To capture as many as possible of the genes regulated by abiotic stress in P. euphratica, 17 different normalized and subtracted cDNA libraries were prepared from control and stress exposed trees. The treatments included several physiologically relevant abiotic stresses; salt, cold, drought, flooding and ozone (Additional data file 1). In addition, as a first step to elucidate the molecular mechanisms behind adaptation of P. euphratica to desert conditions, trees growing in the Ein Avdat valley were used for library construction (Figure 1). In total, 13,838 ESTs were obtained from the 17 libraries, with an average length of 478 nucleotides. The subtraction efficiency was estimated by 'electronic northern' analysis, that is, counting the abundance of various ESTs in different libraries (Additional data files 2 and 3). In the subtracted libraries prepared from leaves, all abundant photosynthesis genes were significantly removed (Additional data file 2) and in the subtracted libraries prepared from root tissues, genes with a putative role in abiotic stress were significantly enriched (Additional data file 3). The ESTs were submitted to GenBank with the accession numbers AJ767069 to AJ780913.

The 13,838 ESTs were analyzed and annotated within an openSputnik EST database as previously described 20. Clustering and assembly of the EST sequences yielded 7,841 unigene features. These unigenes were exhaustively annotated for features including, but not restricted to, probable peptide sequence, Uniprot 21 matches, sequence overlap with complete and incomplete plant genome collections, and Gene Ontology assignments 22. The entire database, including all annotative attributes and complete electronic northern analysis for all 17 EST libraries, can be viewed at the openSputnik website 23.

The P. euphratica EST sequence collection was compared to sequences taxonomically oriented to the rest of the plant kingdom. The P. euphratica unigenes were anchored using BLAST methods to derive plant sequence datasets using an arbitrary expectation value of 1e-10. The datasets were constructed to represent both completed genomes (P. trichocarpa, Arabidopsis thaliana and Oryza sativa) and openSputnik unigene collections from clades of taxonomically related sequences. The numbers of sequences that could be matched to a sequence collection are shown in Table 1. The P. euphratica EST collection was highly similar to the P. trichocarpa genome assembly 24; 97.6% of the ESTs can be found in the genome. Currently, approximately 250,000 ESTs from other Populus species are available from GenBank (Additional data file 4, dbEST release 062405). The P. euphratica EST collection overlaps these ESTs by 74%, which leaves approximately 2,000 novel ESTs in the present P. euphratica collection, indicating that the strategy of using normalized and subtracted stress enriched libraries has been successful at identifying novel Populus ESTs. Furthermore, the overlap of ESTs between P. euphratica and P. tremuloides, P. trichocarpa, P. tremula × P. tremuloides, P. euramericana, P. alba × P. tremula, P. trichocarpa × P. deltoides, P. nigra, P. tremula and P. deltoides varied between 21% and 63% (Additional data file 4). The coding content of Populus and Arabidopsis genomes has previously been shown to be highly similar 19. Consistent with this, the P. euphratica EST collection also has a high overlap (69%) with the Arabidopsis genome. Interestingly, 763 sequences (approximately 10%) were only present in this P. euphratica EST collection and in the P. trichocarpa genome and not in any other plant EST or genome collections. This further validates the use of normalized and stress subtracted libraries for identifying novel ESTs. A total of 54 sequences with significant length and coding potential are not present in any other sequence collections, and may thus represent P. euphratica specific genes.

To estimate the representation of stress related ESTs in the P. euphratica collection, a comparison was also made to other EST collections or DNA microarray experiments performed with stressed plants (P. euphratica, P. trichocarpa × P. deltoides, A. thaliana and Thellungiella halophila) (Additional data file 5). Depending on the particular species or stress treatment, about 36% to 75% of known stress induced genes 617252627 have an equivalent EST in the P. euphratica EST collection (Additional data file 5).

P. euphratica unigene sequences were functionally and structurally annotated using Gene Ontology 22, and the complete assignments are summarized in Table 2 using the plant specific 'GOSlim' terms 28 under the three main categories of biological process, cellular component and molecular function. A broad range of functions and processes were represented in the P. euphratica ESTs. Annotation of the EST sequences with respect to GO terms for molecular function, biological process, and cellular component can be accessed at the openSputnik website 23.

From the collection of 13,838 EST clones, a uni-gene set (selected using the CLOBB algorithm 29; data not shown) was reamplified. In addition, 491 ESTs were amplified twice and are duplicated on the array. All PCR products were re-sequenced to verify the identity of the EST. The ESTs that could be postively identified through sequencing correspond to 7,342 distinct unigenes derived from the clustering of the sequences (HPT2 and CAP3 methods as described in 20). To estimate the number of different genes on the array by allowing for the possibility of split unigenes, non-overlapping ESTs or unbridged assemblies, the unigenes were anchored into putative homologous groups using the available, but partial, Populus genome assembly. Using a stringent BLASTN expectation value of 1e-20, this identified 6,340 homologous groups. The data on the array may account for as many 6,340 distinct genes, and an undetermined number of paralogues.

The P. euphratica EST collection as described above contains several known abiotic stress regulated genes, but this is largely inferred from controlled laboratory experiments. To test which genes might be important for acclimation and adaptation to 'real' stress conditions, the P. euphratica microarray was used to further characterize trees grown in one of its natural habitats, the Ein Avdat valley, located in the Negev desert in Israel. In addition to measuring gene expression, we also analyzed leaves and soil for ionic content and the carbon stable isotope ratio δ¹³C/¹²C to estimate the level of salt and drought experienced by the trees.

We divided the valley into three different areas based on the distance from the small river running through it, distinct phenological differences of the trees growing on the canyon slope, as well as variations in the cambial activity and the width of the annual rings of the trees (Figure 1) 30. Trees growing close to the river have a wide trunk and a large and healthy crown (area A). Trees in the transition area (area B) have narrower trunks than trees in area A, but large and healthy crowns. Trees growing on rocky, dry ground on the upper slope and far from the water source (area C) have a narrow trunk and a small crown and a very constrained growth. Control trees were selected at the Ein Avdat parking lot, 1 km away from the valley (area P), where the trees were periodically irrigated to provide optimal water supply (see Materials and methods). The trees at the parking lot, planted in 1989, originate from root suckers of trees growing in the valley. Leaves were sampled from at least nine individual trees in each area A, B and C, and from seven trees in area P. P. euphratica displays a foliar dimorphism and juvenile leaves differ largely from adults. In this case, only adult, heart shaped leaves where collected from the shaded and unshaded regions of the crown.

Soil samples for the analysis of ion content were collected beneath each tree (close to the trunk at approximately 10 cm depth) in the four different areas. Na⁺ content was significantly higher (approximately 10 times) in the valley soil (areas A, B, and C) than in the soil from area P (Figure 2a). Other ions, including K⁺, did not differ significantly among the sites. P. euphratica trees growing in areas A and B accumulated more Na⁺ in their leaves compared to trees from the irrigated area P (Figure 2b). Ca²⁺ content was higher in leaves from areas A and C compared to those from area P.

An estimation of water use efficiency and thus the level of drought stress encountered by trees was obtained by measuring the carbon stable isotope ratio δ¹³C/¹²C (Figure 2c); δ¹³C/¹²C is a time-integrated index for the intrinsic water use efficiency, which is expected to increase during drought due to stomatal closure 31. The trees from the irrigated area P were more water-use efficient (less negative δ¹³C/¹²C) than those in the valley. This suggests that the trees from the parking lot suffered more from water shortage than the trees in the valley (due to intermittent irrigation). An alternative and more likely interpretation, however, is that trees in the valley were exposed to a stress leading to strongly reduced carbon assimilation, in this case probably a combination of Na⁺ and heat stress. In conclusion, trees from Ein Avdat areas A, B and C were exposed to salt and heat stress, but apparently not to drought stress.

For the microarray analysis, RNA was extracted from each of the nine individual trees sampled in areas A, B and C. Before labeling, RNA was pooled from three trees to give three biological repeats for each area. The same control RNA was used in all hybridizations and was prepared from a pool of leaves from seven trees growing at the parking lot (area P). Genes, grouped into functional categories, that consistently displayed higher or lower transcript levels in area A, B or C when compared to area P are shown in Additional data file 7 and are displayed graphically in Figure 3. Despite the highly different phenotypes of the trees in each area, their gene expression profiles were in most cases almost identical. Welch ANOVA analysis indicates that transcript levels for 22 genes (26%) showed significant differences between areas A, B and/or C, usually being either higher or lower in area C (indicated with an asterisk in Additional data file 7 and marked in red in Figure 3). About one third of the genes with higher transcript levels in area A, B and C have a putative role in response to abiotic stress. These include an aldehyde dehydrogenase and metallothioneins with putative roles in defense against oxidative stress. A plastid terminal oxidase that may play a role in photo-oxidative stress also had increased transcript levels.

In contrast to the relatively large number of genes with a putative role in defense against abiotic stress, no genes encoding transcription factors or other components of transcriptional regulation showed altered expression, although 153 ESTs on the array have the GO annotation 'regulation of transcription'. Consistent with this, only a few genes with a putative role in signal transduction (receptor-like Ser/Thr kinase, cyclic nucleotide and calmodulin-regulated ion channels and a phospholipase C) had increased transcript levels (Additional data file 7). As the salt stress imposed on the trees in the Ein Avdat valley is of a constitutive nature, the transcript level profiles detected in the microarray analysis are likely to be the end result of long term acclimation to saline desert conditions. Thus, genes responding rapidly after a stress treatment (including transcription factors) might be expected to be absent from the gene expression profile.

The transcript levels of several aquaporin water channels have been shown to be reduced during drought stress and to a smaller extent during salt stress 32. This may help in protection against water loss under abiotic stress conditions 33. In areas A, B and C, two aquaporins had decreased transcript levels. Other genes with lower transcript levels included those encoding chlorophyll a/b-binding proteins, chalcone synthase and dehydration-responsive protein RD22.

Seven ESTs representing genes with higher or lower transcript levels in the valley-grown trees than in area P trees, and spots with high and low hybridization intensities were used for quantitative real time RT-PCR (qPCR). An EST (AJ775831) encoding glucosidase II alpha subunit, which had displayed a constitutive expression level in all analyses performed with the P. euphratica DNA microarray, was used as an internal control to which gene expression was normalized (ΔC_T). The same RNA samples from areas A, B, and C and the area P control that were used in the DNA microarray analysis were used as templates in qPCR (Table 3). For most of the genes, transcript fold ratios were close to the microarray results in both analyses, indicating the reliability of the P. euphratica DNA microarray. One gene (AJ780552) displayed a higher fold-induction in the PCR analysis, which probably reflects the higher dynamic range of qPCR compared to array analysis 34. For only one gene (AJ769631) where the DNA microarray data implied a minor change in expression levels did the qPCR not show the same differential expression.

The accumulation of 22 different metabolites in leaves of P. euphratica trees was examined by gas chromatography coupled to mass spectrometry (GC-MS). The metabolites were analyzed initially in the leaves used also for the transcriptome analysis in areas A, B and C but not in P. The analysis was repeated during the following year for areas A, B, C and P. We present data from this last sampling as metabolite contents in areas A, B, and C did not change with sampling dates. Metabolite profiles of leaf extracts from six individual plants from each area (A, B, C, and P) were compared (Table 4). Trees from area A (which accumulated more Na⁺; Figure 2a) displayed significantly higher levels of the amino acids β-alanine, valine and proline than trees from other areas, including the less saline area P. Proline has previously been shown to accumulate to high levels in Na⁺ or osmotically stressed P. euphratica 35. Likewise, glycerol, glyceric acid, and myo-inositol had also significantly higher concentrations in trees from area A compared to trees from the other areas. Fructose and mannitol, on the contrary, were detected in significantly lower concentrations in trees from the area A. No significant and/or clear differences between the trees from different areas were detected in any of the other metabolites analyzed.

For one of the metabolites detected, galactinol, the corresponding biosynthesis gene (galactinol synthase) had approximately threefold higher transcript levels in areas A, B and C (Additional data file 7). Galactinol itself did not display any significant increase in the leaves. Nevertheless, galactinol is a precursor of raffinose, which showed significant increases in areas B and C (Table 4). Significant changes were also detected in myo-inositol, which is a precursor of galactinol 36.

Soil in the Ein Avdat valley contains high concentrations of Na⁺ (Figure 2a). This is also reflected in the vegetation composition in the valley, which includes several halophytic species (Juncus maritimus Lam, Suaeda vera, Atriplex halimus, Imperata cylindrical, Limonium pruinosum and Zygophyllum dumosum Boiss.). Maintenance of cellular ion homeostasis is important for vital activities in all organisms. High salinity, in addition to osmotic stress, causes ion imbalances and consequently cell death. K⁺ is one of the major cations in the cytosol and is responsible for the maintenance of membrane potential. Under normal growth conditions, K⁺ concentration in plant cells is higher than Na⁺ concentration. High Na⁺ in soil causes inhibition of K⁺ uptake, resulting in a severe increment in the Na⁺/K⁺ ratio, which is detrimental for photosynthetic activity and causes inhibition of growth and ultimately cell death 37. P. euphratica trees grown in highly saline soils (Figure 2a) accumulated high Na⁺ concentration in leaves (approximately twofold more Na⁺ than K⁺; Figure 2b) and still continued growing without visible symptoms of severe stress. This clearly indicates that P. euphratica has developed an efficient mechanism to tolerate high salt levels. Accordingly, in vitro exposure of P. euphratica to salt showed that Na⁺ is preferentially accumulated in the cell wall, which thus protects the cytosol from Na⁺ toxicity 38. The carbon isotope ratio δ¹³C/¹²C revealed that the trees were likely not suffering from severe drought stress (Figure 2c). This may be due to the fact that P. euphratica trees produce a deep and extensive root system, giving the trees access to deep water even when they were far from the river (area C) 939.

Survival and growth of P. euphratica trees under the unfavorable conditions of high soil salinity and heat stress suggest that they are well adapted to the local desert conditions. Thus, sequencing of cDNA libraries in combination with DNA microarray analysis was used to elucidate the mechanisms underlying this adaptation. Construction of the cDNA libraries was based on maximizing the number of sequences relevant to stress tolerance and adaptation, in particular salt stress. To reach this goal, normalized libraries from Ein Avdat valley-grown adult trees and subtracted libraries from NaCl and several other abiotic stress-treated juvenile trees were constructed (Additional data file 1).

The EST collection obtained was compared to other poplar stress EST collections and salt DNA microarray experiments to estimate the amount of stress related genes obtained (Additional data file 5). Gu et al. 6 used a similar strategy and employed the suppression subtractive hybridization (SSH) method to identify genes upregulated by salt in P. euphratica. Of the 58 sequences reported, 43 (74%) were present in our EST collection. Christopher et al. 17 sequenced 928 ESTs from systemically wounded leaves of hybrid poplar (P. trichocarpa × P. deltoides) and performed a macroarray analysis on systemically wounded leaves. Of the genes listed as having probable or potential herbivory or wounding functions, 20 of 33 sequences (60%) were represented in our EST collection. In the macroarray analysis, they identified 18 genes with more than a tenfold increased transcript level in response to wounding; 12 of these 18 sequences (67%) were present in our EST collection. In our collection, the majority of highly wound-induced genes were obtained from the normalized control libraries, indicating that P. euphratica trees may constitutively express defense genes. Whether genes that could contribute to salt tolerance of P. euphratica also are constitutively expressed remains to be determined. Wounding- and poplar mosaic virus-regulated genes have also been identified in P. trichocarpa × P. deltoides using microarray analysis 25. Of the 150 genes induced twofold or more by virus and/or wounding, 113 (75%) were present in the P. euphratica EST collection (Additional data file 5), indicating that even though all cDNA libraries were constructed from trees exposed to abiotic stress, there were many ESTs related to biotic stress present in the P. euphratica EST collection.

As only three Populus array experiments relating to stress have been published, a comparison was also made to transcript profiling of salt stressed A. thaliana. Seki et al. 26 analyzed 7,000 genes in A. thaliana and found that 227 genes displayed more than fivefold increased transcript levels and 103 genes decreased levels in response to the NaCl treatment. A comparison of these genes with the P. euphratica EST collection showed that 81 (36%) of the Arabidopsis genes with increased transcript levels in response to salt stress have a direct match in our EST collection (Additional data file 5). Among the genes with reduced transcript levels, 56 (54%) were present in our EST collection. Not surprisingly, the P. euphratica genes with a match to the Na-regulated Arabidopsis genes were predominantly from the normalized Ein Avdat library and the subtracted libraries, whereas the down regulated genes mostly had matches in the control libraries. Overall, this comparison with known stress regulated genes indicates that a wide collection of stress related genes was obtained in the P. euphratica EST collection.

A uni-gene set of the P. euphratica EST collection was printed on microarrays, which were used to profile gene expression in leaves harvested from Ein Avdat areas A, B and C. Leaves from trees growing in the least saline soil (area P) were used as controls. In A. thaliana, typically 5% to 30% of the genes displayed changed levels of transcripts in response to various abiotic stress treatments 2640. In striking contrast, the number of P. euphratica genes that displayed different transcript levels expressed in the Ein Avdat experiment was only approximately 1% of the genes on the array. Furthermore, the trees in areas A, B and C had similar transcript profiles (Additional data file 7). For some genes, for example an early light-induced protein, there was a significantly higher expression in area C compared to the other areas (Additional data file 7). These genes with higher expression in area C might be important for adaptation to the highly stressful conditions on the dry, rocky slope (Figure 1). The majority of the genes with altered transcript levels belong to only a few functional categories: 'protein metabolism', 'response to abiotic or biotic stimulus', 'metabolism' and 'biological process unknown'. One reason for the relatively low number of genes with different transcript levels could be that the Ein Avdat trees have been growing in the valley for decades and are well acclimated to these growth conditions, and thus may have their defense mechanisms fully activated. This could also explain why so relatively few genes involved in signal transduction and no genes involved in control of transcription were identified as displaying different transcript levels. A systematic genotyping of the Ein Avdat trees with Simple Sequence Repeat (SSR) markers did not detect any polymorphism among the trees, suggesting that the P. euphratica trees are ramets of the same genotype, probably propagated through root suckers (I Beritognolo and R Muleo, personal communication; see also 3941). Thus, the trees at each area may be expected to respond similarly to stress conditions, which also might explain their almost identical transcript profiles (Additional data file 7).

Interestingly, some of the genes with increased transcript levels in areas A, B, and C (galactinol synthase, aldehyde dehydrogenase, β-amylase, ferritin and cysteine protease) have previously been shown to be upregulated in Arabidopsis during combined heat and drought stress 42. The galactinol synthase gene was only found in the Ein Avdat EST library, indicating that galactinol (or raffinose as indicated by the metabolite profiling; Table 4) might be used as a compatible solute during salt stress. Galactinol synthase is also one of the genes with increased transcript levels in controlled laboratory experiments with salt exposed P. euphratica 38. Abiotic stress causes the formation of reactive oxygen species that can react further with cellular components such as lipids and proteins. None of the 'classic' enzymes involved in antioxidant defense, including catalase, ascorbate peroxidase and superoxide dismutase, showed altered transcript levels. However, two other proteins, aldehyde dehydrogenase and metallothioneins, might fulfill a similar role in antioxidant defense. A result of oxidative damage is the formation of reactive and toxic aldehydes. In A. thaliana, an aldehyde dehydrogenase (AthALDH3) gene was induced by salt stress and other abiotic stresses 43. Transgenic plants overexpressing AthALDH3 displayed increased tolerance to salt and dehydration stress and accumulated less reactive aldehydes derived from lipid peroxidation 43. The increased transcript abundance of an aldehyde dehydrogenase in P. euphratica suggests that it could use this defense strategy to reduce damaging effects from oxidative stress. Metallothioneins are low molecular mass cysteine-rich proteins that can bind heavy metals and have a proposed role in detoxification of heavy metals. Expression of poplar metallothioneins in a cadmium sensitive yeast strain conferred cadmium tolerance 44. The transcript levels of poplar metallothioneins increase during senescence, heavy metal treatment and virus infection 132544. The presence of several cysteines in the metallothioneins suggests that they might be involved in the detoxification of reactive oxygen species or in maintenance of redox levels. Recently, recombinant rice and watermelon metallothinoneins have been shown to possess high superoxide and hydroxyl radical scavenging capacity 4546. Metallothionein ESTs were highly abundant in Ein Avdat and control libraries (Additional data file 2). Furthermore, the microarray analyses indicated that they were more highly expressed in trees growing in areas A, B and C than in trees growing in the less saline area P (Additional data file 7). This suggests that metallothioneins might play a role in detoxification of reactive oxygen species in stressed trees.

The genes with highest fold-induction in transcript levels in the Ein Avdat valley encode cysteine proteases. Cysteine proteases have previously been shown to be induced under both salt and drought stress 47. Their role could be to enhance degradation of stress-denatured proteins. Alternatively, elevated cysteine protease expression is frequently observed during senescence in Populus 1315. As the samples for array analysis were harvested late in autumn, it is possible that the trees in the valley (particularly in area C) might develop senescence faster than the trees at the parking lot, and this might lead to elevated cysteine protease expression.

Recently, the halophyte salt cress (T. halophila) has been established as a model for research on salt tolerance due to its high sequence identity with A. thaliana 4849. Comparative microarray experiments performed on A. thaliana and T. halophila with the same NaCl treatment revealed that the salt sensitive A. thaliana responded to the stress by modified expression of a relatively large number of genes, while the expression of only a few genes (six times less) changed in the tolerant T. halophila. Furthermore, direct hybridization of control RNA from both species on the same array showed that the tolerant species over-expressed several stress-related genes under non-stressed conditions 49. Thus, the salt tolerant species constitutively expresses defense genes and does not show further induction of genes upon stress treatment. Forty-nine percent of the stress related genes constitutively expressed in T. halophila were present in the P. euphratica EST collection (Additional data file 5). P. euphratica resembles T. halophila in several ways; both are tolerant to salt stress and display constitutive expression of genes related to, and induced by, biotic stress. Further support for the hypothesis that P. euphratica constitutively expresses defense genes and responds with few changes in gene expression following stress comes from our preliminary array experiments using salt stressed P. euphratica and P. tremula (which is salt sensitive); P. tremula displayed 10 times more genes with modified transcript levels than P. euphratica (B Vinocur, M Brosché, J Kangasjärvi, A Altman, unpublished data). To be fully confirmed, this hypothesis requires additional study, especially direct microarray hybridization comparisons between P. euphratica and stress sensitive poplars such as P. tremula or P × canescens.

To complement the transcript profiling we also analyzed the accumulation of 22 metabolites in the leaves collected from the experimental sites. Increased concentration of several amino acids such as β-alanine, valine and proline was observed, particularly in area A (Table 4). Rizhsky et al. 42 reported similar increases in different amino acids such as isoleucine, leucine, valine, β-alanine and tyrosine as a response to heat and drought stress, and proline, cysteine and serine have increased concentrations following cold stress 50. Among the amino acids analyzed, β-alanine had the highest accumulation in P. euphratica leaves from areas A and B. β-Alanine is the precursor of β-alanine-betaine, a quaternary ammonium compound that has similar osmoprotective function as glycine-betaine and accumulates in species belonging to the highly salt tolerant family Plumbaginaceae 51. Therefore, it is possible that amino acid accumulation supports increased production of metabolites that are part of a defense under very saline conditions.

The concentration of several metabolites such as fructose, glycerol, malate, maltose, mannose, galactinol, myo-inositol, putrescine, raffinose, sucrose, and trehalose increase dramatically in response to heat, drought and cold stress in Arabidopsis 4250. In P. euphratica, however, the changes in stress responsive carbohydrates and organic acids were of limited extent at the four different field sites. Considering the high accumulation of Na⁺ in field-grown leaves (Figure 2b), as well as in controlled salt experiments 38, it is possible that sodium itself can act as an osmolyte as in halophytes.

The Ein Avdat valley is located in the Negev desert in Israel at 450 m above sea level and a latitude of 30° 49' 30" N, and a longitude of 34° 46' 0" E. The average rainfall is 100 to 200 mm per year with most of the rain falling during winter time. Adult heart shaped leaves were collected from trees in areas A, B, and C for the cDNA library during the summer seasons of 2000 to 2001. Material was also harvested from young in vitro plantlets of a clone (clone B2) derived from a single tree from area B. Samples for microarray analysis were taken from areas A, B, and C and from the Ein Avdat parking lot (area P, 1 km from the valley) on 27 November 2003, and for ion, isotope and metabolite profiling on 27 November 2003 and 13 October 2004. The parking lot trees were irrigated with non saline tap water for 24 hours once a week.

For normalized control libraries, leaves, shoots and roots were harvested from 8 month old P. euphratica seedlings. Seeds were obtained from the Forest Administration Division, Xinjiang Corps of Production and Construction, Urumqi, PR China. The seeds were sown on sand that had a high content of clay (6%) and germinated at 12 hours light with 250 μmol m^-2 s^-1 photosynthetically active radiation, 25°C, and a relative air humidity of 60% in a climate chamber (Weiss, Giessen, Germany). The seedlings were potted into soil (Fruhstorfer Erde N, Archut, Germany), transferred to a greenhouse with natural light and watered daily with Long Ashton macronutrient solution 52.

For stress treatments, leaves and roots were harvested from P. euphratica subjected to the following treatments: elevated CO₂ (plants submitted to an atmosphere of 700 ppm for 60 days); different irradiance levels (plants under three different irradiance regimes corresponding to 48%, 18% and 8% external irradiance) for 60 days; drought stress (plants submitted to a 21 day cycle of drought with 5% to 3% volumetric soil humidity) before harvesting; flooding stress (plants submitted during 21 days to a complete submergence of the root system); ozone (plants submitted to 200 nl l^-1 of O₃ for 7 and 20 h); cold and freezing (leaves harvested from plants acclimated at 5 days of +2°C and from plants exposed to two and seven days of -2°C following the acclimation period); salt stress (leaves and xylogenic cambium harvested from plants exposed to 150 mM NaCl for 0.5 h, 1 h, 1.5 h, 2 h, 5 h, 10 h, 24 h and 48 h and roots harvested after 40 minutes, 1.5 h, 8 h and 20 h); cadmium stress (plants were stressed with 50 μM CdCl₂ and roots harvested after 12, 24 and 48 h).

Leaves were frozen in liquid nitrogen and freeze dried for two days. Dry powder samples (250 mg) were then extracted with 5 ml 65% HNO₃ at room temperature overnight to release the free ions and then digested for 2 h at 180°C. The supernatants were diluted in deionized water and ion analysis was carried out using an ICP-AES Spectroflame (Spectro Analytical Instruments, Kleve, Germany). The data were analyzed with the Tukey multiple comparison algorithm (p = 0.05). Soil samples were dried at 70°C and subsequently digested with a nitric acid pressure system according to Heinrichs et al. 53. Quantification of elements was carried out by ICP-OES (inductively coupled plasma-optical emission spectrometry; Spectro Analytical Instruments) at l = 559 nm. One way ANOVA tests were performed with the Tukey honestly significant difference (HSD) post hoc test. For isotope measurements freeze-dried leaves were finely ground using a mill (CB2200, CEP Industrie, Dept Sodemi, Saint-Ouen l'Aumône, France). ¹³C content in bulk leaf tissue was measured with an isotope mass spectrometer (Finnigan Mat; Delta S, Bremen, Germany). ¹³C content was expressed as the ¹³C/¹²C ratio relative to the Pee Dee Belemnite standard as a per mil deviation from this standard. The precision of the spectrometric analysis (standard deviation in δ) of the laboratory standard (ground leaves of Quercus ssp.) was 0.15‰ (n = 201).

RNA was isolated from P. euphratica leaves or roots using the method described in Chang et al. 54. Total RNA from root samples was DNase I treated before mRNA isolation. mRNA was isolated using oligo-tex beads (Qiagen, Hilden, Germany). Normalization of mRNA populations was performed as described in Sasaki et al. 55. Normalized cDNA libraries were constructed from mRNA using the SuperScript plasmid system with Gateway technology for cDNA synthesis and cloning (Invitrogen, Carlsbad, CA, USA). Subtracted libraries were constructed by two different methods. First by using the PCR select cDNA subtraction kit (BD Bioscience Clontech, Palo Alto, CA, USA; also called SSH 56), with the resulting PCR products cloned into pGEMT-Easy T-vector (Promega, Madison, WI, USA). The second subtraction method used restriction enzyme digestion with cap-selection. Tester cDNA was prepared by synthesizing cDNA from 500 ng mRNA with an oligo-dT adaptor primer (5' GACTAGTTCTAGATCGCGAGCGGCCGCCCTTTTTTTTTTTTTTTTVN3') in the presence of a 'cap select' primer (5'AAGCAGTGGTATCAACGCAGAGTGTCGACGGG3'), which binds to the end of a newly synthesized cDNA (see 57 for a description of the CapSelect technique). A driver was prepared by single-strand PCR amplification from control libraries and by double-stranded cDNA synthesis from control mRNA. The driver was then hybridized to the tester in a ratio of at least 50:1 for 24 to 48 h. After hybridization, double stranded molecules (representing common transcripts) were degraded by digestion with a panel of restriction enzymes (RsaI, NlaIII, BfaI, Sau3AI; New England Biolabs, Beverly, MA, USA). The remaining undigested single-stranded cDNAs were size selected and subsequently amplified by PCR using primers designed for the 5'cap and 3' poly-A tail (5'AAGCAGTGGTATCAACGCAGAGT3' and 5'GACTAGTTCTAGATCGCGAGC3'). The resulting PCR products were digested with NotI and SalI, size selected, and cloned into pSPORT1 (Invitrogen, Carlsbad, CA, USA). The subtracted libraries were subjected to a stringent quality control to determine how many clones from each library to sequence. Two parameters were tested, 'subtraction efficiency' and 'complexity'. Essentially, subtraction efficiency was judged by assessing the removal of abundant control cDNAs. Complexity was judged by determining the percentage of different cDNAs in the library. To test this, 96 clones were sequenced from each library. To pass 'subtraction efficiency', a maximum of one of the obtained sequences should correspond to any of the five most abundant cDNAs from the normalized libraries (rbcS, lhcb, ferredoxin, metallothionein, PSII 10 kDa protein). Furthermore, at least 70% of the sequences had to be sequences not already present from the normalized libraries. To pass 'complexity', the number of independent sequences had to be at least 60%. The following libraries passed the quality control: P8, P9, P10, P11, P15, P16 and P17.

EST sequencing was performed as described in Laitinen et al. 58. The sequences have GenBank accession numbers AJ767069 to AJ780913. All the sequences were quality checked before clustering and integration into a Sputnik EST database as described in 2358. Unigene sequences are assigned a unique id based on their 5'-most EST sequence. A prefix 'C_' denotes that the unigene represents a multi-member 'cluster' of sequences. Singleton sequences are prefixed with 'S_'. Peptide sequences were derived for all unigenes using the ESTScan application 59. Prior to the ESTScan predictions, a P. euphratica specific ESTScan model was created by training with open reading frames identified through BLASTX of the unigenes against the Swissprot database with results filtered using the expectation value of 1e-10. The peptide predictions were used to estimate the amount of protein coding sequence (CDS) within both individual EST and cluster consensus sequences. Sequences were annotated for homology using the BLASTN and BLASTX algorithms against a non-redundant protein sequence database, the Swissprot database, the A. thaliana and O. sativa genome databases, and sequence databases containing the aggregated openSputnik consensus sequences for Asterid, Eurosid, Caryophyllid, and monocot genomes. Sequences were additionally functionally characterized in context with the MIPS Funcat and Gene Ontology GO terms 28. GO terms were derived using two independent methods: one using the Swissprot to Gene Ontology mappings, the other using the InterPro to Gene Ontology mappings.

The ESTs were clustered using the CLOBB algorithm and based on the clusters, a uni-gene set were selected for reamplification as previously described 2958. All PCR-products were sequenced to verify correct identity. The Populus microarray contains 7,662 different populus ESTs (of which 491 ESTs were amplified twice and are duplicated on the array, for a total number of 8,153 populus ESTs; these ESTs correspond to 7,342 distinct unigenes derived from the clustering of the sequences (HPT2 and CAP3), and spiking and negative controls that were amplified from the Arabidopsis functional genomics consortium control set 60 and from additional Caenorhabditis elegans clones. The number of genes corresponding to the arrayed ESTs was estimated by anchoring ESTs to the available Poplar genome scaffold using the BLASTN method, with results filtered at 1e-20. Sequences were merged if they overlapped or appeared within the same 5 kbp interval on the assembled genome scaffold.

For printing, 5 μL of purified fragments were transferred to a 384 printing plate together with 5 μL of 6 × SSC solution. The ESTs were printed onto poly-lysine coated slides with a Genemachines OmniGrid 100 (Genomic Solutions, Ann Arbor, MI, USA) using 16 SMP3 pins (TeleChem International, Sunnyvale, CA, USA). Printing conditions were 48% to 50% humidity and 21°C.

Three biological repeats were used from each of the areas (A, B and C; Figure 1) in the Ein Avdat valley; each of the independent biological repeats contain leaves from two or three trees. The parking lot (area P) control sample was a pool of leaves from seven trees. To avoid bias in the microarray evaluation as a consequence of dye-related differences in labeling efficiency and/or differences in recording fluorescence signals, dye labeling for each paired sample (Area/Parking lot control) was reversed in two individual hybridizations. Thus, a total of six hybridizations per area were obtained. The complete protocols for probe labeling and hybridization, normalized data and raw data files are available from the ArrayExpress database 61 under the accession E-MEXP-182.

Images were analyzed in GenePixPro 5.1 (Axon Instruments, Union City, CA, USA). Visually bad spots or areas on the array and low intensity spots were excluded. Low intensity spots were determined as spots where less than 55% of the pixels had intensity above the background + 1SD in either channel. Using these criteria, 31% to 58% of the spots gave a positive signal from the 18 hybridizations. The data from GenePixPro 5.1 were imported into GeneSpring 7 (Silicon Genetics, Redwood City, CA, USA) and normalized with the Lowess method. The background subtracted median intensities were used for calculations. Up- and downregulated genes were selected using two criteria: the gene should consistently have twofold increased or decreased transcript levels in all biological repeats, there should be a signal from the spot from at least four of six hybridizations for each area, and the gene induction or repression should be statistically significantly different from a ratio of 1.0 determined with Students t test in GeneSpring. Gene expression in areas A, B, and C was compared using Welch ANOVA assuming that groups have unequal variances. After Benjamini and Hochberg false discovery rate correction for multiple testing, a false discovery rate of 0.05 or less was considered statistically significant. Tukey's post-hoc test was applied to identify the areas that differ from each other.

The microarray results were verified with qPCR. Reverse transcription was performed with 6 μg of DNase I treated total RNA with SuperScript III according to the manufacturer's instructions (Invitrogen). The reverse transcription reaction was diluted to a final volume of 100 μl, and 1 μl was used as template for the PCR using SYBRgreen PCR master mix (Applied Biosystems, Foster City, CA). PCR was performed in duplicate using ABI Prism 7000 default cycling conditions (Applied Biosystems). The primers used for PCR are listed in Additional data file 6. The raw threshold cycle (Ct) values were normalized against glucosidase II alpha subunit (shown to have a constant expression in all experiments performed on the P. euphratica DNA microarray) to obtain normalized ΔCt values that were then used to calculate the difference in expression levels between areas A, B and C and the parking lot control.

The extraction protocol was modified from Roessner-Tunali et al. 62 and Schauer et al. 63. Briefly, frozen leaf tissue powder (100 mg dry weight) was extracted in 1.4 ml of 80% (v/v) aqueous methanol in a 2.0 ml microcentrifuge tube. Ribitol (120 μl of 0.2 mg ml^-1 water) was added as an internal standard prior to incubation. The mixture was extracted for 15 minutes at 70°C under 200 rpm. The extract was vigorously mixed with 1,500 μl water and 750 μl chloroform andsubsequently centrifuged 15 minutes at 4,000 rpm. Aliquots of the methanol/water supernatant (150 μl) were dried in vacuum overnight. The dry residue was modified for GC-MS analysis according to Schauer et al. 63. Residues after reduction were redissolved and derivatized for 90 minutes at 37°C in 60 μl of 30 mg ml^-1 methoxyamine hydrochloride in pyridine followed by a 30 minute treatment with 120 μl of N-methyl-N-[trimethylsilyl]trifluoroacetamide at 37°C. A retention time standard mixture (10 μl of 0.029% (v/v) n-dodecane, n-pentadecane, n-nonadecane, n-docosane, n-octacosane, n-dotracontane, and n-hexatriacontane dissolved in pyridin) was added before trimethylsilylation. Sample volumes of 1 μl were then injected onto the GC column on a splitless mode.

The GC-MS system was run as previously described 63 with some modifications. It was composed of a Pal autosampler (Ophir analytic Tel Aviv, Israel, representative of CTC analytic, Zwingen, Switzerland), a TRACE GC 2000 gas chromatograph, and a TRACE DSQ quadrupole mass spectrometer (Restek, Bargal Analytics, Tel Aviv, Israel, representative of ThermoFinnigan, Hemel Hempstead, UK). GC was performed on a 30-m Rtx_5Sil MS column with 0.25 μm film thickness (ThermoFinnigan). The injection temperature was set at 297°C, the interface at 280°C, and the ion source adjusted to 200°C. Helium was used as the carrier gas at a flow rate of 1 ml min^-1. The analysis was performed under the following temperature program: 5 minutes of isothermal heating at 70°C, followed by a 5°C per minute oven temperature ramp to 350°C, and a final 5 minute heating at 330°C. Mass spectra were recorded at 2 scan sec^-1 with a scanning range of 40 to 600 m/z. Both chromatograms and mass spectra were evaluated using the XCALIBUR v1.3 program (ThermoFinnigan). A retention time and mass spectral library for automatic peak quantification of metabolite derivatives were implemented within the NIST 2.0 method format. Substances were identified by comparison with authentic standards, as described in Roessner-Tunali et al. 62.

The levels of the compounds were calculated as the relative response ratio of peak areas of different compounds related to the peak area of ribitol (which served as an internal standard), and normalized with respect to the dry weight of the sample (according to Nikiforova et al. 64). Data are presented as the average of six separate measurements and analyzed statistically with the Tukey multiple comparison algorithm (p = 0.05) incorporated into JMP statistical software (JMP, Statistics and Graphics Guide: Version 3, SAS Institute Inc. Cary, NC, USA).