Plantlets from P. × canescens (P. alba × tremula) clone INRA717 1-B4 and P. euphratica clone B2 from the Ein Avdat valley in Israel were multiplied by in vitro micropropagation , transferred to aerated hydroponics using Long Ashton nutrient solution  for acclimation to greenhouse conditions, subsequently potted into soil (Frühstorfer Erde, Lauterbach-Wallenrod, Germany) and grown for 3 months in a greenhouse at 20 to 25°C, with a relative air humidity of 40 to 60% and additional 16 h of light (neon lamps: L58W/25 and 58W/840, Osram, Munich, Germany, and TLD 58W/840 Philips, Amsterdam, Netherlands) between 6:00 a.m. and 10:00 p.m. to reach a continuous photosynthetic active radiation (PAR) of 150 μmol m-2s-1. Fully expanded leaves were harvested at a stem height of approximately 70 cm (upper third of the stem) after four h in the light period. The leaves were shock-frozen in liquid nitrogen and kept at -80°C until further analysis.
Gas exchange (net photosynthesis, respiration, and transpiration) was measured (1 min means) with a portable gas exchange system (HCM-1000 or GSF-3000, Walz, Effeltrich, Germany) under ambient conditions. For analysis of dark respiration plants were connected to the system and kept for 30 min in darkness prior to analysis. For each plant (n = 5) the gas exchange values of 5 min were averaged.
Of each species, 5 biological replicates were measured. Frozen leaf tissue was ground in a ball mill (Retsch, Haan, Germany). Fine powder was diluted in methanol to a methanolic concentration of 70% to give highest ion density inside the electrospray, without eliminating those neutrals which are highly water soluble. High-resolution mass spectra were acquired on a Fourier Transform Ion Cyclotron Resonance Mass Spectrometer (APEX Qe, Bruker, Bremen, Germany) equipped with a 12-T superconducting magnet and an Apollo II Electrospray (ESI) source. The ionization source was run in the negative operation mode to generate mono charged negative ions. Thus, all important multi-functional organic compounds which bear at least one acidic site could be deprotonated to give rise to a detectable anion in the mass spectrometer . Each sample was introduced into the ionization source at a flow rate of 2 μL min-1 by a microliter pump with a nebulizer gas pressure of 20 psi and a drying gas pressure of 15 psi (heated to 200°C). Each sample was measured three times and the signal intensities of each detected ion were averaged before running the multi-dimensional statistical data analysis.
Spectra were externally calibrated on clusters of arginine (10 mg l-1 in methanol); calibration errors in the relevant mass range were always below 0.05 ppmv. The spectra were acquired with a time domain of 1 Megaword with a mass-to-charge ratio (m/z) range of 146-2000. The spectra were zero filled to a processing size of 2 Megawords. Before Fourier transformation of the time-domain transient, a sine apodization was performed. No fragmentation experiments were performed in this study. Thus, the whole mass range could be scanned, and 300 scans were summed on in each acquisition. The ion accumulation time in the ion source was set to 0.1 s.
FT-ICR mass spectra were exported to peak lists at a signal to noise ratio (S/N) of two. From those lists, High to Low signal intensity and Low to High signal intensity profiles were obtained by use of the Hierarchical Clustering Explorer HCE Version 3.0 . Ions with sharp differences between the acquisitions of the two species were highlighted as High to Low and Low to High signal intensity profiles. The m/z values of all ions were clustered by the use of the average linkage method with similarity/difference measure type: Euclidean distance. The search method is model-based and the used distance measure was Pearson's correlation coefficient of 0.8. For all obtained hits, t-tests were calculated to validate a statistical significance at α = 0.1.
Metabolite identification and annotation of the m/z values was performed via the MassTRIX web site with the following parameter settings: 'Scan mode negative ionisation (correct for H+ loss)', 'Max. error 1.0 ppm', 'Database KEGG with isotopes', 'Organism Populus trichocarpa' . Identified compounds and differentially expressed genes added to the query as EC numbers were automatically mapped to KEGG pathways using KEGG/API via MassTrix. Metabolites with the same total formula could not be distinguished.
Analysis of soluble and cell wall bound phenolics
Frozen leaf tissue was ground in a ball mill (Retsch, Haan, Germany). Fine powder (60 mg per sample) was extracted with 2 ml of 50% methanol in an ultrasonic bath (60 min, 40°C; Sonorex Super RK 510 H, Bandelin electronics, Berlin, Germany). The extract was centrifuged, the pellet was extracted once again and the supernatants were combined for analysis of soluble phenolics photometrically with the Folin Ciocalteus method . Catechin (Sigma-Aldrich, Deisenhofen, Germany) was used for calibration and the phenolic concentrations were expressed as catechin equivalents.
To determine cell wall bound phenolics, the pellet was washed with n-hexane, dried, and weighed. This pellet representing cell walls was extracted with 1 M NaOH as described earlier . The extracted phenolic compounds were measured with the Folin Ciocalteous reagent as above.
Analysis of soluble and non-soluble condensed tannins
Condensed tannins (proanthocyanidin) were determined after . For extraction of soluble tannins, plant materials ground to a fine powder were extracted with 50% methanol, centrifuged, the pellet was re-extracted and the extracted solutions were combined. Aliquots were mixed with a reagent (1.6% methanol, acid butanol [5% concentrated HCl in butanol (v/v)] and 0.2 ml of 2% FeNH4(SO4)2·12H2O in 2 M HCl [FeNH4(SO4)2·12H2O/HCl = w/v], incubated at 95°C and used to measure the extinction at 550 nm. The amount of tannins was expressed as absorbance per dry mass.
To determine non-soluble tannins, the pellet was lyophilized, weighed and resuspended in 1 ml of 100% methanol. The slurry was combined with acid butanol reagent and processed as above.
Frozen tissues were ground to a fine powder with a ball mill (Retsch, Haan, Germany) pre-cooled in liquid nitrogen and extracted in DMSO/HCl (dimethylsulfoxid:25% HCl = 80:20 (v:v)) at 60°C for 30 min. The mixture was centrifuged and aliquots of the supernatant were mixed with 0.2 M citrate buffer (pH 10.6). This extract was mixed with a reaction mixture (0.75 M triethanolamine-buffer, 4 mM NADP, 10 mM ATP, 9 mM MgSO4, pH 7.6) for enzymatic determination of soluble sugars as described previously . Glucose was measured after addition of hexokinase (3 mg ml-1, Roche Diognostics GmbH, Mannheim, Germany), fructose after addition of phosphoglucose isomerase (10 mg ml-1, Roche Diognostics GmbH, Mannheim, Germany) and sucrose after incubation with β-fructosidase (5 mg ml-1, Roche Diognostics GmbH, Mannheim, Germany). Calibration curves were produced with glucose, fructose and sucrose. The sucrose content of the extracts was below the detection limit of the method.
In silico sequence analysis
DNA sequences of twenty genes were compared between P. × canescens, P. euphratica and P. trichocarpa. For five genes, full length sequences were available for P. euphratica and P. × canescens cDNA ('IAA-amino acid hydrolase', Ill3 [ GenBank:AJ744952, GenBank:AJ744953]; 'Na/H antiporter type D', NhaD1 [ GenBank:AJ561195]; 'salt induced serine-rich' and 'temperature induced lipocalin', SIS and TIL [ GenBank:FJ238515, GenBank:FJ238514 and GenBank:FJ238513, GenBank:FJ238512]; 'gibberellin regulated protein', GAST, [GenBank:FJ238511, GenBank:CF231013]). Criterion for the selection of further genes was their representation by probe sets on the microarray and the availability of sequence data for P. euphratica and P. × canescens in public databases. Probe sets both with and without significantly different signal intensities between P. euphratica and P. × canescens were randomly chosen, and sequence data for the corresponding P. trichocarpa gene model retrieved from the Populus genome project (Table 1) . Because most sequence data is available from EST datasets and thus as cDNA sequences, open reading frame (ORF) sequences of the P. trichocarpa gene models were used to screen the GenBank database for homologs in P. euphratica and P. × canescens by using the Megablast algorithm . Pair wise alignments and base identity calculations of the cDNA sequences were generated with the GeneDoc software . An ANOVA was conducted to test for significant differences between the sequence identity values at α = 0.05.
Leaf material of 3 plants was pooled. Three biological replicates per poplar species were analyzed (i.e. a total of 9 nine plants per species was included). Frozen leaf tissue was milled in a ball mill (Retsch, Haan, Germany). Total RNA was extracted from 500 mg of frozen plant powder according to Chang et al.  with minor modifications: No spermidine was applied in the extraction buffer, and 2% β-mercaptoethanol was used. RNA was additionally purified using an RNeasy Mini Kit (Qiagen, Valencia, CA). Total RNA yield and purity were determined spectrophotometrically (BioPhotometer, Eppendorf, Hamburg, Germany) at A260 and A280. RNA integrity was assessed on an Agilent 2100 Bioanalyzer (Agilent, Santa Clara, CA) at the Microarray Facility (Tübingen, Germany).
Of each species, 3 biological replicates were analyzed on the GeneChip® Poplar Genome Array (Affymetrix, Santa Clara, CA). Synthesis of one-cycle cDNA and biotin-labelled cRNA, fragmenting of cRNA, hybridization to the Poplar Genome Array, washing, staining and scanning was performed as stated by Affymetrix (GeneChip® Expression Analysis Technical Manual) at the Microarray Facility (Tübingen, Germany). Data were deposited at ArrayExpress [EMBL: E-MEXP-1928].
Statistical analysis of the raw signal intensity data was conducted using the following functions from packages released by the bioconductor project, implemented in R [78, 79]. Background correction, quantile normalization and summarization of the Affymetrix CEL output files resulting in a raw list of normalized genes was computed by robust multi-array average (RMA) using the 'rma' function from the 'affy' package . Statistical testing for differentially expressed genes was performed on this raw list with the 'sam' function from the 'siggenes' package . In the SAM analysis, Δ was chosen to obtain FDR = 0.05. Annotation of the genes was carried out via the PopArray-Database . Gene Ontology (GO) terms were matched via The Arabidopsis Information Resource (TAIR) .
Gene lists were subjected to several filtering processes after : genes that were not expressed were removed by calculating Affymetrix' MAS 5.0 change calls using the 'mas5calls' function from the 'affy' package; only genes with two or three 'present' calls in any one of the two species were considered. Probe sets presumably affected by differences in binding efficiencies were removed. For this purpose, SDs of prey signals were calculated for each bait set on a microarray and then averaged over all six microarrays for each gene. The resulting data were compared with the range of the SDs of all prey set signals. A gene was removed from the list when its prey signal SD was in the upper 5% quantile of all 'present' bait set SDs on the microarray. The setting of the threshold for this filter affects the stringency of the results. For the purpose of this study the upper 5%-quantile was used.
Probe sets for which no annotation was available were removed. In cases where multiple probe sets were matched with the same JGI P. trichocarpa gene model, duplicate probe sets were removed so that each gene model was represented only once. The probe set with the lower p-value in the SAM analysis was used for further analysis if both genes showed the same direction of change in gene expression. Probe sets annotated by the same gene model with contrasting expression ratios were removed.
For evaluation of over- and underrepresented GO slim terms, the frequencies of annotations within list of significant genes were compared to the frequencies within the defined 'gene universe'. Statistical testing was conducted by calculating the cumulative hypergeometric distribution function, using the 'phyper' function in R. Benjamini-Hochberg correction was applied to the resulting p-values using the 'p.adjust' function.
For statistical analysis of overrepresentation with 'The Ontologizer' , a gene ontology file for Populus was adapted. In this file, all JGI Populus trichocarpa gene models represented on the Affymetrix GeneChip were linked to the GO identifier of their closest Arabidopsis matches as listed by TAIR . The 'population' of the Ontologizer conforms with our 'gene universe', whereas the two lists of genes higher expressed in one of the poplar species each correspond to a 'study set'. As parameter settings, term-for-term analysis with Benjamini-Hochberg correction was used.
For quantitative Real-Time PCR (qRT-PCR), primer pairs were designed for actin 9 [GenBank:AJ778775 (P. euphratica) and GenBank:CX656348 (P. × canescens)] as a reference gene and 10 transcripts that were differentially expressed in P. euphratica and P. × canescens (Additional file 1, Table S4). All primers met the following conditions: 1) 100% sequence identity with both P. euphratica and P. × canescens cDNA; 2) fragment lengths between 90 and 180 base pairs (bp); 3) calculated salt-adjusted melting temperatures Tm(salt) between 58 and 62°C. Primer design was performed with the Oligo Explorer, and suitable primers were tested for similar Tm(salt), primer dimers and primer loops by Oligo Analyzer (both Gene Link, Hawthorne, NY, USA).
Total RNA was DNAse treated with a Turbo DNA-free kit (Ambion, Austin, TX) and transcribed to cDNA with a RevertAid™ First Strand cDNA Synthesis Kit (MBI Fermentas, St. Leon-Rot, Germany). For each gene, three technical repeats were analyzed for three biological samples. The qRT-PCR was performed on an iCycler (Bio-Rad, Hercules, CA). 250 ng cDNA were used in a 25 μl reaction with 1 × ABsolute qPCR SYBR Green Fluorescein Mix (ABgene, Surrey, UK; including Thermo-Start DNA Polymerase) and 10 μM primer. First denaturation and activation of the Taq-polymerase occurred at 95°C for 15 min, followed by 45 cycles of denaturation at 95°C for 15 s, annealing at 56°C for 30 s and elongation at 72°C for 30 s.
qRT-PCR outputs of three biological replicates per species and gene were analyzed using the MyiQ software (Bio-Rad, Hercules, CA). Primer specificity was assessed by melting curve analysis. Statistical analysis of expression ratios and standard error was conducted using a pair wise fixed reallocation randomization test implemented in Excel with the Relative Expression Software Tool (REST)-384 [84, 85]. All primer pairs showed less than 5% differences in PCR efficiency between P. euphratica and P. × canescens.
Data are shown as means (± SE). Five to six plants were analyzed per species. Statistical analysis (ANOVA) was performed using Statgraphics (Centurion XV, St Louis, MO). Significant differences at P ≤ 0.05 were calculated by ANOVA followed by a Tukey-Kramer test .