally diverse, many of these genes had been reported in studies on individual genes as strongly abiotic stress-regulated. These included the abscisic acid (ABA)-responsive PP2Cs ABI1, ABI2 and HAB2; the ABA-responsive transcription factors ATHB7, ATHB12 and growth regulators ATAF1, NAP, BLH1, SPL; proline biosynthesis genes P5CS2, ALDH2, P5CDH and several alcohol dehydrogenases; the CBL-interacting protein kinases CIPK11/PKS5 and CIPK21; and other abiotic stress-responsive genes such as RAB18, RD21A, RD29A, RD29B, KIN2, ERD7, ERD10, ATTPS11 and MT2A. Finally, several G-proteins also appeared in C4.
The cluster was further characterized by the presence of genes for transporters, several of the ABC type, and for amino acids, proton-dependent oligopeptides, and sulphate (Table S1), the calcium-transporting ATPase ACA8, two auxin polar transporters including APP1, and several mitochondria and chloroplast carrier protein genes involving in energy transfer. Identified were genes for several chloroplast-located proteins, previously suggested as stress-responsive (ATFER1, ELIP1), several leaf senescence markers (ASP3/YLS4, ACD1, DIN2), and genes involved in lipid degradation and β-oxidation pathways were greatly upregulated.
As a smaller group, aspartyl proteases, cysteine proteases, genes involved in the protein ubiquitination pathway, autophagy gene APG8h, and vacuolar processing enzyme γ-VPE were also included in C4. Not surprising, many redox homeostasis genes were greatly upregulated, including glutaredoxins, glutathione peroxidases (ATGPX2 and ATGPX6), and glutathione S-transferases (ATGSTZ1 and ATGSTU7). Many lignin biosynthesis genes, the wax synthesis gene WAX2, and a few cell-wall modification genes formed a group that indicated the process of secondary growth. Finally, a number of NAM transcription factors, zinc-finger family protein genes, and two WRKY transcription factors were included in this cluster.
Differences between the two species: genes upregulated in Arabidopsis only (C3)
The 318 genes in C3 were greatly upregulated in Arabidopsis, yet were mostly unaffected in Thellungiella. In contrast to C4, very few genes in this cluster have been characterized previously as osmotic/salt/ABA-responsive. At first glance, many of these have defense-related functions. These consist of genes for a number of MATE efflux family proteins (presumably involved in toxin resistance), a metallothionein (MT1A), the universal stress protein RD2, senescence-associated SAG13, other short-chain dehydrogenase/reductases, hydrolases, jacalin lectin family proteins, disease-resistance proteins, a terpene synthase, and trypsin inhibitors. The associated WRKYs 17, 25 and 33 might be grouped as defense-related transcription factors.
Interestingly, a group of genes specifying calcium-binding proteins, and calcium-dependent (calcium-binding) and independent (lipid-mediated) vesicle trafficking functions appeared almost exclusively in C3. They included ECA1, CPK9, BON1, several SEC14 genes, an SEC1 gene, and two SNAREs (VAMP7B and VAMP7C). The appearance of some genes in fatty acid synthesis in C3 could be consistent with the increased intracellular trafficking.
Many genes encoding Myb transcription factors, AP2s, bZIPs, two NAMs, and zinc fingers appeared in C3 as specifically upregulated in Arabidopsis. Many protein kinases and PP2Cs, similar to C4, also appeared in this cluster. Additionally, we found a few MAP kinases, including ATMKK9, ATMPK11, ATMPK5 and WNK4, in this group. The presence of a variety of translation initiation factors, ribosomal subunits RPL10B and RPL10C (both also have an RPL16 domain that binds directly to 23S rRNA) could be consistent with the appearance of the MAPKs.
Differences between the two species: genes upregulated in Thellungiella only (C2)
The 128 genes in C2 were upregulated steadily at a comparable level in Th-150 and Th-250, and downregulated in Arabidopsis. As expected, as presumably species–specific functions they were relatively difficult to categorize. As many as half the genes are identified as functionally unknown by both The Arabidopsis Information Resource (TAIR) and The Institute for Genomic Research (TIGR). The remainder fell into many divergent and often unique categories. MSBP1, ATTRX4, ATGSTU20 and other redox homeostasis genes were included. Differently from cluster C3, we observed very few transcription factors in C2. Instead, Thellungiella induced small nuclear ribonucleoproteins and a Methyl-CpG binding protein (known to function in gene silencing). A few peptidyl-prolyl cis-trans isomerases, including PIN1AT (known to facilitate protein folding), were specific to this cluster. Further C2-specific genes were MMT (a methionine S-methyltransferase), a UbiE/COQ5 methyltransferase involved in coenzyme synthesis, a CoA-synthase homolog, two cytosolic aminopeptidases, and several genes specifying protein-targeting functions. Evaluating the collection of functions in C2, it may be suggested that, rather than activating multiple transcription and translation processes, Thellungiella diverged from the Arabidopsis model by enhancing functions in protein interaction and modification, and in protein redistribution processes.
Thellungiella prefers sodium (C1)
C1, the largest cluster with 464 genes, was composed of distinctly differently regulated genes. Arabidopsis repressed the expression of these genes at both time points, most strongly at 24 h. Thellungiella showed no regulation at 3 h at both concentrations of NaCl. At 24 h, however, Th-250 showed slight downregulation, and Th-150 showed upregulation. One-third of the genes in C1 were annotated by TAIR and TIGR as functionally unknown/no ontology. The rest could be grouped clearly into different categories. Similarly to C2, C1 showed a large number of genes for protein interactions, modification and translocation, including protein kinases and phosphatases, chaperonins, DNAJ proteins, and protein targeting. Genes for cytosolic ribosomal subunits abounded, suggesting that protein synthesis pathways were induced in Th-150, conceivably to sustain growth, or to restructure the proteome, or to make up for initial losses. The upregulation of FKBP15-2, Sec61 gamma, and cell-cycle genes (CDC2A, ROC2) appeared to support this view. In addition, increased transcription involved other anabolic functions: DNA repair (e.g. RecA); nucleotide synthesis and salvage pathways (e.g. UMPS, PUR5, ADK2, UPT1); fatty acid synthesis (e.g. CAC2, SMT3, LACS2, KAS I, KAS III); and amino acid synthesis (e.g. NMT1, MAM1/IMS3, CARA).
No plastid ribosomal subunits were found in C1, which may suggest that plastid growth was restrained. However, many genes encoding proteins involved in glycolysis, the TCA and Calvin cycles, mitochondrial electron transport/ATP synthesis, and even tetrapyrrole biosynthesis appeared in C1. The non-photochemical quenching gene NPQ4/PSBS was severely downregulated in Arabidopsis, yet stayed constant in both Th-150 and Th-250.
It was noteworthy that, although C1 contained a comparably large number of transcriptional regulators like cluster C3, these fell into diverse categories including small nuclear ribonucleoproteins, histone deacetylases, chromatin-assembly factors, various RNA- or DNA-binding proteins, or transcriptional coactivators. Only a few have been classified as transcription factors (CUC2, SVP, MS1, RAP2.7, WRKY54). Their presence, however, was consistent with the conclusions drawn for cluster C2.
The seven small clusters
Clusters C5–C12 included 118 genes altogether. There were from seven to 26 genes associated with these clusters, indicating that their regulation patterns were relatively insignificant. A description of these clusters is given in Appendix S1.
Pre-stress expression differences between Thellungiella and Arabidopsis
Thellungiella transcripts generally hybridized to the printed Arabidopsis probes with intensities that were equal to those seen in homologous hybridizations. After intensity normalization, standard deviation filtering and t-test (P = 0.05), 2620 genes remained as having statistically significant intensity differences between Thellungiella and Arabidopsis. Among these, 1207 had higher intensities in Arabidopsis and 1313 had higher intensities in Thellungiella, but in both species very few transcripts showed large differences. When a threefold difference was used as a cut-off, there were 42 transcripts with higher intensities in Arabidopsis, and 51 higher in Thellungiella (Table 2).
Table 2. Significant intensity differences between Thellungiella and Arabidopsis transcripts
P-values calculated using two-tailed Student's t-test with two-sample unequal variances. ITh and IAt values calculated by averaging raw intensities, after background subtraction, extracted from the 12 Thellungiella and six Arabidopsis 24-h control slides. AGI, Arabidopsis Genome Initiative.
Glutaredoxin family protein
F-box family protein, similar to SKP1 interacting partner 2 (SKIP2)
Protease inhibitor/seed storage/lipid transfer protein family protein
DNA-directed RNA polymerases I, II and III, 7-kDa subunit, putative
GDSL-motif lipase/hydrolase family protein
GDSL-motif lipase/hydrolase family protein
GDSL-motif lipase/hydrolase family protein
U-box domain-containing protein, protein ubiquitination
ChaC-like family protein, similar to E. coli pH-independent Ca2+/H+ antiporter
RALFL27, similar to tobacco rapid alkalinization factor (RALF)
Glycoside hydrolase family 28 protein
FAD3, ER omega-3 fatty acid desaturase, biosynthesis of 18:3 fatty acids from phospholipids
Tryptophan synthase-related protein
Mother of FT and TF1 protein (MFT), a flowering-time modulator
RAB18, a dehydrin
Gar1 RNA-binding region family protein
Flavin-containing monooxygenase family protein
Ras-related GTP-binding family protein
b 5′-adenylylsulphate reductase (APR3)/PAPS reductase homolog (PRH26)
Putative GTP-binding protein, similar to SAR1
ATGPX4, a glutathione peroxidase
Expressed protein with a DNA ligase domain
Glycosyl hydrolase family 17 protein
Expressed protein, weakly similar to MTD2
ATGSTU26, a glutathione S-transferase
Glycosyl hydrolase family 17 protein
CYP78A5, a P450 that marks the shoot apical meristem boundary in Arabidopsis
PAP8, a purple acid phosphatase
Laccase, putative similar to high-pI laccase (LAC2-1)
Sarcosine oxidase family protein
Hydrolase, alpha/beta fold family protein
GDSL-motif lipase/hydrolase family protein
60S ribosomal protein L35a (RPL35aA)
ATK5, a kinesin protein involved in microtubule spindle morphogenesis
Phospholipid/glycerol acyltransferase family protein
Fimbrin-like protein, putative; similar to fimbrin-like protein ATFIM1
RNA-binding protein, putative
SOUL, heme-binding family protein
Dehydrin, similar to dehydrin ERD10
Among the 51 transcripts with higher intensity in Thellungiella, several groups emerged. The first group included four GDSL lipases and a lipid transfer protein, together with FAD3, and a few cell wall-synthesis genes. The second group contained five genes involved in redox control, including ATGPX4, ATGSTU26, APR2 and APR3; and three abiotic stress-inducible genes including RAB18 and TCH2. The third group included eight genes with a function in transcriptional regulation and growth modulation, including MFT, ATK5 and CYP78A5. Finally, we identified four genes functioning in protein-ubiquitination pathways. Transcripts for genes encoding transcription factors were not present. When all 2620 genes were considered in terms of differences in gene families, a few families stood out (Table S3). More highly expressed in Thellungiella were genes with functions in ABA synthesis and ABA responsiveness, lipid transfer proteins and GDSL lipases, heat-shock proteins, peptidyl-prolyl cis-trans isomerases, histone genes and small nuclear ribonucleoproteins, redox control proteins, and C2H2 zinc fingers. In contrast, genes involved in ethylene response, cyclins, transposons and bZIP transcription factors showed higher intensities in Arabidopsis.
Verification by real-time RT-PCR
Quantitative RT-PCR analyses were conducted for all six experimental conditions using 11 transcripts chosen from different fuzzy k-means clusters, apparently up- or downregulated or unaffected (Figure 5; Table S2). The results, while the absolute fold-regulation varied between microarrays and quantitative amplifications for these transcripts, showed a correlation of 0.83.
With a focus on the polar fraction of metabolites (Table 3), drastic differences distinguished the two species. Thellungiella maintained higher metabolite levels in both the absence and presence of salt stress compared with Arabidopsis. Most of the differences, and higher concentrations in Thellungiella, could be attributed to increases under stress in the amount of hexoses, disaccharides and additional complex sugars, the identity of which has not been determined conclusively; and in the amount of sugar alcohols, especially inositol and galactinol. The increase in the category amino acids in both species was nearly exclusively based on an increase in proline, a known stress response in both species (Bressan et al., 2001; Taji et al., 2004), which was more pronounced in Thellungiella. Among the entry metabolites for nitrate into the amino acid pool, Thellungiella contained much higher amounts for glutamic and aspartic acids, the corresponding amines. In addition, Arabidopsis and Thellungiella were distinguished by different amounts of organic acids, especially malate and citrate which decreased slightly during the stress. Again, Thellungiella showed much higher amounts than Arabidopsis under pre-stress conditions, which was also the case for inorganic ions, particularly phosphate.
Table 3. Metabolite analysis
Mean values are expressed as percentage of total signal (internal standard, ribitol, was taken as 100%). Thellungiella contains many of these metabolites at higher amounts under control conditions, especially fructose, sucrose, complex sugars, malate and proline. The nature of all complex sugars has not been identified unequivocally. Samples were taken from control plants and plants treated at 150 and 250 mm NaCl, all harvested at the 24-h time point. ND, not detected; difference control versus stress: *P = 0.01; **P = 0.05.
Unknown (putative complex sugar)
Transcript profiles have been reported, using a cDNA-based platform of 7000 Arabidopsis full-length cDNAs, in salt treatments that compared Thellungiella plants with Arabidopsis plants exposed to a lethal (250 mm) concentration of NaCl (Taji et al., 2004). The results indicated less pronounced stress responses by Thellungiella compared with Arabidopsis. The authors interpreted their results by further identifying constitutively higher expression of genes in Thellungiella as the cause for its superior performance. We used a 70-mer oligonucleotide-based platform of >25 000 Arabidopsis genes, with more reasonable and physiologically comparable stress conditions for both species. Rigorous statistical analyses, careful selection of clustering methods, and a comparison with Affymetrix data ensured high confidence in data interpretation. Combined with pre-stress intensity analysis and metabolite profiling, the results revealed genes highlighting a clear distinction between the two species. In addition, we used plants of comparable developmental age, taking into account Thellungiella's slower growth, and performed experiments when complete rosettes had formed and before flowering was initiated.
Selection of suitable tools for analysis
To depict the expression pattern for 2090 genes that passed one-way anova with high statistical confidence, we introduced the fuzzy k-means clustering method (Gasch and Eisen, 2002) which, to our knowledge, has not been applied before to plant transcriptome studies. In conventional k-means clustering, each gene is forced into one and only one cluster, and therefore co-expression and co-regulation patterns will be masked by only marginally relevant genes, resulting in misinterpretations. In fuzzy k-means clustering the boundary of each cluster is set by a selected membership cut-off value. By optimizing this cut-off value, irrelevant genes can be removed. Furthermore, PCA guaranteed capture of the most informative centroids; automated convergence of very similar centroids during computation ensures uniqueness of the final clusters. Compared with conventional k-means clustering, the requirement of accurately estimating k is greatly reduced, leading to increased precision. This approach placed 75% of the regulated genes into five major clusters with distinct expression patterns. In contrast, a test using conventional k-means clustering gave results that were difficult to analyse, due to the ambiguous placement of a subset of the genes and similarities between different clusters (results not shown).
Arabidopsis salt-stress responses
The importance of maintaining ion homeostasis during salt stress has long been recognized. The Salt-Overly-Sensitive (SOS) pathway that connects a calcium signal through a calcium-binding protein (SOS3) via a kinase transmitter (SOS2) to a sodium exclusion (SOS1) function (Zhu, 2002, 2003) has been established as a major ionic (Na+) stress-regulatory system. Also shown has been a salt-protection function provided by transgenic overexpression of Na+/H+ (or alkali cation) antiporters of the NHX/NHE family which confine sodium to vacuoles, or by maintaining potassium homeostasis (Apse et al., 1999; Blumwald, 2003; Sottosanto et al., 2004).
However, transcripts for these functions are not readily observed in microarray studies (Kreps et al., 2002; Seki et al., 2002). These genes could have low transcript intensities as one reason, or their induction might be post-transcriptional. Possibly most relevant is that many of these genes have been characterized in Arabidopsis mutant screens at low Na+ concentrations, while they exert an effect that is insufficient to save Arabidopsis at moderate or high Na+. Accordingly, studies done with Arabidopsis at higher Na+ concentrations point to the importance of osmotic adjustments, the pivotal role of ABA, and genes and pathways shared with other abiotic, and even biotic, stress responses (Fowler and Thomashow, 2002; Kreps et al., 2002; Seki et al., 2002, 2003). A salt stress that challenges growth is not merely ionic in nature; it requires elicitation of a complex response network (Zhu, 2001).
Comparisons with Affymetrix gene-chip results
In an additional evaluation, our results were compared with publicly available Affymetrix gene-chip data from Arabidopsis 3-h and 24-h 150 mm NaCl stress experiments (AtGenExpress, TAIR gene chip database: http://www.arabidopsis.org/info/expression/ATGenExpress.jsp). The two data sets, although based on different platforms including different numbers of genes, and processed through different statistical methods, shared common trends of regulation. For the 3-h time point, 81% of upregulated genes and 82% of downregulated genes from our study (t-test, P = 0.05) were regulated in the same direction as suggested by the Affymetrix data set. For the 24-h time point the numbers were 79% for upregulation and 88% for downregulation (S.M. and Q.G., unpublished data). This comparison provided additional confidence with respect to the microarray platform used, as well as suggesting that the Arabidopsis salt stress-response mechanisms are now relatively well defined at the level of transcripts.
Similarities between the Arabidopsis and Thellungiella salt-stress transcriptomes
The similarities are clearly depicted in the fuzzy k-means clusters C0 (downregulation, 411 genes) and C4 (upregulation, 272 genes). Cluster C0 proved simplest with respect to its composition: ribosomal subunit genes and genes in photosynthesis functions represented 12 and 20%, respectively, of the downregulated, functionally known genes in C0, suggesting repression of protein synthesis and plastid activities. Additionally, cell-wall synthesis and modification were equally reduced, together with other growth and expansion processes. Interestingly, in other large-scale analyses of the Arabidopsis abiotic stress transcriptome, transcripts for ribosomal proteins rarely changed (Fowler and Thomashow, 2002; Kreps et al., 2002; Seki et al., 2002, 2003). However, an analysis of Affymetrix gene-chip data (AtGenExpress, TAIR gene chip database: http://www.arabidopsis.org/info/expression/ATGenExpress.jsp) revealed 70% of the ribosomal subunit genes as significantly downregulated only in osmotic and salt-stress experiments (S.M. and Q.G., unpublished data). This suggests that the osmotic or salt-stress treatments typically used are of such severity that Arabidopsis ceases to grow, and that the reactions measured are largely defense responses.
Cluster C4 included a large number of abiotic stress-responsive (especially ABA-responsive) genes, many identifying functions in osmolyte production. Also observed were glutaredoxins, glutathione peroxidases and glutathione S-transferases, suggesting a necessity for enhanced redox homeostasis. Further upregulated were transcription factors and growth regulators, suggesting remodelling of growth; a variety of transporters, probably involved in nutrient redistribution, signalling molecules and energy; and many proteases and protein ubiquitination genes, the latter suggesting protein-recycling processes. Consistent with cluster C0, leaf senescence markers were upregulated, showing decreased chloroplast activity. The upregulation of secondary cell-wall and wax biosynthesis suggest attempts at preventing water loss.
Similarities in regulation between Arabidopsis and Thellungiella showed that both salt-sensitive and salt-tolerant plants, encountering increased salinity, transcriptionally rapidly curb growth-related functions, including (principally) chloroplast activities. Abscisic acid was clearly the crucial hormone used by both species. At 3 h these adjustments were already well under way in both species. Presence and upregulation of protein kinases (including CDPKs) and phosphatases (especially PP2Cs) suggested responses involving protein modification.
Differences between salinity-sensitive Arabidopsis and tolerant Thellungiella
Clusters C1, C2 and C3 (910 genes) outline differences between Arabidopsis and Thellungiella. C3, clustering genes upregulated in Arabidopsis but downregulated in Thellungiella, identified mainly general defense-related genes and transcription factors, together with protein translation–initiation functions. Even at 150 mm NaCl, Arabidopsis initiates what may be termed a global defense response that apparently requires the presence of newly synthesized proteins. In contrast, cluster C2, upregulated in Thellungiella at both concentrations of NaCl (and especially at 250 mm) and downregulated in Arabidopsis, was enriched in genes with a function in protein folding, post-translational modification and protein distribution. Here the choices of pathways diverged dramatically between both species. It is tempting to suggest that the ‘Thellungiella way’ conserved resources and energy, hence making it a more superior strategy under stress conditions.
The composition of cluster C1 is also illuminating. Apart from genes suggesting modifications at the protein level, as in C2, we observed stress recovery already at the 3-h time point. These processes included enhanced nucleic acid salvage, DNA repair, carbon assimilation, and amino acid and lipid synthesis. In addition, transcripts in glycolysis, TCA cycle, Calvin cycle and ATP synthesis had reverted to the pre-stress status at the 24-h time point. Although some chloroplast ribosomal protein genes remained downregulated, transcripts for cytosolic ribosomal subunit proteins were upregulated. It appeared that Thellungiella was able to evaluate stress intensity quickly, and took distinct, different strategies facing 150- and 250-mm NaCl stress.
A recent study, reporting 6578 unigene expressed sequence tags (ESTs) from cold-, drought- and salt-stressed cDNA libraries of Thellungiella (ecotype Yukon) (Wong et al., 2005) concluded that salt-induced responses diverged significantly between Thellungiella and Arabidopsis. Our results agree with this conclusion. In this study, 569 out of the 2090 genes analysed here had at least one matching EST, evenly distributed over the 12 clusters that show overall sequence identity of 92–95% with their Arabidopsis homologs, further supporting the suggestion that the Arabidopsis-based microarray platform is suitable for Thellungiella transcript profiling.
Metabolites and stress tolerance
Fundamental differences, reflecting what the transcript profiles appear to predict, are prominently reflected when comparing metabolite profiles (Table 3). Thellungiella exhibits higher pre-stress concentrations of several compounds that have been shown to have protective functions in the osmotic imbalance that is a component of salt stress (Hasegawa et al., 2000). The concentrations for all major metabolite groups were often several-fold higher in Thellungiella and, in some groups, stress (even short-term stress) caused even more disparity. The high amounts of hexoses (approximately fivefold), sugar alcohols (twofold) and organic acids (two- to threefold, malate, citrate, but not fumarate) in Thellungiella were unexpected. Some of these increased during the 24-h stress period. Proline, the amino acid that increased most dramatically in both species, as previously noted (Bressan et al., 2001; Taji et al., 2004); inositols; and increased amounts of hexoses and complex sugars are substantiated by increased expression levels of transcripts involved in their metabolism.
Transcription factors with species specific patterns
The majority of transcription factor transcripts identified at significantly regulated levels showed co-ordinated behaviour. As a group, the plant-specific transcriptional regulators NAM (no apical meristem), including ATAF1 and CUC1, were all upregulated significantly in at least one condition or species. WRKY transcription factors, known to have regulatory functions in pathogen defense and other stress responses (Eulgem et al., 2000; Ulker and Somssich, 2004), were also upregulated, especially in Arabidopsis. However, Myb transcription factors showed diverse responses, and none appeared in C4 as upregulated in both species. AP2 transcription factors mainly appeared in C3 as upregulated only in Arabidopsis. bZIP transcription factors, having higher intensities in Arabidopsis under control condition, were further induced mainly in Arabidopsis.
The largest group of responders, however, included transcripts for 40 zinc-finger proteins. Although originally identified as nucleic acid-binding proteins, zinc-finger domains are versatile in structure and involved in diverse functions, particularly in protein–protein interactions (Takatsuji, 1999). In our experiments, the regulated zinc-finger genes appeared as one of the largest functional categories, found in every cluster. For instance, C2H2-type zinc fingers have been suggested as stress regulators (Sakamoto et al., 2004). They showed generally higher transcript intensities in Thellungiella under control condition, and many were induced especially in Arabidopsis during salt stress. The precise functions of all observed zinc-finger proteins remain to be elucidated.
Differences in transcript intensity
Recent studies have probed the possibilities of acquiring both expression ratios and absolute or relative gene-expression intensities from long oligonucleotide arrays (t'Hoen et al., 2004). We analysed all 24-h control intensities. Even before normalization, we observed comparable hybridization intensities, and the Pearson correlation coefficient between Thellungiella and Arabidopsis slides, based on the averaged intensity for each gene, was 0.90. After stringent statistical analysis, 2620 genes showed significant intensity differences between the two species. Consistent with the observation by Taji et al. (2004), with high confidence, we observed ABA biosynthesis and ABA-responsive genes with higher intensities in Thellungiella (Table S3). We further identified genes with a function in redox control, protein folding, histone genes, and cell-wall synthesis genes as more highly expressed in Thellungiella.
In order to see whether differences in expression intensity were important for salt-stress tolerance, the expression-intensity ratio between Thellungiella and Arabidopsis was calculated for each gene in the computed fuzzy k-means clusters (data not shown). The only significant difference identified cluster C2. Surprisingly, the averaged ratio in C2 was 1.5, demonstrating that Thellungiella not only had higher intensity of genes in this cluster with diametrically opposite regulatory behaviour, but elevated their expression when stressed. Therefore we consider these genes, 129 in total, and the way Thellungiella enhances their expression, as prime representatives in salt-stress acclimation and major contributors to the tolerance lifestyle of this species. One-third of these genes are labelled as functionally unknown in the TAIR database; and only 18 out of 129 genes have been studied in Arabidopsis. None of these 18 genes has been indicated before as salt stress-relevant, and it seems possible that they represent stress-inducible paralogs of genes that are not regulated in the same way in Arabidopsis. The high-intensity signal on the Arabidopsis platform and upregulation of these genes in Thellungiella could indicate their involvement in Thellungiella's superior stress performance. Also, the high signal intensity may point to the existence of novel functions carried out by paralogs of Arabidopsis genes in Thellungiella. Only the sequencing of the Thellungiella genome will resolve this question.
With a platform that nearly completely encompassed the genome, our study identified salt-stress response as a defined network in both Arabidopsis and Thellungiella, with common responses. Many of the Arabidopsis-specific reactions to high salinity appear to be injury-based, representing extreme defense reactions. Advanced bioinformatics tools then led to the discovery of response pathways that distinguished Thellungiella from Arabidopsis. Thellungiella appears superior in managing metabolite composition and maintenance of energy, while Arabidopsis expended energy at low stress levels on a multiplicity of pathways. The response precision shown by Thellungiella, we suggest, identifies genes that will very probably be novel and important for our understanding of plant salinity stress tolerance.
Plant growth and stress treatments
Arabidopsis thaliana (Col-0) and Thellungiella halophila (Shandong) were used (Inan et al., 2004). Surface-sterilized seeds suspended in Murashige and Skoog (MS) medium (Gibco, Carlsbad, CA, USA) containing 2% sucrose were germinated at 24°C with gentle shaking (16/8 h light/dark; approximately 150 μmol photons m−2 s−1). Two-week-old seedlings were transferred to pots filled with isolite artificial soil (Sundine Enterprises, Arvada, CO, USA) for hydroponic culture. Plants were supplied with 0.5× Hoagland's nutrient solution with 4× increased Fe.
Four-week-old Arabidopsis and 6-week-old Thellungiella, similar in size and before bolting, were irrigated with 150 mm NaCl (At-150, Th-150), and one set of Thellungiella plants with 250 mm NaCl (Th-250). Plants remained in the presence of the chosen NaCl solution, or nutrient solution for controls. After treatment for 3 and 24 h, respectively, plants, at least 10 per sample, were frozen in liquid N2. Two biological repeats, grown separately at different times, were used. We chose these time points because a previous manuscript (Taji et al., 2004) had shown that, within the initial 12 h salt stress, both species showed an increase in NaCl uptake; after 24 h Arabidopsis showed a marginally higher concentration of sodium.
Total RNA was isolated (RNeasy, Qiagen, Carlsbad, CA, USA). Glass microarray slides consisting of 70-mer oligonucleotide probes (http://ag.arizona.edu/microarray) were used in hybridizations. RNA samples (70 μg each) for control and treatment conditions were reverse-transcribed (SuperScript III, Invitrogen, Carlsbad, CA, USA), and hybridization was carried out according to http://atarrays.tigr.org/arabprotocols.shtml (for detailed protocol see Appendix S2). To avoid bias in microarrays as a consequence of dye-related differences in labelling efficiency, dye labelling for each paired sample (stress/control) was swapped in one of three independent hybridizations. In total, 36 microarray hybridizations were carried out.
Microarray evaluation and statistical analysis
Signal intensities for each array element were collected (GenePix 4000B, Axon Instruments, Union City, CA, USA) and images analysed (GenePix Pro 4.0). Spots with intensities lower than background or with an aberrant spot shape were flagged by the GenePix software and checked manually. The resulting GPR files were analysed by TIGR-TM4 (http://www.tm4.org;Saeed et al., 2003) and converted by ExpressConverter ver. 1.5 to generate co-ordinated MEV and ANN files. MEV files were processed through tigr-midas ver. 2.18. Total intensity normalization, Lowess (Locfit) normalization, standard deviation regulation, and intensity filtering were done within each slide. After normalization, columns IA and IB in dye-swap slides were swapped so that each column IA represented the control data and each column IB the stress data.
Statistical analyses were carried out using mev ver. 3.0.3, FuzzyK (http://rana.lbl.gov/FuzzyK/software.html) and spotfire (Spotfire, Somerville, MA, USA). In mev, a one-class t-test for each experimental condition, and a one-way anova for all six experimental conditions, were carried out separately to reveal patterns of regulation. In both cases permutation with a very high stringency (anova:P = 0.01, permutation = 1000; t-test: P = 0.01, permutation = 64) was used to find significantly differently regulated genes. Adjusted Bonferroni P-value correction was applied at the same time in order to reduce false discovery rate. The one-way anova output, log2-ratio of stressed to control, was then analysed with spotfire for hierarchical clustering and with FuzzyK (Gasch and Eisen, 2002) for fuzzy k-means clustering, including a PCA. Throughout we use the names of Arabidopsis homologs to identify Thellungiella genes.
Fuzzy k-means analysis was done with the FuzzyK program (http://rana.lbl.gov/FuzzyK/software.html) with the parameter k = 20, and 13 centroids were generated. Genes were grouped into a cluster if they had a 0.50 or higher membership value with a certain centroid. All together, 12 clusters were obtained corresponding to 12 centroids, and the last centroid (C12) was removed because no gene had a membership value higher than 0.50. The results were visualized via the software MapleTree (http://mapletree.sourceforge.net).
Transcript intensity analysis was carried out as follows: the raw intensities after background subtraction for each gene were extracted from the control column of all 24-h slides, 18 in total. After removal of all control spots, total intensity normalization was done with all 18 slides. Then the mean intensity value and standard deviation were calculated separately for each gene in the 12 Thellungiella and six Arabidopsis slides. Genes with a ratio of standard deviation to mean >0.50 were then filtered out. Finally, a t-test (P = 0.05) was performed with the remaining genes to find significant differences in transcript intensities.
In order to understand better the salt-stress response at the transcript level, the fuzzy k-means clustering results and transcript intensities under control conditions were combined and analysed. For each gene appearing in the fuzzy k-means clustering results, the ratio r = ITh/IAt was calculated, with ITh and IAt representing averaged transcript intensity calculated from all 24-h Thellungiella control slides and 24-h Arabidopsis control slides. The individual ratios within each cluster were averaged, resulting in 12 averaged ratios.
For verification of the hybridization results, real-time PCR was conducted. RNA (2 μg) from each experimental condition was used for first-strand cDNA synthesis exactly as described. Primer sequences are listed in Table S2. Thellungiella actin (CX129618) cDNA primers and Arabidopsis Ubiquitin-10 cDNA primers were used as internal controls. Detection of RT-PCR products used the fluorescent dye SYBR-green (Applied Biosystems, Foster City, CA, USA) and the ABI PRISM/Taqman 7700 Sequence Detection System (Applied Biosystems). Three repeats per gene were done, and averaged threshold-cycle numbers were used to calculate changes (log2).
Aliquots of the material used for microarrays in liquid N2 were used and metabolites were extracted (Fiehn et al., 2000a,b). Tissue (50 mg) was extracted and separated into polar (methanol/water) and apolar/lipid (chloroform) phases. Internal standards (acetic acid, proline, ribitol, trehalose) were added to a separate set of samples before extractions. Vacuum-dried polar fractions were derivatized (Fiehn et al., 2000b). Samples (0.5 μl; split ratio 40:1) were injected into a GC-HP5890 (Spectron-HP, Kirkland, WA, USA) connected to a Micromass 70-VSE magnetic sector mass spectrometer (injection temperature 250°C; interface 315°C; ion source kept at 200°C). Mass spectral m/z range was 50–800. Masses were compared with data of the National Institute of Standards and Technology (NIST) and its replicates collection (NISTREP), our own libraries, and reference compounds for peak identification.
The authors thank Dr John Quackenbush and the TM4 group, TIGR, for help with statistical analyses. Also, we thank Drs Valeriy Poroyko, Alexander Ulanov and Dong-ha Oh, University of Illinois at Urbana-Champaign (UIUC), for discussions and advice on manuscript preparation. Space constraints made it impossible to list the publications characterizing the 341 functionally characterized salt stress-regulated genes included in Table S1. The work has been supported by NSF DBI-0223905/DBI-0500617, the UIUC Plant Biology and Physiological and Molecular Plant Biology graduate programs, and UIUC institutional funds.