SuperSAGE: the drought stress-responsive transcriptome of chickpea roots

Background Drought is the major constraint to increase yield in chickpea (Cicer arietinum). Improving drought tolerance is therefore of outmost importance for breeding. However, the complexity of the trait allowed only marginal progress. A solution to the current stagnation is expected from innovative molecular tools such as transcriptome analyses providing insight into stress-related gene activity, which combined with molecular markers and expression (e)QTL mapping, may accelerate knowledge-based breeding. SuperSAGE, an improved version of the serial analysis of gene expression (SAGE) technique, generating genome-wide, high-quality transcription profiles from any eukaryote, has been employed in the present study. The method produces 26 bp long fragments (26 bp tags) from defined positions in cDNAs, providing sufficient sequence information to unambiguously characterize the mRNAs. Further, SuperSAGE tags may be immediately used to produce microarrays and probes for real-time-PCR, thereby overcoming the lack of genomic tools in non-model organisms. Results We applied SuperSAGE to the analysis of gene expression in chickpea roots in response to drought. To this end, we sequenced 80,238 26 bp tags representing 17,493 unique transcripts (UniTags) from drought-stressed and non-stressed control roots. A total of 7,532 (43%) UniTags were more than 2.7-fold differentially expressed, and 880 (5.0%) were regulated more than 8-fold upon stress. Their large size enabled the unambiguous annotation of 3,858 (22%) UniTags to genes or proteins in public data bases and thus to stress-response processes. We designed a microarray carrying 3,000 of these 26 bp tags. The chip data confirmed 79% of the tag-based results, whereas RT-PCR confirmed the SuperSAGE data in all cases. Conclusion This study represents the most comprehensive analysis of the drought-response transcriptome of chickpea available to date. It demonstrates that – inter alias – signal transduction, transcription regulation, osmolyte accumulation, and ROS scavenging undergo strong transcriptional remodelling in chickpea roots already 6 h after drought stress. Certain transcript isoforms characterizing these processes are potential targets for breeding for drought tolerance. We demonstrate that these can be easily accessed by micro-arrays and RT-PCR assays readily produced downstream of SuperSAGE. Our study proves that SuperSAGE owns potential for molecular breeding also in non-model crops.


Background
Chickpea (Cicer arietinum L.) is one of the most important grain legume crops worldwide and a major source of protein for millions of families in developing countries. Despite considerable investment in breeding, average chickpea yield in major producer countries such as India stagnates at 0.6-0.7 Mt hectare -1 since many years. This low yield is far below the crop's potential of 3-5 Mt hectar -1 under optimal conditions. Chickpea is mostly grown in low-input, rain-fed agriculture in Mediterranean-type environments as inter-crop between cereals, and on residual moisture from monsoon rains on the Indian subcontinent. Due to insufficient, untimely and erratic rainfall in these semi-arid and arid areas, the crop often suffers from drought at the end of the cropping season. In future, global warming and soil erosion will even worsen conditions for many crops including chickpea. Thus, drought represents the major constraint to increase chickpea yield, and drought tolerance therefore is a major aim of chickpea breeding. However, drought tolerance is a complex trait and hard to achieve by conventional breeding methods. Understanding of the molecular mechanisms underlying drought tolerance is therefore needed for successful, knowledge-based crop improvement [1].
Molecular genetics and genomics of stress-responses in model plants such as Arabidopsis revealed that abiotic stresses such as drought, salinity and cold stress are characterized by ionic-and osmotic-disequilibrium components; eliciting general as well as specific responses and mechanisms of stress-protection [2]. These studies underpinned the importance of early responses to the various stresses for the survival of the plants [3]. Much of our current understanding of stress-response mechanisms comes from genome-wide analysis of gene expression, facilitated by the availability of microarrays carrying a comprehensive set of genes.
In chickpea, like in many other under-researched crops, extensive microarray-based studies are not yet possible, because the necessary EST data are not available. For example, no more than 7,580 chickpea ESTs are publicly available at the National Centre of Biotechnology Information (NCBI, http://www.ncbi.nlm.nih.gov/), as compared to at least 1,463,500 ESTs available for Arabidopsis. In the absence of such resources, researchers often use less comprehensive approaches as e.g. suppression subtractive hybridisation (SSH) libraries [4], which do not per se allow the quantification of expression of differentially expressed genes. SSH results are therefore often used for the generation of macro-arrays for subsequent analysis of gene expression [5]. These authors used this cumbersome approach for the discovery of 101 dehydration-responsive transcripts in chickpea roots.
Open-architecture, whole-genome transcription profiling technologies such as SuperSAGE [6], however, provide a solution to the problem of lacking EST and genomic data. SuperSAGE is an improved version of the Serial Analysis of Gene Expression (SAGE) technique [7]. In the past years, it has demonstrated a high versatility due to its longer tag size (26 bp) [8,9]. In principle, SAGE and all its variants rely on the assumption that a small, defined part of a cDNA, a so-called "tag", characterizes this cDNA, and that counting the number of times a particular tag occurs in the tag population faithfully reflects the abundance of the respective mRNA in the transcriptome. Since 10.000 to 100.000 tags are sequenced in a single experiment, a comprehensive profile of the transcriptome is generated.
Here we report on the stress responses of 80,238 transcripts representing 17,493 unique 26 bp tags (UniTags) from roots of the drought-tolerant chickpea variety ICC588 early after onset of desiccation. We discuss the stress-regulated transcription of genes involved in signal perception and transduction, ROS scavenging and metabolism, osmotic and ionic stress-related pathways, regulation of water and ion homeostasis, as well as several reported effector proteins. To test the reliability of the present results we use microarrays carrying stress-responsive as well as constitutively expressed 26 bp tag sequences. To further confirm the SuperSAGE results with a third method, we use SYBRgreen and commercially available TaqMan assays produced from 3'-and 5'-RACE sequences from selected chickpea mRNAs. Finally, we compare our transcription profiles from drought-stressed chickpea roots to results obtained from chickpea root SSH libraries [5] and microarray experiments in Medicago truncatula [10], and discuss similarities and differences. This study is the first of a series characterizing stress responses of chickpea on a molecular level as a prerequisite for production of expression markers and microrrays for highthroughput germplasm and expression (e)QTL analysis at the onset of knowledge-based breeding for stress-tolerance in this important protein crop.

The combination of high-throughput 454 sequencing with SuperSAGE
Drought libraries are part of a project, which aims at evaluating the transcriptional responses of chickpea upon diverse abiotic stresses, including several other treatments and various tissues (e.g. salt-stressed roots and coldstressed leaves; data not shown). For the sequencing of all the libraries, a single 454 plate divided into two sections was used, from which a total of 380,000 reads were extracted. After eliminating: i) incomplete reads, ii) twinditags, and iii) ditags without complete library-identification DNA linkers, a total of 330,000 26 bp tags were obtained for further analysis. From these, about 50,000 tags were singletons, that were excluded from analysis. Finally, 280,000 tags remained for evaluation (data not shown). Subsequently, sub-datasets were constructed for each experimental situation. For the present study, 82,238 26 bp tags from control and dehydrated roots were analyzed.

Abundance of UniTags and annotation to public databases
A total of 82,238 26 bp tags from roots of the drought-tolerant variety ICC588 either subjected to 6 h desiccation (53,141) or from well-watered controls (28,897) were sequenced, and represented 17,493 unique transcripts, so called UniTags. Less than 1% percent of these occurred in very high copy numbers (> 5,000 counts.million -1 ), whereas 23% and 75% of the transcripts were present between 100 to 1,000 and less than 100 copies.million -1 , respectively (Table 1). UniTags from control and stress libraries were deposited in the Gene Expression Omnibus (GEO) public domain under accessions GSM321783 and GSM321790, respectively.
Annotation of the 17,493 UniTags matched 3,858 (22.0%) to well characterized sequences from the Fabaceae family available in public databases. Of these, 53% matched to sequences from Medicago truncatula, 29% to Cicer arietinum, 6% to Pisum sativum, 3% to Glycine sp., 2% to Medicago sativa, and 7% to other legume genera. In many cases, TIGR gene index annotations from legumes bridge automatically to characterized Tentative Consensus (TC) sequences mostly from Arabidopsis, rice, and maize. Of the remaining 13,635 (78.0%) non-assignable 26 bp tags, 5,685 were significantly homologous to anonymous EST or DNA sequences, whereas 7,950 found no match at all. A summary of the primary data is given in

Annotation of virtual tags generated from chickpea ESTs deposited in public domains
In order to test the validity of the annotation of chickpea 26 pb tags through sequence homology with other legumes, virtual tags generated from chickpea EST sequences deposited in the NCBI data bank were extracted, and their direct annotation was compared with the annotation of longer homologous ESTs from the model legume M. truncatula. After retrieving the complete set of chickpea sequences deposited in the NCBI EST database (7,500 sequences), a total of 3,544 different in silico-generated 26 bp tags were selected to be directly BLASTed against the nr NCBI (Fabaceae) nucleotide database ( Table 2). From these, a total of 998 tags revealed high homology hits. After exclusion of anonymous entries, 253 tags were linked to Uniprot entries or to characterized non-protein coding RNAs (Table 3). In parallel, the same 3,544 tags were BLASTed against the plant EST NCBI (M. truncatula) and TIGR (M. truncatula) databases separately, where 1,143 and 680 sequences, respectively, found at least one high homology hit. Then, the complete target sequences from each BLAST were retrieved and re-blasted against the nr NCBI (Fabaceae) database. A total of 632 (NCBI M. truncatula ESTs) and 630 (M. truncatula TIGR ESTs) sequences, each representing a different 26 bp tag, revealed high homologies with nr NCBI entries. From these, 369 and 213, respectively, were non anonymous ( Table 3). The results from the direct BLAST were compared with each of the M. truncatula EST-briged homology searches for commonly annotated sequences, either by  Table 3, see also Additional file 2]. The present results reveal that in more than 90% of the cases, the assignment of tags to Uniprot-linked accessions is congruent with the annotation of longer ESTs from other legumes. However, databases overloaded with anonymous entries considerably reduce the amount of biologically interpretable data.

Differential gene expression in response to drought stress of chickpea roots and assignment of 26 bp tags to Gene Ontology (GO:) functional categories
We calculated the natural logarithm of expression ratios [here denoted as R (ln) ] of the 26 bp tags from control versus stressed roots as well as significance levels (P) according to Audic and Claverie [11] of up-and down-regulation for each transcript using the software package Discov-erySpace 4.01 (Canada's Michael Smith Genome Sciences Centre). Differences in abundance of tags in control and stressed roots were considered relevant at R (ln) > 1 (> 2.7fold change). At this threshold, 7,532 (43%) tags were significantly differentially expressed in stressed as compared to control roots. Of these, 2,557 were up-, and 4,975 down-regulated. A total of 880 transcripts (5.0%) showed more than 8-fold difference in expression (R (ln) > 2.0, P < 0.05). Of these, 291 were up-, and 589 down-regulated under stress. As depicted in the Venn diagram ( Figure 1), a considerable number of tags occurred exclusively in either the control tissue or under stress.
Correlation of R (ln) to defined standard functional gene categories (i.e. biological processes) in the Gene Ontology (GO:) database revealed that the majority of the most upregulated transcripts are assigned to the GO: biological process "Metabolism" (with the exception of extensin [O65760_CICAR], and S-receptor kinase-like protein [Q70I30_LOTJA] transcripts). As depicted in Figure 2, Stress Perception and Signalling (i.e. Intracellular Signalling Cascades, P = 0.997), Small GTPase-mediated Signal Transduction (P = 0.994), RNA Metabolism (P = 0.989), and Cellular Carbohydrate Metabolism (P = 0.989) were the most represented GO: functional categories in desiccation-stressed roots. Further, over-representation of transcripts involved in Transport (P = 0.943), Proteolysis (P = 0.926), Oxidative Phosphorylation (P = 0.886), and Stress Response (P = 0.878) indicate the mechanisms by which the roots adapt to the stress. Over-representation of tran-Venn diagram showing down-regulated, constitutively expressed, and up-regulated chickpea 26 bp tags Figure 1 Venn diagram showing down-regulated, constitutively expressed, and up-regulated chickpea 26 bp tags. Venn diagram of the quantitative tags classification into down-regulated, constitutively expressed, and up-regulated transcripts in control and drought-stressed chickpea roots. * Tags not differentially expressed, including tags found in either the control or the drought-stress library, respectively (at low statistical significance). Several differentially expressed tags are observed in both libraries.
scripts involved in GO: functional category "Oxygen and Reactive Oxygen Species Metabolism" (P = 0.808) suggests that ROS play an important role as side stress, but also as signalling molecules.
Members of a gene family assigned to over-or under-represented GO: categories, respectively, could be up-as well as down-regulated to different extents. For example, whereas UniTagSTCa-6374, annotated to the putative extensin (Q9FSY9_CICAR), was one of the most downregulated transcripts, UniTag STCa-19021 representing another extensin isoform (O65760_CICAR) was the most Up-regulated tag in our data set. However, both isoforms are grouped in the same GO:category.

Differential expression of drought stress-related subtranscriptomes in chickpea roots
Since the present genome-wide expression analysis revealed a plethora of differentially expressed 26 bp tags with and without match to genes of known function, it is impossible to display or discuss all of them in the frame of this paper. Instead, in Figure 5 and Figure 6 we elaborate in more detail on the expression of genes and gene families belonging to the GO: biological processes "Signal Transduction", "Stress Sensing", "Regulation of Transcription", "Transport", "Post-transcriptional Regulation" and "Pathway Inhibitors", all involved in early responses to stress. In these categories, the gene itself or members of its family have known functions in stress-perception, stresssignalling and stress-responsive regulation of transcrip- tion and chromatin structure. Further, we closely look at transcripts encoding proteins involved in Reactive Oxygen Species (ROS) scavenging, and transcripts related to ROSmediated signal transduction cascades ( Figure 6). As examples for the regulation of effector genes down-stream of the signalling cascades, we will detail the differential expression of genes coding for water-channel proteins, so called aquaporins, and of genes coding for proteins involved in osmolyte metabolism ( Figure 5, Figure 6).

Confirmation of SuperSAGE results by qRT-PCR
Our genome-wide expression analysis results were exemplarily confirmed by quantitative real-time (qRT) PCR using TaqMan probes XPTm-Ca-22356 (O65741_CICAR, mRNA for putative transmembrane channel protein) and XPTm-Ca-7975 (anonymous drought-induced EST) available from GenXPro GmbH, as well as SYBR Green ® assays.
Oligonucleotides for SYBR Green assays were deduced

Profiles confirmation via microarray hybridization of spotted SuperSAGE-derived oligos
To characterize the chickpea transcriptome under drought stress, sequence information from SuperSAGE profiles was used to design an Agilent 16 K microarray, onto which 3,000 selected 26 pb tags were spotted for a comparison of both profiling techniques. On the microarray, the majority of the oligonucleotides with original tag sequences were spotted twice (twin-replicas). Additionally, oligonucleotides with different mismatch numbers from each original tag as well as a small sub-set of longer RACE-derived sequences were also included. After statistical treatment of the different internal twin-replicas, normalization, and mismatch background correction, reproducible signals from different hybridization rounds (i.e. RNA replicas, and dye-swapped samples) were Drought stress under-represented GO biological processes in chickpea Figure 3 Drought stress under-represented GO biological processes in chickpea. Under-represented GO: biological processes as deduced from transcript abundancies (annotated to UniProt entries) in drought-stressed chickpea roots. Representation of GO: terms was calculated by the program ermineJ 2.0. Categories with values below 0.2 are less represented in a given data set.
selected for cluster analyses together with expression ratios of the 26 bp tags. A total of 79.05% probes on the micro-array shared the same tendency of expression with the respective tag when hybridized with cRNAs prepared from drought-treated roots from ICC588. From a total of 1,056 spots showing congruent results among different replicas and dye-swap experiments, 425 and 417 were upand down-regulated in both techniques, respectively, whereas 214 spots showed in-congruent results ( Table 6). UniTag expression ratios from both techniques are deposited in the main data matrix [see Additional file 1].

Comparison of different transcript profiling techniques: SuperSAGE versus SSH macroarrays and microarrays
We tested, whether our genome-wide transcription patterns are typical for roots of chickpea and other legumes, even if the profiles were obtained with other methods. Therefore we compared the present results to two studies addressing similar questions. In chickpea, Boominathan and co-workers [5] investigated whether pre-exposure to a dehydration shock improved adaptive responses of the roots during subsequent dehydration treatment. These authors identified 101 dehydration-inducible transcripts by repetitive rounds of cDNA subtraction, differential DNA-array hybridization, and Northern-blot analysis. Additionally, responses to exogenously applied abscisic acid (ABA) were also monitored. Since one analyzed time point was set 5 h after onset of drought, the results of this experiment should be at least partially comparable to our study in which RNA was isolated from roots 6 h after onset of drought stress. However, the results were not 100% congruent. It is important to note, that micro-or macro-arrays do not reliably differentiate between different transcript isoforms from gene families. Hybridization signals may integrate the hybridization intensities over all closely related transcripts, whereas SuperSAGE generates Drought stress over-represented GO cellular components in chickpea Figure 4 Drought stress over-represented GO cellular components in chickpea. GO: cellular components assigned to localization as deduced from transcript abundancies (annotated to UniProt entries) in drought-stressed chickpea roots. Representation of GO: terms was calculated by the program ermineJ 2.0. Categories with values above 0.8 are better represented in a given data set.
Heat map profiles of tags representing genes involved in signalling, and response-related processes Figure 5 Heat map profiles of tags representing genes involved in signalling, and response-related processes. Heat map profiles of tags representing genes involved in Ca 2+ -dependent signalling, general signal transduction, transcription regulation, protein-protein interactions, stress, ABA-dependent signalling, drought-response, damage and repair responses, and chromatin organization.
absolute numbers for each transcript variant (isoform). Due to the much deeper coverage of the transcriptome by this technique, almost all differentially expressed transcripts represented in the SSH libraries have at least one, but usually more counterparts in our 26 bp tags libraries.
A cluster analysis of transcription profiles obtained by macroarray hybridization and the differential expression of 26 bp tags from drought-stressed roots 5 h or 6 h after onset of the stress, respectively, is shown in Figure 8.
Similarly up-regulated (category C in Figure 8) and differentially regulated transcripts (i.e. down-regulated in the SuperSAGE experiment, but up-regulated on the macroar-ray, section D in Figure 8) under both conditions are listed in Table 7.
Investigating regulatory and protective mechanisms leading to desiccation tolerance (DT) in Medicago truncatula seeds, Buitink and co-workers [10] published another, at least partially comparable study. These authors used the 16 k Medicago microarray to monitor changes in the transcriptome of desiccation-sensitive 3-mm-long radicles at different time points during incubation in a polyethylene glycol (PEG) solution mimicking the effects of desiccation. These experiments identified several specific expression profiles at different time scales. A cluster analysis Heat map profiles of tags representing genes involved in channelling and detoxification-related processes Figure 6 Heat map profiles of tags representing genes involved in channelling and detoxification-related processes. Heat map profiles of 26 bp tags representing genes involved in water and ion channelling, ROS detoxification, and compatible osmolyte accumulation, with profiles of 97 26 bp tags annotated to CYP450 genes.
qRT-PCR confirmation of SuperSAGE results Figure 7 qRT-PCR confirmation of SuperSAGE results. TaqMan™ RT-PCR confirmation of SuperSAGE data by relative curve quantification using primers and TaqMan™ probes designed from 3'-and 5'-RACE sequences, respectively, that are derived from corresponding 26 bp tags. A) TaqMan™ assay with tag primer XPTm-Ca-22356 (derived from a 3'-RACE product; the original tag represents O65741_CICAR, the mRNA for a putative trans-membrane channel protein). Result: no difference between control and desiccated roots, i.e. constitutive expression (black arrow). B) TaqMan™ assay with tag primer XPTm-Ca-7975 (derived from a 3'-RACE product; the original tag represents an anonymous drought-induced EST). Result: earlier C T for the cDNA from desiccated roots (double headed arrow). Control (red), drought-stressed roots (blue).
comparing the results from desiccation-stressed Medicago radicles and drought-stressed chickpea roots is depicted in Figure 9. Transcripts up-regulated in both species (category D), regulated in opposite directions in the two species (section B), and transcripts down-regulated in chickpea and Medicago as a reaction to stress (section A) are listed in Table 8.

Differential expression of natural antisense transcripts (NATs)
As expected from an open-architecture technique, tags synthesized on coding strands of template cDNAs and, in addition, tags originating from the opposite (anti-sense) strands were discovered. Here, at least 170 tags matching to the reverse sequences of ESTs in the public databases were detected. These tags could represent potential natural antisense transcripts (NATs). In Lotus japonicus, SAGE tags representing NATs were induced during nodulation [12], and in Arabidopsis endogenous siRNAs derived from a pair of natural cis-antisense transcripts regulate salt tolerance [13]. Thus, it is possible that NATs are also involved in stress tolerance in chickpea. However, Galante and coauthors [14] demonstrated that a considerable portion of such NATs found in the databases are artefacts. In drought-stressed chickpea roots, we found -inter aliasup-and down-regulation of NATs for transcripts of several members of the aquaporin gene family and differential expression of NATs for phosphatase transcripts. However, since we did not exclude genomic priming by DNAse digestion of the RNA prior to cDNA synthesis, confirmation of the identity of these potential NATs will be subject to further studies.

Discussion
We analysed the expression of approximately 80,000 transcripts from un-stressed control and drought-stressed chickpea roots, respectively. It was previously estimated that the total number of average-sized transcripts per cell in higher plants ranges from 100.000 to 500.000 [15]. Thus, 80.000 26 bp tags provide ~1-fold coverage for transcripts present at a minimum of 6,2 copies per cell. There- Since, as we have shown, 75% of transcripts are present in less than 100 copies million -1 , and a small portion of transcripts is represented by more than 1000 copies million -1 , we probably missed a number of transcripts with less than 10-14 copies per cell. These may include transcripts for highly interesting pleiotropic proteins such as transcription factors possibly present at only 0,001 copies per cell [16]. Additionally, sampling larger organ sections with millions of cells obscures the specific transcript profiles of component cells. As a consequence, transcripts found in high copy numbers in just a few specialized cells are under-represented. As elegantly demonstrated for maize [17] and Arabidopsis [18], there are strong differences in transcription profiles between adjacent tissues. Therefore, the detection of rare transcripts should be improved by increasing the number of sequenced transcripts (as can be done with next-generation sequencing platforms such as Illumina's Solexa or ABI's SOLiD) in combination with a more selective sampling technique such as laser-capture microdissection [19].
Also, annotation of 55% of the 26 bp tags to the databases was straightforward. Considering that roughly 1,300 chickpea EST sequences are publicly available (at the time of annotation), most tags matched to sequences from the related model legume Medicago truncatula rather than to chickpea ESTs. On the other hand, a large number of significant hits represented fully uncharacterized database accessions, a fact that handicaps functional interpretation of the present chickpea transcription profiles more than the size of the 26 bp tags.
Additionally, a relevant problem arose from the interpretation of the assignment of different distinct 26bp tags to single UniGenes or tentative consensus sequences (TCs; [20]). We decided to classify such tags as isoforms of transcripts from members of gene families rather than assigning them to a particular gene, even if the expression patterns suggested functional homology. In addition, we could not differentiate whether these tags with similar annotation came from alternatively spliced transcripts of the same gene. Since in Arabidopsis and rice at least 21% of genes produce alternatively spliced transcripts [21], evidence beyond sequence homology of the SuperSAGE tags (i.e. 3'-and 5'-RACE sequences) is needed to decide between these two possibilities.

Differential expression of members of large gene families
One major advantage of the differential expression of 26bp tags as compared to macro-and micro-arrays is the very good differentiation between transcripts from different members of large genes families, which comprise the majority of all genes. Gene family members have more or less conserved sequences, and similar or different functions. For example, the cytochrome P450 (CYP) gene superfamily comprises more than 272 members grouped in 44 families in Arabidopsis thaliana, and more than 458 members in rice [22]. In the present study, 97 26bp tags revealed high-homology hits with members of the CYP450 superfamily, from which 33 could be assigned to their respective family through BLASTing against the Arabidopsis-CYP database (http://132.192.64.52/blast/ P450.html, data not shown). The involvement of CYP superfamily members in numerous catalytic reactions on a spectrum of substrates suggests that the transcription profiles of its members reflect this diversity. In accordance with the expectation, 43 (44.3%) of the 97 CYP-annotated 26bp tags changed their expression profiles at least 2.7fold with R (ln) (absolute value) > 1.0, whereas 47 (55.7%) showed only slight differences or constitutive levels. Among the 26bp tags assigned to drought stress-related CYPs, hits to CYP707A family (STCa-23852) which harbours ABA 8'-hydroxylases (key enzymes in ABA-catabolism; [23]), revealed a very slight up-regulation (almost constitutive levels), indicating that the turnover of ABA is already active under our experimental condition and time-points. UniTag STCa-18410, assigned to a CYP81F4 member, stands out for its extreme down-regulation (15fold). Although its function is not known, CYP81F4 members generally change their expression profiles dynamically after induction of water stress [24]. Considering the differential expression and diverse functions of individual members of gene families, whole-genome transcription profiling is only useful, if it differentiates between the different genes and their transcript isoforms in such families. Faithful discrimination between, and individual quantification of expression of these isoforms is therefore one advantage of the longer 26 bp tags.
However, also methodological inconsistencies may result in the observation of several different tags from a single gene [25]. To prevent methodological artefacts and assure the validity of the detected transcript variants, confirmative procedures such as double NlaIII digestion were standard in all libraries. Additionally, in silico routines for the exclusion of artefacts were applied (e.g. elimination of twin ditags and singletons [7]).
In the following sections, we discuss the expression of only a subset of drought-regulated genes and gene family members. We focussed on genes known to be involved in stress-perception, signalling and transcription initiation, Comparison between SuperSAGE expression profiles and macroarray-generated data (Boominathan and co-authors [5]). Three main categories from the cluster analysis in Figure 8 are detailed below. Section A: Constitutive tags versus upregulated transcripts on the macroarray. Section B, D: Non-corresponding SuperSAGE and macroarray transcript profiles. Section C: Up-regulated tags and up-regulated transcripts on the macroarray.  because at least some of them are well characterized, and often clear-cut evidence is available for their role in stressresponses [3]. As has been shown for tolerance to saltstress in rice, these genes or their products regulate early events in drought responses, differentiating between stress-tolerant and -susceptible genotypes [26]. However, genes involved in stress perception and signalling are not necessarily most up-or down-regulated.

Stress perception and signalling in drought-stressed chickpea roots
Dehydration-related stresses such as drought and salinity have ionic as well as osmotic attributes that elicit signal transduction cascades resulting in activation of effector genes to adapt the metabolism of the plant to the stress. In the model arising from research in Arabidopsis or rice, the first step of signalling is the perception of the stress through G-protein coupled receptors (GPCR), inositol polyphosphates, or receptor-like kinases (RLKs; [2]). In drought-stressed chickpea roots, 36 RLK transcript variants were detected. One of these increased in abundance more than 20-fold, whereas fourteen were 2 to 8-fold upregulated under stress, indicating a potential role of these transcripts in stress perception. Calcium transients are major signalling events in plants [27]. Thus, entry points and interconnecting links in major stress-related signal- Comparison between SuperSAGE expression profiles and 16K-microarray-generated expression data (Buitink and co-authors [10]). Four main categories from the cluster analysis in Figure 9 are detailed below. Section A: Down-regulated tags and down-regulated transcripts on the microarray. Section B: Contrasting SuperSAGE and microarray expression profiles (down-versus up-regulated) Section C: Contrasting SuperSAGE and microarray expression profiles (up-regulated vs. constitutive) Section D: Up-regulated or constitutive tags and 16K-microarray transcripts, respectively ling cascades involve Ca 2+ sensors and proteins regulated by Ca 2+ -concentration gradients between apoplast and cytoplasm [28]. In drought-stressed chickpea roots, Super-SAGE revealed moderate to significant changes in expression levels of Ca 2+ -responsive genes. Transcript isoforms encoding Ca 2+ sensors (e.g. calcineurin-B-like proteins, CBLs), Ca 2+ -channel proteins, and inositol-3-phosphate (IP 3 )-gated Ca 2+ -release (e.g. phospholipase C), were generally up-regulated (for an extensive characterization see [29,30]). In contrast, transcript isoforms transcribed from genes involved in downstream events like fine-tuning of, and interconnecting between, signalling cascades were up-as well as down-regulated in reaction to the stress. These included transcripts encoding a wide range of kinases such as Ca 2+ -dependent protein kinases (CDPKs; [31]), calcineurin-B-interacting protein kinases (CIPKs, [32])., and protein phosphatases class 2C.
Since the interplay between kinases and phosphatases balances activation and inactivation of proteins and with it the cross-talk between signaling cascades and metabolic pathways [33], measurement of transcriptional activity of kinase-encoding genes is important for understanding drought-response homoeostasis. Mitogen-activated protein kinases (MAPKs), however, do not seem to interact much with early drought-stress signalling in chickpea roots, since of 6 MAPK transcripts detected, 2 were downregulated more than 8-fold, and 4 similarly expressed as in control roots. In addition, the only MPKK detected was constitutively expressed, whereas from three 26 bp tags annotated to MPKKKs, two were down-regulated at least 4-fold (STCa-8893, STCa-10844), and one transcript was up-regulated at least 6-fold (STCa-2124).

Regulation of 14-3-3 transcripts
Signal transduction and regulation of metabolism achieved via phosphorylation-mediated transition of protein states require that phosphorylated proteins physically interact with specialized adapter proteins to fulfil their regulatory role. An example for such adapters are the phosphoserine/threonine-binding 14-3-3 proteins [34]. For example, 14-3-3 proteins are phosphorylation targets for SnRK 2.8, a member of the sucrose non-fermentingrelated kinase family, that is down-regulated in plants deprived of nutrients and with reduced growth [35].
Plants have large 14-3-3 gene families, and various 14-3-3 isoforms have varying affinities to target proteins. In Arabidopsis, at least 15 expressed members of the 14-3-3 gene family exist and exhibit high cell-and tissue-specificity as well as diverse expression levels [36]. In rice, at least four 14-3-3 transcript isoforms are induced by drought and salt-stress [37].
In drought-stressed chickpea roots, we detected 18 isoforms of 14-3-3 protein transcripts. However, contrary to Comparison of transcription profiles from Medicago dessi-cated young radicles (16k-microarray) with SuperSAGE chickpea profiles Figure 9 Comparison of transcription profiles from Medicago dessicated young radicles (16k-microarray) with SuperSAGE chickpea profiles. Comparison of transcription profiles from dessicated young radicles using the Medicago 16k-microarray (lane 1, Buitink et al. 2006) with SuperSAGE data (lane 2) from drought-stressed chickpea roots. A total of 147 26 bp tags could be linked through the UniProt data base and be used for comparison. Transcripts are clustered in groups, that follow similar expression patterns. Four main categories (detailed informations in Table 8) can be distinguished: Section A: Down-regulated transcripts on the microarray and down-regulated 26 bp tags. Section B: Contrasting microarray expression and SuperSAGE profiles (up-vs. down-regulated). Section C: Contrasting microarray and SuperSAGE profiles (up-regulated vs. constitutive). Section D: Up-regulated or constitutive 16K-microarray and SuperSAGE transcripts, respectivelyTaqMan™ RT-PCR confirmation of SuperSAGE.
results from drought-stressed rice, only three of these were up-regulated more than 2-fold, whereas ten were downregulated more than 4-fold ( Figure 5). Similar differential expression of 14-3-3 transcript isoforms was detected in young tomato roots under normal growth conditions as well as in response to salt stress and potassium and iron deficiencies, suggesting that especially one isoform (TFT7) may mediate cross-talk between the salt stress and potassium and iron-deficiency signalling pathways, respectively [38].

Transcription factors and involvement of ABA in early drought stress responses in chickpea roots
Another large and complex class of genes encode transcription factors (TFs). We could identify 124 UniTags from TF transcripts classified into 26 TF families ( Figure  10), whereas 8 remained un-classified. The majority of TF-UniTags annotated to the bZIP TF family (18), followed by UniTags matching to the HDZ (14), HMG (13), and WRKY (10) TF families. As described for the bZIP type TF family in Figure 5, expression profiles of transcription factors-encoding transcripts may display diverse regulation tendencies. This is also true for UniTags derived from transcripts encoding MYB family members, though these were reportedly involved in signal transduction under waterdeficit [39]. This observation may be related to the constitutive and even slight down-regulation of transcripts for proteins involved in ABA synthesis such as 9-cis-epoxycarotenoid dioxygenase [40], though Boominathan and co-authors [5] observed a strong up-regulation of a particular mRNA for this enzyme under drought stress, that probably escaped our detection. The difference in expression of genes involved in ABA-dependent signalling such as the isoforms of MYB TFs [41] may suggest that, as in maize, ABA as signal may be restricted to very specific regions of the root [42]. From six MYB transcription factors detected, two were up-regulated, and 3 were downregulated. Of 18 members of the bZIP TF family (see above) to which AREB factors belong [43], four were upregulated more than 3-, whereas three were down-regulated more than 4-fold.
In contrast, ABA-independent signalling seems to prevail: two out of five Drought-Responsive-Element-Binding (DREB2) TF isoforms thought to be regulated independently from ABA [44] were at least 3-fold up-regulated (STCa-4170, STCa-4212). Additionally, two further DREB transcript variants revealed constitutive levels, whereas one was 3-fold down-regulated (STCa-13360). Besides DREB2 TFs, WRKY TFs also seem to be involved in drought-stress responses in chickpea, because of the 10 detected family members, two were at least 2-fold (STCa-4132, STCa-10200), and two 4-fold up-regulated (STCa-11618, STCa-11619). Also, UniTag STCa-15340, homologous to an alfin-1-like TF transcript from alfalfa, belonged to the 40 most up-regulated transcripts under drought stress in chickpea (Table 4). In alfalfa, transgenic overexpression of a TF belonging to this class enhances expression of the endogenous MsPRP2 gene and improves salinity tolerance [45]. Thus, up-regulation of STCa-15340 indicates the involvement of alfin-1-like TFs in salt-as well as in drought-stress responses. Though already highly informative, our analysis of TF gene expression in drought-stressed chickpea roots certainly tapped the tip of an iceberg only, since we could assign only 124 UniTags to this class of genes. In fact, there are more than 1500 TF genes in the Arabidopsis genome [46]. Considering that we could assign only 22% of the 26 bp tags to well characterized entries in the public data bases, the number of TFs in our data set should be approx. 5 to 6 times as high as the one we could assign, i.e. approx. 650. This estimate is in good agreement with results from sugarcane, where 237,954 ESTs contained 600 TF sequences [47].

ROS scavenging and ROS-triggered signalling-related genes
Plants generate singlet oxygen-, superoxide-, peroxide-, and hydroxyl-radicals (ROS) that trigger a wide range of partly genetically fixed responses. ROS are released by either NADPH oxidases or peroxidases, that may exist alone or in combination in different plant species. Plant cells perceive changes in the concentrations of ROS as second messengers, and transform them into signals that change the transcription of genes [2]. On the other hand, disturbances in metabolism and photosynthesis by environmental stresses lead to ROS accumulation, which, if not controlled, can rapidly reach toxic levels in the plant cell [48]. Major ROS scavenging enzymes include superoxide dismutase (SOD), ascorbate peroxidase (APX), mono-dehydroascorbate reductase (MDAR), dehydroascorbate reductase [49], glutathione-S-transferase (GST) [50], glutathione peroxidase (GPX), glutathione reductase (GR), and catalase (CAT) [51]. The balance between the activities of these enzymes could be crucial for determining the steady-state level of ROS.
Our data reflect the complexity of ROS signalling and scavenging. For example, 10 out of 29 peroxidase isoforms were significantly up-regulated (data not shown), but only one NADPH oxidase was detected and moderately up-regulated. However, one transcript each for a hypersensitive-induced response protein and radicalinduced cell death 1-1 protein were detected, and both were significantly up-regulated. Of the ROS scavengers, six out of seven SOD transcripts, and one out of two transcripts coding for each catalase and dehydroascorbate reductase were more than 2-fold up-regulated, whereas two of the three ascorbate peroxidase (APX) transcripts were moderately or even significantly down-regulated. Since APX catalyses the H 2 O 2 -dependent oxidation of L-ascorbate (vitamin C;[52]), this potent ROS scavenging mechanism seems to be under-used in early responses of chickpea roots to drought. Considering that STCa-7166 representing an NADP-dependent isocitrate dehydrogenase transcript belonged to the most up-regulated transcripts in our study (R (ln) 3.58, Table 4), scavenging of ROS by gluthatione and recycling of oxydized gluthatione by this enzyme seems to prevail instead. This assumption is supported by the strong up-regulation (R (ln) 3.08, Table  4) of cysteine synthase (CS)-encoding UniTag STCa-2982.
In rice, as a consequence of CS up-regulation, both the total glutathione pool and reduced glutathione concentration were significantly increased in response to aluminium stress [53].
Still another stress-responsive gene family encodes glutathione-S-transferases (GSTs), key defence enzymes against xenobiotic toxicity, and has at least 56 members in rice [54]. From 14 chickpea 26 bp tags annotated to GSTs, four isoforms (STCa-977, 2175, 20830 and 12384) were at least 2-fold down-regulated, and three isoforms revealed up-regulation (STCa-3042, 12502 and 22470). Apart from ROS-scavenging, GSTs may also function in stress tolerance through signalling [55]. Considering the importance of these proteins for managing ROS-related stress, the GST transcript variants strongly up-regulated in chickpea roots under drought stress are potential targets for molecular breeding for drought tolerance. these crops. For example, the Arabidopsis genome harbours at least 35 genes coding for different aquaporins, that are differentially expressed under different stress conditions and, whereas one family member is up-, the other may well be down-regulated. In chickpea roots, we detected at least 42 different 26 bp tags from aquaporin transcripts, representing the three classes (nodulin-, plasma membrane-, and tonoplast-intrinsic forms). Interestingly, the over-expression of a certain aquaporin isoform in transgenic Arabidopsis resulted in altered expression patterns of other aquaporin isoforms with consequences for seed germination, seedling growth, and stress responses of the plants under various stress conditions [57]. These results suggest a concerted transcriptional regulation of at least a subset of aquaporin genes. Like in Arabidopsis, drought also elicited differential responses in the different members of the aquaporin gene family also in chickpea. It will be interesting to learn, whether the differentially expressed putative NATs corresponding to several of the sense aquaporin isoforms detected in chickpea are involved in the regulation of other gene family members. Also, whether differences exist in the expression of aquaporin isoforms between drought-tolerant and -susceptible genotypes of chickpea needs to be determined.

Changes in transcription of genes regulating compatible osmolyte accumulation
Beyond the broad repertoire of signalling cascades and signalling interactions that plants have on their defence lines against drought stress, the accumulation of sugars, sugar alcohols, amino acids, and polyamines, acting as compatible osmolytes against the osmotic disequilibrium is one of the most widespread strategies of plants to enhance their tolerance against drought stress [58]. Several genes involved in biosynthesis, transport as well as intermediate and catabolic pathways related to this strategy have been genetically and functionally characterized during the past years [59][60][61][62][63][64][65][66][67]. After screening the C. arietinum UniTags database for the behaviour of genes related to the above processes, transcription profiles and number of isoforms from at least 12 relevant genes were investigated in detail ( Figure 6). Related to sugar accumulation, one UniTag annotated to trehalose-6-phosphate synthase (STCa-18759, 2-fold down-regulated), and three 26 bp tags annotated to trehalose-6-phosphate phosphatase (STCa-9149 3-fold upregulated; STCa-11438 3-fold down-regulated; STCa-21065 constitutive) were observed. Trehalose plays an important role as compatible osmolyte and signalling molecule under drought stress [60,62]. However, since we do not observe strong up-regulation of genes encoding threhalose-6-phosphate synthase, we are hesitant to conclude that threhalose accumulated to high concentrations under our assay conditions. Additionally, one significantly up-regulated UniTag STCa-11968 repre-senting a galactinol synthase gene as well as at least three tags representing transcripts related to sucrose metabolism and transport with more than 2-fold expression changes (STCa-19100, STCa-8449, Ca-SS-16426) revealed, that the dynamics of sugar metabolism, transport and accumulation could be altered as a response to drought stress in chickpea. The positive role of galactinol synthase in stress tolerance has already been reported elsewhere [64]. Several 26 bp tags representing amino acid transport-and accumulation-related genes were detected. For proline, a compatible osmolyte [67], one 6-fold upregulated UniTag with homology to a proline/betain transporter (STCa-24308) as well as moderate down-and up-regulation of two 26 bp tags representing a negative regulator for proline accumulation (proline dehydrogenase; STCa-8454, STCa-8455) were revealed. This suggests that prolin accumulation may occur to some extent under drought stress in chickpea. The transcript levels for betaine aldehyde dehydrogenase UniTag STCa-14752, the key enzyme for glycine betaine synthesis [68], did not markedly change upon drought stress, indicating that this osmolyte, otherwise reported to accumulate under water stress [69], is not important for an early stress response.
Since the accumulation of polyamines as compatible osmolytes is discussed as protection against stress, we paid special attention to transcripts encoding arginine decarboxylase and spermidine synthase, that share important roles in putrescine and spermidine accumulation [70]. The detected transcripts were 2-fold (STCa-8875; arginine decarboxylase) and 3-fold (STCa-611; spermidine synthase) up-regulated. Our results indicate a differential influence of drought stress on mechanisms for compatible osmolyte accumulation as an early stress response in chickpea.

Comparison of SuperSAGE versus macro-and microarrays
Though drought-stress responses from roots of adult chickpea plants were compared to the roots of much younger seedlings from Medicago ([10], Table 7), the similar expression of several genes (section D, Figure 9) suggests similar reactions of the roots of both legumes to drought stress, independently of the developmental stage of the roots. Genes similarly up-regulated in both species inter alias include a certain extensin isoform, actually the most up-regulated transcript in stressed chickpea roots (O65760_CICAR). Extensins are hydroxyproline-rich proteins strengthening cell walls, and are often activated by mechanical stress [71]. Like other proline-rich cell-wall proteins, extensins require hydroxylation of prolines to 4hydroxyprolines to form the cell wall matrix [72]. Consistent with extensive hydroxylation of extensin, UniTag STCa-542 representing a prolyl 4-hydroxylase alpha subunit-like protein belonged to the most up-regulated transcripts in drought-stressed chickpea roots (R (ln) 2,722).
At the same time, STCa-1804, encoding an expansin-like protein, was most down-regulated (R (ln) -3,095) in stressed chickpea roots. Expansins weaken cell walls [73], and thus down-regulation of the expansin gene also would contribute to an increased strength of the cell walls of drought-stressed roots. In addition, UniTag STCa-24349, representing a gibberellin 2-beta-hydroxylase transcript, belonged to the most up-regulated transcripts in stressed chickpea roots (R (ln) 2,28, Table 4). The encoded enzyme inactivates gibberellin through βhydroxylation [74] and thus, its up-regulation should result in significantly deceased levels of bioactive hormone and a reduction of cell divisions and extensions. We conclude that in chickpea and Medicago, drought impairs division and extension of cells and results in growthretarded roots with strengthened cell walls.
Other transcripts similarly up-regulated in both chickpea and Medicago inter alias code for dehydrin, DREB, 2C protein phosphatase, UDP-glucose phosphorylase and blue copper protein which are discussed elsewhere in this paper. Taken together, a comparison of drought stress responses in chickpea and Medicago reveals many interspecies similarities and suggests to exploit the huge resources available for Medicago to test the functions of differentially expressed transcripts in chickpea.

Conclusion
Applying SuperSAGE to the analysis of abiotic stress responses in chickpea for the first time, our study presents the most comprehensive transcriptome profile of this crop available to date. It increases the number of chickpea ESTs from approximately 1,900 to more than 80,000, and the number of unique transcripts to more than 17,000.
The study identified major drought-stress signalling cascades resulting in differential expression of effector genes, and hints to the importance of ROS and N starvation as side stresses resulting from drought. Our study revealed, that (1) genes involved in photosynthesis and energy metabolism were down-regulated, (2) many genes involved in early responses to biotic and abiotic stresses were up-regulated, while (3) many other stress-responsive genes were down-regulated, and (4) regulatory genes encoding e.g. transcription factors or signal transduction proteins were both up-and down-regulated. We conclude that follow-up transcription profiling studies of responses to drought in chickpea must take into account the potentially deleterious effect of the stress on SNF and thus, on N supply to the plant in order to prevent mixing up responses to different stresses.
One important fact arising from our study is the unexpectedly high number of differentially expressed isoforms of members of large gene families, that was also observed in SAGE libraries from Lotus japonicus, where different levels of transcription induction among leghemoglobin gene paralogs were found [12]. These findings highlight the efficiency of tag-based techniques to discriminate different gene family members. At the same time, they underpin the necessity to -experimentally and linguisticallydistinguish between certain transcript isoforms (and the underlying genes) rather than summarizing them under a common term.
With this work, we aimed at identifying candidate genes as targets for molecular breeding for drought tolerance in chickpea. Numerous studies confirmed the polygenic nature of drought tolerance, for which single QTLs have only little individual effect. Considering the large number of genes located at QTLs for drought tolerance and related traits in cereals, comparing our transcription profiles to genes mapped to drought QTLs in these crops may help to decide whether SuperSAGE has identified such potential breeding targets.

RNA isolation and construction of SuperSAGE libraries
Total RNA was isolated from control and stressed roots using a modified CTAB procedure [75] followed by precipitation of the RNA in 3 M LiCl at 4°C overnight. From approximately 1 mg of total RNA, poly(A) + -RNA was purified using the Oligotex mRNA Mini Kit (QIAGEN, Hilden, Germany) according to the manufacturer's batch protocol. Subsequent steps for construction of SuperSAGE libraries were performed as detailed by [76]. However, instead of concatenation of di-tags and subsequent cloning and sequencing, amplified ditags were directly sequenced by 454 Life Sciences, Branford, CT, USA.

Tags quantification and data analysis
For each library, 26 bp long 26 bp tags were extracted from the sequences using the GXP-Tag sorter software provided by GenXPro GmbH, Frankfurt am Main, Germany. Library comparison and primary statistical treatment was carried out using the DiscoverySpace 4.01 software (Canada's Michael Simith Genome Sciences Centre, available at http://www.bcgsc.ca/discoveryspace). Scatter plots of the distribution of the expression ratios (R (ln) ) and significance of the results were calculated according to Audic and Claverie [11]. Additionally tags assigned to TIGR TCs indicating weak similarity to characterized genes were not selected. For targets from legumes different from chickpea, a maximum of three mismatches was allowed. The expected number of random matches (E value) was kept under 0.009 for individual TIGR databases, and 0.0009 for larger databases (e.g. NCBI nr restricted to fabaceae hits). Low complexity regions were rejected, whereas gap costs were set to 5-2 (NCBI BLAST standard setting).

Annotation test of in silico generated chickpea 26 bp tags using M. truncatula ESTs
In order to test the validity of the annotation of chickpea tag sequences through homologies with other legumes, 7,500 chickpea EST sequences deposited in the NCBI data bank were used to generate virtual 26 bp tags. Initially, all ESTs were screened for CATG sites using the BioEdit software, version 7.0.5.3 http://www.mbio.ncsu.edu/BioEdit/ BioEdit.html. Subsequently, all ESTs harbouring more than 30 bp between the most 3' CATG site and the end of the sequence were selected. After virtual 26 bp tag extraction, duplicate fragments were excluded, and the remaining tags BLASTed against public EST/mRNA databases following three main routes: I) BLASTing against the nonredundant (nr) NCBI nucleotide database (Fabaceae mRNA accessions), II) against the plant EST database at NCBI (M. truncatula accessions), and III) against M. trun-catula ESTs deposited in the TIGR gene indices. Complete EST/mRNA high homology target sequences derived from BLASTs (II) and (III) were retrieved and reBLASTed against the nr NCBI database (Fabaceae). The results obtained by these two BLAST strategies were compared with strategy (I) after exclusion of anonymous entries (e.g. AFLP fragments, shotgun sequencing clones, whole genome entries, whole chromosomes, BAC clones, etc.). BLAST parameters were set as described in the previous section.

Cluster analysis and functional category distribution analysis
Cluster analysis of the expression ratios (R (ln) ) used the software package Cluster 3.0 http://rana.lbl.gov/Eisen Software.htm. A distance matrix for the R (ln) was calculated with Pearson's correlation distance method [78]. Transcripts were clustered using the average linkage clustering routine under hierarchical clustering. P values for the most represented GO: biological processes observed after 6 hours of desiccation were calculated and correlated with the UniTag expression ratios (R (ln) ) using the "Receiver Operator Characteristic" (ROC) routine of the ermineJ 2.0 software package (University of British Columbia, 2006, http://www.bioinformatics.ubc.ca/ ermineJ. P-values for the representation of GO: categories are calculated according to [79] as indicated by the software developers.

Rapid amplification of cDNA ends (3'-RACE) using tag sequences as PCR primers
To test the versatility of the 26 bp tag-derived oligonucleotides for direct use as 3'-RACE PCR primers, cDNA amplifications were carried out with an initial denaturation step of 94°C for 2 min, followed by 30 cycles each of 94°C for 40 sec, 55°C for 1 min, and 72°C for 1 min, with a final extension step at 72°C for 4 min. Reactions contained 15-20 ng cDNA template, 10 pmol 26 bp tagbased primer, 10 pmol oligodT (t)14-NV primer, 200 μM dNTPs, 0.4 U Taq DNA polymerase (Genecraft, Germany) in buffer containing 1.5 mM MgCl 2 supplied by the provider. After amplification, products were separated in 1.5% preparative agarose gels. Bands corresponding to unequivocal amplicons were excised, and DNA extracted with Qiaquick cleanup columns (QIAGEN, Hilden, Germany). Cloning of PCR products as well as colony PCR screening followed stanrdard blue-white screening procedures [80]. Positive clones were sequenced via ABIprism multi-colour fluorescence-based DNA analysis system (APPLIED BIOSYSTEMS, Foster City CA, USA).

Confirmation of SuperSAGE expression profiles via qRT-PCR
Parallel RNA extractions of the same tissue, from which the SuperSAGE libraries were derived, were carried out as described in a previous section. Approximately 500 ng of total RNA were further processed to poly(A) + -RNA via Oligotex matrix (QIAGEN, Hilden, Germany). cDNA was synthesized using the Superscript III double-stranded cDNA synthesis kit (INVITROGEN, Karlsruhe, Germany).
Resulting cDNA was quantified with two parallel methods: i) Nanodrop spectrometer measurement (NANO-DROP, Willmington DE, USA, and ii) Caliper chip quantification (CALIPER, Hopkinton MA, USA).
SYBR green oligonucleotide deduction was carried out with the software package Primer Express, version 2.0, provided by Applied Biosystems (Foster City, CA, USA) with 3'-or 5'-RACE products from selected 26 bp tags as starting points. The two TaqMan assays used in this study were provided by GenXPro GmbH, Frankfurt, Germany, and used according to the protocol included in the kit.
The real-time PCR reactions for SYBRgreen and TaqMan assays used the Power-SYBRgreen PCR master mix and the TaqMan-Universal PCR Master mixes, respectively (Applied Biosystems). RT-PCR amplifications were carried out in a StepOne RT-PCR System machine with the following temperature profile for SYBRgreen assays: initial denaturation at 95°C for 10 min, followed by 40 cycles of 95°C for 10 sec. and 60°C for 20 sec. (annealing and elongation). TaqMan assay temperature amplification profiles consisted of an initial denaturation at 95°C for 10 min, followed by 40 cycles of 95°C for 10 sec. and 65°C for 30 sec. Amplicon quality was checked by an additional melting curve gradient with fluorescence measures after each temperature step. The amplification of the target genes at each cycle was monitored by SYBRgreen-or Taq-Man probe-released fluorescence. The Ct, defined as the PCR cycle at which a statistically significant increase of reporter fluorescence is first detected, was used as a measure for the starting copy numbers of the target gene. Relative quantitation of the targets amplified via SYBRgreen assays was performed by the comparative ΔΔCt method. Genes amplified by TaqMan assays were quantified via the Relative Standard Curve Method (Applied Biosystems). The efficiency of each primer pair was checked with cDNAs from control and 6h-desiccation as standard templates. The RT-PCR data were normalized with the relative efficiency of each primer pair.

Confirmation of expression profiles via microarrays
SuperSAGE expression profiles were confirmed by direct spotting of a selection of 26 bp tags onto a 16 K Agilent microarray (AGILENT TECHNOLOGIES, Santa Clara CA, USA). Three thousand UniTags with different expression levels under drought, salt, and cold stresses (salt and cold stress expression profiles are not approached in the present paper) were selected. From the 3,000 Tags, a subset of 2,796 oligonucleotides was spotted in duplicate onto different sections of the microarray. Additionally, for each of the 3,000 selected Tags, oligonucleotides with mismatches were spotted onto the microarray in three sets as follows: i) mismatch at position 7; ii) mismatches at positions 7 and 13 and iii) mismatches at positions 7, 13, and 20, respectively. Background correction was achieved by the Feature Extraction softwareTM (Agilent Technologies), subtracting the mismatch intensities for each spotted Tag. Microarray design, spotting and hybridizations were carried out by ARRAY-ON GmbH, Gatersleben, Germany, according to the AgilentTM protocols (AGILENT TECHNOLOGIES, Santa Clara CA, USA).