Core genome responses involved in acclimation to high temperature

Plants can acclimate rapidly to environmental conditions including high temperatures. To identify molecular events important for acquired thermotolerance, we compared viability and transcript profiles of Arabidopsis thaliana treated to severe heat stress (45 o C) without acclimation, or following two different acclimation treatments. Notably, a gradual increase to 45 o C (G-acclimation; 22 o C to 45 o C over 6 hr) led to higher survival, and to more and higher-fold transcript changes, than a step-wise acclimation (S-acclimation; 90 min at 38 o C plus 120 min at 22 o C before 45 o C). There were significant differences in the total spectrum of transcript changes in the two treatments, but core components of heat acclimation were apparent in the overlap between treatments, emphasizing the importance of performing transcriptome analysis in the context of physiological response. In addition to documenting increases in transcripts of specific genes involved in processes predicted to be required for thermotolerance (i.e. protection of proteins and of translation, limiting oxidative stress), we also found decreases in transcripts (i.e. for programmed cell death, basic metabolism and biotic stress responses), which are likely equally important for acclimation. Similar protective effects may also be achieved differently, such as prevention of proline accumulation, which is toxic at elevated temperatures and which was reduced by both acclimation treatments, but was associated with transcript changes predicted to either reduce proline synthesis or increase degradation in the two acclimation treatments. Finally, phenotypic analysis of T-DNA insertion mutants of genes identified in this analysis defined eight new genes involved in heat acclimation, including cytosolic ascorbate peroxidase (APX2), and the transcription factors HsfA7a and NF-X1. conductance, Ten other mutants shown to be homozygous but to not have a heat stress phenotype were HSFB1 At4g36990; expressed protein At5g67600; LTI78 At5g52310; DREB2A At5g05410; Fer1 At5g01600; immunophilin At4g25340; stress induced protein At4g12400; ROF1 At3g25230; galactinol synthase At2g47180; and HSFA2 At2g26150. These mutants were not backcrossed to wild type because of the absence of phenotype.


Abstract
Plants can acclimate rapidly to environmental conditions including high temperatures. To identify molecular events important for acquired thermotolerance, we compared viability and transcript profiles of Arabidopsis thaliana treated to severe heat stress ( . There were significant differences in the total spectrum of transcript changes in the two treatments, but core components of heat acclimation were apparent in the overlap between treatments, emphasizing the importance of performing transcriptome analysis in the context of physiological response. In addition to documenting increases in transcripts of specific genes involved in processes predicted to be required for thermotolerance (i.e. protection of proteins and of translation, limiting oxidative stress), we also found decreases in transcripts (i.e. for programmed cell death, basic metabolism and biotic stress responses), which are likely equally important for acclimation. Similar protective effects may also be achieved differently, such as prevention of proline accumulation, which is toxic at elevated temperatures and which was reduced by both acclimation treatments, but was complex network of changes in plants, only one of which is the production of HSPs/molecular chaperones.
The response of the Arabidopsis transcriptome to heat stress has been examined by several groups (Rizhsky et al., 2004;Busch et al., 2005;Lim et al., 2006;Schramm et al., 2006;Kilian et al., 2007). These studies, however, have not been designed to determine how the observed transcriptional changes correlate with damage from the stress treatment, or with the acquisition of thermotolerance. All of these studies noted the up-regulation of transcripts for well-characterized HSPs, including Hsp101, Hsp70s, and small HSPs. Each study also identified additional transcripts that increased dramatically with heat treatment, a number of which were observed by more than one group. These include members of the DREB2 family of transcription factors, galactinol synthase (At2g47180) and other enzymes in the raffinose oligosaccharide pathway, and ascorbate peroxidase 2 (APX2; At3g09640).
We have been interested in defining genes whose expression is causally related to the ability of plants to acclimate to high temperature, i.e. to acquire thermotolerance. In previous work we used both forward and reverse genetics to identify genes/pathways essential for this process (Larkindale and Knight, 2002;Hong et al., 2003;Larkindale et al., 2005b;Larkindale et al., 2005a). Here we describe results of transcriptome studies designed to identify additional components important for acquired thermotolerance.
Although previous studies have analyzed transcript profiles of plants subjected to high temperatures, these studies have not included parallel quantitative analysis of plant (or cell) survival, nor did they consider the degree to which treated plants develop resistance to further extreme stress. In order to identify transcript changes specifically associated with thermotolerance, we compared viability and global transcript profiles of plants acclimated plants, which is more typical of the natural environment, is much more effective in protecting plants from subsequent heat stress than the S-acclimation treatment. Transcriptome analysis of heat acclimation not only predicts genes and gene clusters whose increased expression is likely to be involved in thermotolerance, but also indicates that decreases in expression of specific genes may be equally critical.
Examining insertional mutants of highly heat regulated genes has also confirmed the importance of new factors in heat acclimation.

Different heat acclimation treatments induce different degrees of thermotolerance.
To identify changes in gene expression that could be causally related to the acquisition of thermotolerance, we first characterized the effectiveness of two different acclimation treatments in inducing tolerance to an otherwise lethal 45 o C stress (Fig. 1). We reasoned that changes in gene expression common to more than one acclimation treatment are more likely to be required for thermotolerance than changes seen in a single type of acclimation treatment. In one acclimation treatment, typical of treatments we and others have used in the laboratory and designated "S" for step-acclimation, plants were acclimated by heating at a constant, non-lethal temperature (38 o C), followed by a 22 o C period prior to the 45 o C stress (Hong and Vierling, 2000;Larkindale et al., 2005b). The other treatment was designed to mimic temperature changes in the field; the temperature was increased linearly from 22 to 45 o C over 6 h (designated "G" for gradientacclimation). Both acclimation treatments were compared to non-acclimated plants were also applied to 2.5 d-old dark-grown seedlings with similar results (Supp. Fig. 1).
Thus, gradual increases in temperature, which are more likely to occur under natural 6 conditions, allow the plant to acclimate better than short, abrupt exposure to high, but non-lethal temperatures.
Thermotolerance treatments affect many gene transcripts.
To identify genes which could be involved in the acquisition of thermotolerance, wholegenome microarrays (Affymetrix At-H1) were used to analyze transcript levels. Plant samples were taken over the time course of the G, S and D heat treatments as shown in conditions that result in similar viability of G and S-acclimated plants (Fig. 2). Recovery samples were taken when plants were still green and not visibly damaged (G, S and D 45R). Array data were averaged for duplicate samples and filtered as described in Materials and Methods. Of the 22,746 genes on the arrays, 4724 genes passed the filtering criteria and were analyzed further. As compared to unheated samples, large numbers of transcripts increased or decreased during heating, but the G, S and D treatments affected transcript levels differently. Fig. 3A shows the total number of transcripts which increased in abundance either 2 to 5-fold or >5-fold, and also the proportions of transcripts which showed different absolute levels of accumulation. At 45 o C many fewer transcripts change in abundance in non-acclimated plants (D45) than in acclimated plants (S45 or G45), consistent with acclimation protecting processes essential to transcription and RNA stability. Notably, plants subjected to G-acclimation showed the largest number of altered transcripts (~1,600 up, ~2,500 down), as well as more transcripts with greater fold-changes and higher absolute expression levels than in the S or D treatments. The higher transcript levels in G compared to S samples may contribute to the increased heat tolerance of G-acclimated plants.
The category of transcripts which increased 5-fold or more and accumulated to >5,000 AU are likely to represent major effector proteins involved in repair of and recovery from heat damage. There are 185 of these transcripts which are up-regulated by 5-fold or greater in at least one sample, as listed in Supplemental Table 1, approximately one third of which are HSPs/molecular chaperones. In addition to HSPs and other chaperones, there are 5 petidyl prolyl isomerases, and 9 transcripts associated with energy 7 metabolism, in particular, enzymes involved in glycolysis (At4g26270, At2g36460, At5g17310, At5g56630, and At1g79550). There were also 10 non-HSP stress proteins, including two universal stress proteins and several cold-and drought regulated proteins.
Significantly more transcripts decreased in abundance under all of the heat treatments than increased in abundance, but again, the greatest change occurred in the most heat tolerant, G-acclimated plants (Fig. 3B). These data suggest RNA turnover and suppression of transcription are also critical for heat tolerance. In contrast to transcripts that increased, 70-90% of the decreased transcripts were of low abundance (<500 AU), and there were few transcripts in the >5000 AU abundance classes. Interestingly, the >5000AU expression class included 7 pathogen defense-related transcripts (At4g11600, At3g50950, At4g21870, At4g36010, At4g36040, At5g06320 and At5g20630), and four pEARLI-like proteins, which are up-regulated by aluminum and cold stress (Bubier and Schläppi, 2004).

Total genome changes in transcript levels vary with acclimation treatment.
We sought to identify genes common to G and S-acclimation which were unaffected in plants heated directly to 45 o C (D samples), as these may be critical to thermotolerance.
We first used statistical clustering of the complete filtered array data sets to compare overall patterns of transcript abundance. Figure 4A shows the result of Euclidean clustering using average linkage of the different array samples (BRB ArrayTools). For reference, biological replicates had similarities of less than 30AU. Unexpectedly, the G and S-acclimation treatments showed significant differences. The overall pattern of transcript changes during the G-acclimation treatment is similar for all three G samples (GAcc, G45, G45R), but distinct from the S and D samples (Euclidian distance of 95).
Thus, although both G-and S-acclimated plants are thermotolerant, more and different changes occurred during G than in S-acclimation. The non-acclimated D samples are also more similar to S-acclimated samples than to the unheated samples, emphasizing that large numbers of transcripts are observed to increase even when plants are treated at temperatures that ultimately lead to plant death.
Changes in specific sets of transcripts are unique to thermotolerant plants.
To define a subset of transcripts unique to thermotolerant plants we first compared sets of transcripts which accumulated at least 2-fold more in the three thermotolerant samples, S Acc1, S Acc2 and G Acc, relative to the unheated sample (Fig. 4B). We suggest that transcripts common to the thermotolerant samples are likely critical to the acclimation process. These include 377 transcripts common to all three samples, 240 transcripts increased in both S Acc1 and G Acc, and 220 common to S Acc2 and G Acc. We next compared transcript sets increased during 45 o C stress in acclimated (G45 and S45) and non-acclimated plants (D45) (Fig. 4C). There were 229 transcripts increased in all heated samples, which not surprisingly included 33 HSPs/molecular chaperones and 6 heat shock transcription factors (HSFs). We also found 242 transcripts that increased during recovery in thermotolerant plants (S45R, G45R), but not in non-acclimated plants (Fig.   4D).
Using the comparisons above, we defined transcripts that increased uniquely in thermotolerant plants as those genes found in the intersection of the subsets highlighted in bold and underlined in Fig. 4, Panels B, C and D. There were 57 up-regulated genes that fit these criteria (Supp. Table 2). Using this stringent definition, there were only two transcription factors in this thermotolerance-specific subset: heat shock transcription factor HsfA3 (At5g03720) and the DREB2B transcription factor, which has been linked to drought stress (At3g11020) (Nakashima et al., 2000). The list also includes three known stress proteins: chloroplast-localized Hsp70-7 (At4g24280); cold-stress related kin1/COR6.6 (At5g15960), and a universal stress protein (At3g03270). Analysis with MIPS FunCat indicates that compared to the whole genome this up-regulated subset contains a significantly greater percentage of genes related to protein fate (At2g15790, At4g03320, At5g27660 and At1g01940, two of which are petidyl prolyl isomerases) , or involved in energy metabolism (At4g10040, At2g34590, At5g03340, At5g17310). About a quarter of the genes have no assigned function, as is typical of the entire genome.
When the same analysis was performed to identify down-regulated genes unique to thermotolerant samples, 69 genes were identified (Supp. Fig. 3 and Supp. Table 3). The down-regulated genes define diverse functions that are notably distinct from those of upregulated genes. Genes found are related to pathogenesis or disease, five cytochrome P450 genes, and an unusually large number of protein kinases (nine). There are no genes in the protein fate or energy metabolism categories, but five in the cell defense and virulence category (At5g36910, At4g16860, At4g16880, At3g44970 and At5g55990).

Clusters of genes up-regulated during the acquisition of thermotolerance.
While genes expressed uniquely in thermotolerant plants may provide critical functions for optimal survival of heat stress, genes whose transcripts increase in non-thermotolerant plants, as well as differences in absolute transcript level, must also be considered. It is already established that chaperones/HSPs contribute to heat tolerance, and as noted above, transcripts of many chaperones/HSPs increase during heating even in nonthermotolerant plants which subsequently die. Therefore, to define additional groups of genes which might be essential for thermotolerance, we used cluster analysis to identify genes showing similar patterns of regulation. Cluster analysis was performed using Euclidean distance and average linkage of transcript levels of all genes under all of the conditions. We identified 73 clusters, containing from one to 2124 genes (for complete list see Supplemental Table 4). Each cluster was analyzed with respect to known promoter elements, responses in other available array data sets, and expression patterns relative to the other clusters ( Figure 5 and Supplemental Table 5). All clusters with 10 or more genes showed a statistically significant enrichment of specific sequence motifs within 1000 bp 5' of the known or predicted transcription start site when compared to the whole genome. We focused on clusters that contained putative promoter binding sites of known transcription factors and that in many cases contained genes reported to be co-regulated under other stress conditions (Vogel et al., 2005;Sakuma et al., 2006).
It is not surprising that six clusters (#8, #41-45) included many genes with heat shock elements (HSEs, GAAnnTTC) as this motif binds HSFs (Nover et al., 1996). Genes in these clusters, which are up-regulated under all heating conditions (although not necessarily to the same extent), contain the majority of the highest fold-increased transcripts and include more HSPs (28 genes) than any other group of clusters. It is also transcripts in these clusters that have generally been identified in other microarray studies of heat-treated Arabidopsis plants or cell cultures (Rizhsky et al., 2004;Busch et al., 2005;Lim et al., 2006;Schramm et al., 2006). Cluster 45, which is equally induced in all three heat treatments here (S, G and D), includes 18 well-characterized HSPs (Hsp101, 14 small HSPs, and three Hsp70s). Genes in clusters #41-44 maintain somewhat higher levels in G-acclimated plants than in S-acclimated plants. In D plants the transcripts generally appear lower than in either acclimation treatment, but are still significantly induced. Clusters #41-44 comprise ~ 50% unknown proteins, but also include stressrelated proteins such as chaperonin 60s, organelle Hsp100/ClpB proteins, two DnaJs, petidyl prolyl isomerases, a glutaredoxin, and the cold-regulated protein COR6.6. While many of these transcripts are likely necessary for thermotolerance, their accumulation is not sufficient, as they also accumulate in non-acclimated plants.
The site II motif (TGGGCC) (Welchen and Gonzalez, 2005)  Of the clusters not listed in Figure 5, there were a limited number of up-regulated clusters. One small cluster, cluster 17 (six genes), includes two subunits of NADH dehydrogenase and a cytochrome c oxidase, up-regulated in all heated samples. Genes in clusters #4, 5, 34 and 52 (41 genes total) all increase more in acclimated plants, and more in G than in S, but these transcripts are predominantly of unknown function.

Clusters of genes down-regulated during the acquisition of thermotolerance.
Two groups of clusters that were down-regulated exclusively in acclimated plants were identified. One group (clusters #21, 22, 63B and 71) contained the W-box motif found in disease resistance genes (Raffaele et al., 2006) and contained 46 disease resistance genes (including classical pathogenesis response proteins PR1 and PR5), 10 cytochrome P450s and 17 transcripts associated with cell detoxification (mainly GSTs).
The other group of down-regulated clusters (clusters #46, 62, and 68) contained the sequence TATATA, which may be a form of TATA box (Molina and Grotewold, 2005), and included seven disease resistance genes, six transcripts associated with cell detoxification (mainly GSTs), 16 auxin regulated genes, five expansins and eight cytochrome P450s. Interestingly, comparison with other available array data showed that these transcripts were also down-regulated in most of the stress and chemical treatment arrays (Genevestigator), suggesting that these may represent short-lived transcripts.
Most of the remaining clusters of genes not discussed above (as listed in Supplemental Table 4) were down-regulated and contained primarily disease resistance genes, auxininduced genes and enzymes for general metabolism (see Supplemental Table 4).

Transcripts associated with cytosolic protein synthesis increase in acclimated plants
We utilized MAPMAN array data visualization software (Thimm et al., 2004) as another approach to uncover processes related specifically to acclimation. A very dramatic difference between acclimated and non-acclimated plants was revealed in comparisons of transcripts associated with protein synthesis. Many cytosolic ribosomal proteins are significantly up-regulated throughout G-acclimation and are at least transiently upregulated (though to a lesser extent) during S-acclimation, but remain notably unchanged in non-acclimated plants (Supplemental Fig. 3). This observation is consistent with results of the cluster analysis, as these genes fall primarily in clusters containing Site II promoter elements. At the same time chloroplast ribosomal protein transcripts are significantly repressed in acclimated plants only. Transcripts associated with translation initiation, elongation and release are induced by both acclimation treatments, but more so during G-acclimation. The fact that none of these transcripts were induced at either the pre-treated or the D45 time points, explains why they were not identified in previous heat stress transcriptome studies. These data are consistent with the observation that protein synthesis can be acclimated to higher temperatures (Black and Subjeck, 1989), and indicate that early events of acclimation enable enhancement of this transcriptional program.

Differential responses to other abiotic and biotic stresses and to programmed cell death with acclimation
Visualization of changes in stress-related transcripts using MAPMAN also provides insight into potential interactions of stress response networks (Supplemental Fig. 4).
Transcripts induced by other abiotic stresses show varied patterns of change. Coldregulated transcripts accumulate more in S-acclimation than in G-acclimation, which reinforces the observation that clusters identified as enriched for genes with putative DRE elements are mostly increased in S-acclimation only (Fig. 5). Many transcripts associated with oxidative stress are also primarily increased in S-acclimation only, although peroxiredoxins and catalase transcripts accumulate under G-acclimation as well. These data are consistent with the fact that S-acclimated plants are less thermotolerant than Gacclimated plants and suggest that S-acclimated plants experience greater stress than Gacclimated plants.
Another striking transcriptional change in stress transcripts observed in this MAPMAN analysis concerns biotic stress. There is an obvious decrease in large numbers of transcripts associated with biotic stress in acclimated plants only. The magnitude of the decrease is also greatest for G-acclimated plants. This visualization draws together the down-regulation of disease-related transcripts seen in multiple clusters, including those with W-box motifs (Fig. 5).

Different transcriptional programs may achieve similar physiological changes
Transcripts increased in only G or S-acclimation cannot themselves be essential for thermotolerance. However, they may control similar protective physiological processes by different mechanisms. Levels of transcripts involved in proline metabolism illustrate this point. It was recently shown that proline accumulation is toxic to plants at high temperatures (Rizhsky et al., 2004). Free proline levels are controlled primarily by the balance of synthesis and degradation catalyzed by the key enzymes ∆ -1-pyrroline-5carboxylate synthetase (P5CS, At3g55610) and proline oxidase/dehydrogenase (At3g30775) (ProOx) (Hayashi et al., 2000). Notably, during S-acclimation, the transcript for ProOx (cluster 63A) accumulates over 5-fold, consistent with an increased ability to remove proline. In contrast, the ProOx transcript remains low during Gacclimation (Fig. 7A) To further test the importance of proline regulation for thermotolerance, we studied two T-DNA insertion alleles of the ProOx gene (Supp. Fig. 5) and tested ability of the mutants to acquire thermotolerance. Both alleles were unable to acquire thermotolerance normally under either S or G-acclimation conditions ( Fig. 6B and C, 7C), and both alleles accumulated even higher levels of proline than wild type, both under control and heated conditions (Fig. 7B). These results further confirm the importance of controlling proline levels during heat stress. We suggest that a different balance in the regulation of proline synthesis vs. degradation prevents proline accumulation at high temperature in S-and G-acclimated plants.

Identification of genes contributing to thermotolerance.
To test the significance of some of the observed transcript changes, early on in this study we selected 30 genes for mutant analysis (Supplemental Table 6) from the list of 185 genes up-regulated more than five-fold by heat treatment and showing >5,000 AU of expression (Supplemental Table 1). The criteria for selection included the availability of T-DNA insertion mutant lines and excluded, in general, members of gene families with predicted redundancy (although there are exceptions). This list included the ProOx gene described above (At3g30775), and we also added a putative choline kinase gene (for a total of 31 genes), which showed only >4,000 AU of expression at maximum induction, but which was over 10-fold induced, and which was known to be regulated by other stresses (Summers and Weretilnyk, 1993;Tasseva et al., 2004). T-DNA insertion lines for16 these genes were screened by PCR to identify homozygous mutant lines. Of the 16 tested to identify homozygotes, we were not successful in recovering homozygous lines for two mutants (At5g15450 & At3g24500). Failure to obtain homozygotes of At5g15450, ClpB-p1, was clearly due to lethality of the homozygous mutant as determined from subsequent studies (Lee et al., 2007), whereas insufficient material may have been screened in the case of At3g24500, MBF1c, as others have successfully obtained homozygotes from the same lines (Suzuki et al., 2005). Of the 14 homozygous lines obtained, four were found to have a heat acclimation phenotype (At3g09640, At1g79929, At3g51910, At3g30775) (Table 1).
For the other 15 of the 31 lines, at least 12 progeny from each line were tested for phenotype directly, using segregating material obtained from the stock center. Of these 15 mutants, four showed a distinct heat acclimation phenotype (At4g23570, At1g10170, At4g36010, At1g74320), and were subsequently found to be homozygous insertion lines.
The other 11 mutants showed no evidence of heat sensitivity among the seedlings tested.
It remains possible that some of these genes may prove important for acclimation, but they have not been further analyzed. In total, those lines analyzed as homozygotes but not having a phenotype were HSFB1 At4g36990, expressed protein At5g67600, LTI78 At5g52310, DREB2A At5g05410, Fer1 At5g01600, immunophilin At4g25340, stress induced protein At4g12400, ROF1 At3g25230, galactinol synthase At2g47180, HSFA2 At2g26150 (Table 1 and Supplemental Table I). We have not rigorously demonstrated that these homozygous mutants without phenotype are indeed null mutants, although the position of the T-DNA insertions early within the gene structure for all but At5g52310, At4g25340, and At4g12400 is consistent with this interpretation (data not shown).
Of the remaining eight T-DNA insertion mutants with heat acclimation phenotypes, in addition to ProOx, second insertion alleles were obtained for APX2 (At3g09640), and thaumatin (At4g36010), and all mutants were then backcrossed once to wild type and homozygous mutant lines reselected and used to obtain the data in Table 1. Absence of detectable full length RNA in all mutants was confirmed by RT-PCR (Supplemental Fig.   5), indicating that the mutants most likely are null alleles. These eight mutants represent multiple important functions and had heat sensitive phenotypes ranging from mild to severe, although none was as severe as a mutation in Hsp101 (At1g74310), which was completely unviable after the treatments used (Table 1). The most severely defective mutants were APX2, a cytosolic ascorbate peroxidase, and two transcription factors, HsfA7a (At3g51910) and NF-X1 (At1g10170), all showing 20-30% viability compared to wild type under both the step and gradient acclimation treatments. The next most severe mutants were ProOx and SGT1a (At4g23570), a factor implicated in Hsp90 function (Takahashi et al., 2003;Azevedo et al., 2006). The Hsp110 (At1g79920), choline kinase (At1g74320) and thaumatin (At4g36010) mutants showed moderate, but significant phenotypes. These mutants were also tested in the hypocotyl elongation assay, and all behaved as wild type, with the exception of a mild phenotype for the ProOx mutants (not shown). This is consistent with previous data showing that there can be significant differences in heat acclimation phenotypes of mutants at different growth stages (Hong and Vierling, 2000;Larkindale et al., 2005b). Notably, all of these mutants grew identically to wild type in the absence of heat stress, indicating that their heat sensitivity is not compounded by poor growth. In total we conclude that these eight genes, including ProOx, represent genes that are important for acquired thermotolerance.

Discussion
We have defined transcript profiles associated with plant acclimation to heat stress, revealing the complexity of molecular events that contribute to thermotolerance. We to 45 o C than no acclimation. Interestingly, G-acclimation, designed to mimic the gradual increase in temperature that would be experienced in the natural environment, induced greater heat tolerance than S-acclimation, a treatment typical of those used in heat stress and heat acclimation studies in plants, yeast, bacteria and mammalian cells (Knop et al., 1985;Mackey and Derrick, 1990;Hong and Vierling, 2000). The difference in response to these two treatments indicates the importance of placing transcript profiling results in the context of physiological responses. Our whole genome transcript profiling defined specific transcripts (Fig. 4 Table 5) and physiological processes (Fig. 7, Supplemental Figs. 4 & 5) that are up-regulated as well as down-regulated in thermotolerant plants. In the response to heat and the process of heat acclimation we suggest that decreases in gene expression may be of more importance than previously recognized, as non-thermotolerance plants fail to make many of these adjustments. Genes whose transcripts specifically increase associated with thermotolerance, as distinct from genes regulated by heat even in plants that fail to acclimate, have also been identified and can now be tested for their role with available mutants. Finally, among the genes with abundant transcripts that are most highly heat-induced, we have identified eight new genes required for maximum heat acclimation, including the cytosolic ascorbate peroxidase APX2, and the transcription factors HsfA7a and NF-X1.
Several other studies have examined the response of the Arabidopsis transcriptome to heat stress, but not as related to heat acclimation. Rizhsky et al. (2004) Table 5).
Besides the HSE-containing clusters, most of the transcripts in the other clusters showed no simple pattern of change compared to other studies in which the heat stress transcriptome has been examined. As many of these were transiently increased, or changed only under specific treatments, it is not surprising that such differences are seen.
Interestingly, however, many clusters identified here showed significant correlations with patterns of gene regulation in array studies involving other stresses. Several different clusters of transcripts (HSE clusters, DRE clusters, site II clusters 47, 54 and 56, TATATA clusters, and clusters 4, 59, 65 and 70) showed a similar up-and downregulation in response to norflurazon and syringolin as they did to heat (Genevestigator, (Michel et al., 2006)). Other clusters, such as the putatively DRE-regulated clusters, were en masse up-regulated in response to cold and anoxia (Genevestigator;(Branco-Price et al., 2005;Gonzali et al., 2005;Vogel et al., 2005;Oono et al., 2006). The putatively Wbox regulated clusters showed induction in response to biotic stress, but repression during heat acclimation. These data suggest that there may be common factors co-regulating transcripts within a specific cluster under different conditions. The greater thermotolerance of G-acclimated compared to S-acclimated plants appears to be due to bona fide differences in the molecular events leading to thermotolerance between the two treatments. Differences in the amount of time plants were exposed to an acclimating temperature do not appear to account for the greater viability of G-acclimated plants. In the treatments analyzed here, plants begin to express HSPs during Gacclimation when the temperature reaches There were a number of transcriptional changes which occurred only during Sacclimation. This may be attributed to thermotolerance being induced through different pathways by the two treatments, or to differences in the stress being perceived by the plant. Unique S-acclimation transcripts were primarily those in the DRE cluster 63A, which is also up-regulated by cold, drought, hydrogen peroxide and anoxic stresses (Supp. Table 5). The significantly greater up-regulation of general stress-related transcripts during S-acclimation can be seen clearly in Supplemental Fig. 4. This suggests that under S-acclimation the plant is experiencing other stresses secondary to heat stress.
S-acclimation may be a "shock response" whereby the plant responds by repairing stressinduced damage. In comparison, the greater numbers and magnitudes of changes in transcript abundance during G-acclimation suggest a more adaptive response whereby damage is prevented; the plant reduces its general metabolism and may reduce damage that could occur due to build up of toxic intermediates. These data further emphasize how the nature of the applied stress significantly affects the outcome of transcriptome studies and supports the need for carefully controlled conditions and documentation of associated physiological responses.
Despite the differences in transcript profiles between the two acclimation treatments, ultimately the same systems must be protected/repaired. We hypothesize that the same physiological outcome might be achieved in different ways. One example of this is the accumulation of proline, where we suggest that lower levels of this toxic metabolite are achieved by reduction of synthesis in G-acclimation and increased degradation in Sacclimation. We show that the ability to degrade proline through proline oxidase is essential for thermotolerance. Mutants of ProOx showed reduced thermotolerance, had higher proline levels prior to heat stress, and had similar proline levels after heat stress in both acclimated and non-acclimated plants. By comparison, wild-type plants subjected to either S-or G-acclimation did not accumulate high levels of proline during heat treatment. Previous work has shown that high levels of proline result in programmed cell death in Arabidopsis at high temperature (Rizhsky et al., 2004). In that study, proline levels were shown to increase during drought stress, but remain low during simultaneous drought and heat stress, or heat stress alone (6h at 38 o C, temperature used in Sacclimation). Thus we show that both acclimation treatments prevent heat-induced proline induction, but that the mechanism employed may be different for the two treatments.
It is well accepted that protection and refolding of cellular proteins through the HSP network of molecular chaperones are important for survival of high temperature stress (Vierling, 1991;Larkindale et al., 2005a), and many genes for these components appeared in HSE-containing clusters which are up-regulated by all heat treatments. Four of the eight genes we found to have a reduced ability to acclimate to high temperature were in these clusters, APX2, Hsp110, SGT1a, and choline kinase (Table 1). APX2 has been documented as heat stress-induced in several studies (Shi et al., 2001;Larkindale and Huang, 2004;Schramm et al., 2006), and it appears to be regulated by HsfA2 (Schramm et al., 2006). Previous studies also indicate it is involved in survival of high light stress (Rossel et al., 2006;Giacomelli et al., 2007). It remains unclear why this particular cytosolic APX isoform is required for heat tolerance (Arabidopsis has two other cytosolic APX proteins (Panchuk et al., 2005)). Determining whether this results from a specific property of the APX2 isoform, or if it is the result of the heat regulation will be of interest to determine. Hsp110, SGT1a and choline kinase had milder but significant effects on heat acclimation. Hsp110 is a member of the larger Hsp70 family (Lin et al., 2001), but unlike Hsp70s, does not appear to be essential for normal growth.
SGT1a is associated with Hsp90 and RAR1 in the development of R-protein modulate disease resistance (Niikura and Kitagawa, 2003;Takahashi et al., 2003;Azevedo et al., 2006), and this is the first indication it may also be important for abiotic stress. Little is known about choline kinases in plants, but their induction by other stresses, specifically salt stress is documented (Summers and Weretilnyk, 1993;Shank et al., 2001;Tasseva et al., 2004), and the potential importance of this enzyme in heat tolerance could be due to subtle modulation of membrane structure and/or lipid signaling.
Although it has long been known that Hsfs are major transcription factors involved in gene transcription in response to heat stress, the relative roles of the 21 different Arabidopsis Hsfs as well as other transcription factors in heat acclimation is far from resolved. Of the 21 Hsfs, only Hsfs A1e, A2, A3, A7a, B1 and B2b were shown to be heat-induced. Charng et al. (2007) reported that insertion mutants of HsfA2 show a reduced ability to sustain thermotolerance (Charng et al., 2007), although they are not required for tolerance to short term heat stress as tested here. HsfA1a/A1b double mutants show some reduction in their ability to synthesize Hsps, but show little reduced thermotolerance (Busch et al., 2005). Here we show that HsfA7a is one of the Hsfs that contributes to heat acclimation. Determining whether this Hsf has specific targets, or works in conjunction with other Hsfs will be of interest to investigate. Among the other Hsfs, we have recently shown that HsfA3, which is heat-induced and found in the DRE containing clusters in our study, is regulated by the Dreb transcription factors Dreb2A and Dreb2B. Furthermore, HsfA3 RNAi plants and an HsfA3 T-DNA insertion mutant have reduced heat tolerance (Schramm et al., In press). Notably HsfA3 and Dreb2B were the only two transcription factors among the genes whose induction was unique to thermotolerant plants (Supplemental Table 2). Although others have reported heat sensitivity of insertion mutants of DREB2A plants, we have not found a defect in heat acclimation for either DREB2A or DREB2B mutants (unpublished). However, we would expect that double mutants of these two factors would be heat sensitive. Our data also identify another transcription factor involved in heat acclimation, NF-X1. NF-X1 also clustered with the DRE group, although it has no apparent DRE elements. This gene is also upregulated by a wide range of other biotic stresses (e.g. drought, hypoxia, oxidative stress, salt stress, wounding; see Genevestigator), and recent data show NF-X1 mutants have reduced growth and survival under salt stress (Lisso et al., 2006). Thus we can now begin to link heat acclimation and with drought/osmotic stress through the DREB2 and NF-X1 transcription factors.
The potential importance for stress acclimation of genes with the site II motif has not previously been recognized. Site II clusters up-regulated in both S and G-acclimation included splicing factors (At1g36730, At2g18510, At4g03430), a putative elongation factor (At2g38560), ribosomal proteins (At3g27450, At3g11120), and translation initiation factors (At1g36730, At2g04520). Other transcripts within these clusters were associated with ubiquitin and proteasomes (10 transcripts) and with protein folding (20 transcripts, including 6 chaperonins, 3 petidyl prolyl isomerases and 2 Hsp70s).
Importantly, site II cluster 57 (484 genes) was up-regulated exclusively in G-acclimated plants. It includes many additional translation initiation and elongation factors and ribosomal proteins, as well as photosystem subunits. The dramatic increase in components associated with cytosolic protein translation was further emphasized by visualization using MAPMAN (Supplemental Fig. 3). Thus G-acclimated plants appear well-positioned to repair and re-start translation and photosynthesis quickly after heat stress, and to maintain translation and photosynthesis to higher temperatures, no doubt contributing to their higher survival rate. The heat-induced translation initiation components may account in part for decreased ability to acclimate. Indeed, preliminary analysis of T-DNA insertion mutations in eIF1A-2 (At2g04520) and eIF5-2 (At1g36370) indicates that these factors are dispensable for normal growth, but contribute to heat tolerance (not shown).
Little attention has been paid to transcripts decreased during stress treatments, but results here suggest that decreases in specific transcripts play an important role in stress acclimation. Many transcripts associated with growth and general metabolism were down-regulated during S-acclimation and even more so in G-acclimation. Transcripts containing TATATA in their promoters also remained down-regulated in G45R samples, when other transcripts had returned to normal. Molina and Grotewold (2005) determined that only about 30% of the transcripts in the Arabidopsis genome have A/T rich sequences in their promoters and that these transcripts tended to be highly expressed under non-stress conditions. FUNCAT identified transcripts in these clusters as overrepresented in metabolic genes. Acclimated plants therefore appear to be effectively limiting non-essential cellular processes until the return to temperatures permissive for growth.
Our data further indicate that a decrease in the potential for induction of programmed cell death (PCD) is important for acclimation to high temperature in plants. High temperatures in non-acclimated plants have been shown to induce a form of PCD (Swidzinski et al., 2002), and in animal systems PCD is reported to be inhibited by HSPs (Beere, 2004). We saw decreases in cell death promoting transcripts and increases in cell death inhibiting transcripts only in thermotolerant plants. Furthermore, reports on treatments inducing PCD indicate heat-induced genes (especially those in the HSP clusters) are generally repressed, suggesting antagonism between acclimation and PCD pathways (Supp. Table 5). In contrast, chemical treatments which induce PCD in plants (syringolin, secreted by Pseudomonas syringae pv syringae and isoxaben, a herbicide which inhibits cellulose synthesis), mimicked the effects of heat on both up-and downregulated clusters of transcripts (Supp Table 5, Genevestigator). Thus there appears to be a complex interaction between heat stress, acclimation to high temperatures and the induction of PCD in plants.

Methods
Plant growth and heat treatment: Wild type Arabidopsis thaliana plants, ecotype Col-0, were grown on plates as described (Larkindale et al., 2005b). 10 d-old plants were heat-treated as in Fig. 1. All samples were taken in the light except the pretreated, S45 and D45 time points. All RNA and protein samples were taken 8.5 hr into the 16hr light cycle. For each sample plants were pooled from three plates (~300 plants). Microarray samples were repeated in duplicate, other assays in at least triplicate. Thermotolerance assays of 2.5 d dark-grown and 7 d light-grown seedlings were performed as described (Hong and Vierling, 2000;Larkindale et al., 2005b). Homozygous lines carrying T-DNA insertions were identified, ones with acclimation phenotypes were backcrossed once to wildtype plants, and homozygous F2 lines reselected and tested for expression of full length mRNA for the mutant gene (Supp. Fig.5).  Figure 7). For cluster analysis, data were further filtered such that genes were excluded if there was less than a 2-fold change between any individual array and the median value for that gene on all of the arrays, or if the gene was detected on only 1 or 2 of the 11 array samples. 4724 genes passed this filtering and were analyzed further. Clustering of both total transcript accumulation within a specific treatment, and of individual genes was done using Euclidean distance and average linkage in BRB array tools. Genes in each cluster were exported into DMT, and expression of the selected genes across all arrays was compared.
Promoter analysis was performed using 1000bp of upstream gene sequence obtained from TAIR (http://www.arabidopsis.org/). Cis elements were identified by blind searching for common 6-letter 'words' using TAIR's motif analysis selecting only those 'words' with a P value of <10 -4 , and by direct searching for defined non-6 letter cis element sequences. These words were then compared to known cis elements using PlantCare (http://oberon.fvms.ugent.be:8080/PlantCARE/index.html ) and Google (www.google.com ). When similarities to known cis elements were identified in clusters of interest, the original sequence was searched for the entire element and its variants.
Expression was compared to other conditions using Genevestigator (www.genevestigator.ethz.ch (Zimmermann et al., 2004)). For experiments in the database involving heat treatments, the array data were down-loaded and compared with array data from this study. Transcript levels relative to physiological pathways were analyzed using MAPMAN (Thimm et al., 2004) (http://gabi.rzpd.de/projects/MapMan/).

Proline Measurements:
Proline levels were measured using the method described by Bates et al. (1973). Plant tissue was harvested at the timepoints indicated and immediately frozen in liquid nitrogen. Tissues samples were ground in microfuge tubes in 3% (v/v) sulfosalicyclic acid and 10% celite resin, and incubated at room temperature overnight prior to assay. 0.5mL of sample was then added to 2mL ninhydrin (1% w/v in glacial acetic acid: water 60:40 v/v), and the tubes were boiled for 1 h. After cooling, the chromophore was extracted with 5mL of toluene, and the absorbance was measure at 520nm against a tissue-free blank. Amounts of proline were calculated by comparison to a standard curve.

Supplemental Data
SFigure 1: Thermotolerance of dark grown seedlings given different acclimation treatments. 2.5d-old seedlings were subjected to the treatments described in Fig. 1.
Hypocotyl elongation was measured 2.5-d after heat stress, and compared to that of unheated seedlings. Error bars represent the standard deviation of 12 seedlings in a single replicate. The experiment was repeated three times with similar results.  Gene transcripts up-regulated by at least 5-fold in one or more heated treatment, and with maximal expression levels greater than 5,000 AU. Cluster number refers to the analysis presented in Figure 5 in the text and in Supplemental Table 4. Genes in bold are the ones used in the mutant analysis.  mutant was homozygosed prior to looking for phenotype. Normal type indicates those mutants that were never tested for phenotype. Acknowledgements: Funded by NSF grant IBN-0213128 and USDA grant NRICGP 99-351007618 to E.V. Thanks also to Kevin Keisler (Microarray Core Facility, University of Arizona) for help with processing the microarray data, and David Mount for help with analysis. We also thank many past members of the Vierling lab who helped with initial screening and testing of many of the T-DNA mutants discussed in this study.