Citrus fruits are the first fruit crop in international trade in terms of economic value (FAO, 2004). Citrus fruits are typically grown in 140 countries located in tropical and subtropical areas with "Mediterranean" type climates, often facing severe abiotic stresses such as salinity, drought and iron chlorosis. Citrus species also suffer from different diseases and pests that considerably affect tree growth and fruit crop. The survival of the citrus industry is today critically dependent on genetically superior cultivars. However, citrus improvement through traditional techniques is unfortunately very difficult due to the unusual combination of biological characteristics of Citrus species, their low genetic diversity and the long-term nature of tree breeding. Thus, Citrus show many biological characteristics such as gametophytic self- and cross-incompatibility, apomixy, juvenility, heterozygosis, dormancy, and surprising root/shoot interactions, that strongly hamper Citrus breeding. On the other hand, genetic and allelic diversity in Citrus cultivars is very scarce. The global linkage disequilibrium in the cultivated citrus that probably originated from three major taxa, may be the result of an initial allopatric evolution and further limitation for predominant apomixy [1]. The fact that only mutational and/or epigenetic events are apparently involved in the diversification of secondary species, combined with human selection, have strongly reduced global genetic diversity, restricting opportunity for genetic advance.

Genomics has provided new tools for crop improvement, helping to identify and select candidate genes responsible of agronomic characters of interest, and allowing the development of fast methods to incorporate these characters into crop plants. After the completion of the Arabidopsis genome sequence [2] and the publication of sequences of indica [3] and japonica [4] rice, plant researchers have been able to scan these genomes to identify and compare genes of interest. The completion of the poplar genome sequence [5] will supply a model for tree life forms.

EST sequencing projects have facilitated appropriate strategies for gene discovery [6], molecular markers identification [7, 8], and many other functional genomic developments and tools, e.g. microarray approaches [9–11]. In Citrus, previous EST sequencing projects have released more than 130000 ESTs to Gen Bank, mainly from C. sinensis, C. unshiu and C. clementina [12–16]. This information has been used to develop two different microarray platforms, based on cDNA and short oligo sequences [12, 17]. In this work, main efforts have been specifically focussed on the study of pivotal traits for Citrus breeding, such as fruit quality, productivity and salinity tolerance. In citrus there are many aspects of fruit quality such as fruit size, shape, colour, texture, flavour and aroma compounds, and several other organoleptic properties that are acquired during ripening and earlier stages of growth [18–20]. Regarding productivity, pivotal traits to be improved are the capacity for fruit set and the resistance to abscission. Clementine mandarin (Citrus clementina), for example, is a self-incompatible cultivar that shows elevated ovary and fruitlet abscission while sweet orange varieties (Citrus sinensis), that in general exhibit standard fruit-set ratios [21, 22], may lose most of the yield during ripening. In addition, the quantity and quality of water can become a limiting factor to economically viable production. In Citrus, it is notorious for example, that during the periods of drought, leaves and fruits remain attached to the tree until water is available and soon afterwards these organs abscise [23, 24]. It is also well known that salt excess affects the size and quality of the production. The capability of Citrus to tolerate salinity is mostly related to the ability of the rootstock to exclude chloride, although the nature of this mechanism remains unresolved [25]. Tolerant Cleopatra mandarin rootstock (C. reshni), for example, accumulated lower chloride amounts than sensitive Carrizo (C. sinenis × Poncirus trifoliata) [26, 27]. The scion variety also plays a role in salinity damage, and more tolerant varieties such as Clementine are generally preferred to sweet oranges [28]. Toward this goal, cDNA libraries were prepared from the following selected genotypes: the varieties Citrus clementina (cv. Clementina de Nules; elevated fruit quality, low setting) and C. sinensis (cv Washington Navel and Navelina; pre-harvest abscission and low salt tolerance, respectively), and the rootstocks C. reshni (cv Cleopatra; salt tolerant), and C. sinenis × Poncirus trifoliata (cv Carrizo; salt sensitive). The information generated with this effort, complementary to the Spanish Citrus Functional Genomics Project [12], has also been used for the construction of a second generation cDNA microarray of recent release. In the study, special attention has been paid on the methodological aspects, in order to obtain accurate estimates of the number of different transcripts and precise predictions of their function.

The results of this joint initiative of the IVIA, CIRAD and Genoscope designed to provide new information and useful tools for Citrus improvement, will speed up the discovery of genes of major agronomic interest and facilitate the development of new markers and methods to rapidly identify improved genotypes.

Citrus is the main fruit tree crop in the world. However, traditional breeding for Citrus cultivar improvement faces many serious impediments due to the unusual combination of biological characteristics of Citrus, their low genetic diversity and the long-term nature of tree breeding. Genomic technology can overcome these limitations providing new tools, for example, to produce more efficient varieties and rootstocks and to identify new genes, alleles or genotypes of agronomic relevance. Improvement of knowledge of the transcriptome is one of the first tasks that have to be developed in order to understand the developmental biology of the plants and how these respond to environmental stresses. This work that pursues this goal provides a deep insight into the Citrus transcriptome specifically related to three major commercial traits i.e improved fruit quality, higher yield and tolerance to environmental stresses, especially salinity.

Towards this objective, 10 cDNA libraries representing particular treatments and tissues from selected varieties and rootstocks differing in fruit quality, resistance to abscission and tolerance to salinity were generated to provide a large and enriched expressed sequence tag collection. The assembly of these sequences, more than 52600 ESTs, allowed the identification of 15660 transcription units. The results of this analysis are comparable to previous reports in Solanum tuberosum, that detected 19892 unigenes from 61949 ESTs [62], or in Sorghum bicolor, with 16801 unique transcripts derived from 55783 ESTs [63]. The data showed that all sequences from the Citrus species analyzed, from both this study and databases, were almost identical, suggesting that the differential behaviour of these cultivars during normal fruit growth or when facing environmental adverse conditions is more likely associated with differences in gene regulation rather than with sequence divergence. This result is not unexpected since Citrus posses a high level of phenotypic diversity while global genetic diversity, analyzed with molecular markers, appears to be very low or practically null. A large effort was made to determine the real number of different transcripts represented in the EST collection. It is well known that the accuracy of EST clustering is affected by various error sources, such as sequencing mistakes, contaminant sequences and the presence of products of chimeric splicing. The most common error occurs when different ESTs from the same gene are falsely separated into two or more clusters [64]. To overcome this difficulty the level of redundancy was estimated comparing all unigenes with each other, and clustering them in supercontigs, that are more likely to correspond to real transcripts. The level of redundancy was estimated to be 26%, a value similar to that obtained in sugarcane, for example [65]. This first restriction suggests that the likely number of unigenes in the Citrus collection is closer to 13900 rather than to 15660.

It was also crucial to determine the occurrence of contaminant sequences, mostly from microorganisms, since many samples were taken from open field. The presence of contaminant sequences (mainly from bacteria and fungi) is a general problem not attended in any of the EST projects we have examined. For instances, a BLASTN search performed with several contaminant sequences found in this work against the viridiplantae section of the GenBank EST database revealed a considerable number of ESTs regarded as plant sequences that really corresponded to fungi species (data not shown). Thus, the analysis reported in this work may help to prevent the presence of sequences from contaminant species in the databases. Determining the species of the unigenes best hit sequences helped to identify putative contaminants, allowing not only a more precise estimation of the real number of Citrus transcripts but also criteria for microarray EST selection. Since about 400 Unigenes were believed to be contaminant sequences from other species, the number of Citrus expressed genes was reduced to 13500.

A relevant observation of this work is that more than 38% of the 13500 unigenes (5159 sequences) are novel Citrus unigenes. EST sequences were obtained from two kinds of cDNA libraries, normalized full length and standard libraries. The normalized library was generated with a wide variety of reproductive and vegetative tissues, enriched with developing fruits, abscission zones and salinity samples. In the first strategy, the normalization process very effectively increased low abundant transcripts, since the bulk of the novel unigenes described (4673) derived from this library. The standard libraries that were constructed from either samples of fruits, abscission zones or salt-treated organs, were generated with the idea of providing transcripts specifically expressed at these particular tissues and organs without increasing redundancy. To estimate the contribution of the standard libraries in terms of gene novelty, a set of unigenes presumably involved in the three processes of interest was selected and the libraries containing them identified. Although for an accurate evaluation of these contributions confirmation of tissue specificity appears to be mandatory, these preliminary estimates showed that most of the novel genes were certainly recovered in the normalized library. Moreover, the contribution of the specific standard libraries to the unigene set maybe roughly estimated to be between 4 and 15%.

The primary homology searches performed against different databases allowed annotation of most unigenes, with more than 73% of them displaying a similarity degree higher than 60%. These results agreed those obtained in previous Citrus [12] or sugarcane [65] EST projects. It was also shown that most of the sequences that did not produce significant hits in the BLASTX searches were shorter than 500 bp (Fig 2) and probably did not carry coding sequences. Additional efforts were performed to characterize these sequences, with supplementary BLASTN searches against the non redundant and EST nucleotide databases. These analyses gave rise to the suggestion that 647 ESTs of the Citrus unigene set may correspond to Citrus exclusive genes since the significant hits they produced were only for Citrus sequences, in spite of the more than 8.5 million EST sequences derived from plant species deposited at the GenBank,

Further improvement of the annotation was carried out through searches performed against secondary databases, composed of patterns or signatures. Although these prediction methods can work with DNA sequences, the error prone nature of ESTs, mainly shifts in the reading frame (missing or inserted bases) or ambiguous bases, may result in inaccuracies and loss of information. Thus, a crucial step in annotation is the robust translation of the ESTs to yield predicted polypeptides. Polypeptide sequences posses a better template for almost all annotation tools, including InterPro and Pfam, and allow the assembly of more accurate multiple sequence alignments. High quality polypeptide predictions can be applied to functional annotation and post-genomic study in a similar way to those available for completed genomes. In the work presented here, the protein translation was performed with Prot4EST, a prediction pipeline that incorporates freely available software (ESTscan, Decoder, HSP tiling) to produce final translations that are more accurate than those derived from any single method [36]. The use of the interProScan tool [37], allowed simultaneous search for motifs against 9 databases. This search produced significant results for almost 11000 predicted proteins, including 342 unigenes that did not have significant hits in the BLAST searches. From the 20 most abundant motifs found in Citrus with Pfam, 7 of them also were included in the top 20 list at the Pfam database, enforcing the accuracy of the analysis and the representativity of the Citrus EST collection. The molecular functions associated with these 20 motifs can be grouped in 4 categories: protein-protein interaction (47.26%), nucleic acid binding (27.51%), protein modification and binding (14.9%), and calcium metabolism (6.17%), which indicates the relative significance of these cellular functions in Citrus.

The distribution of motifs on the polypeptide sequences predicted to be complete proteins showed that the bulk of sequences displayed a single motif (76%). Proteins carrying 2 ore more motifs showed unlike signatures rather than repeats of the same motif. For instance, the number of proteins with 2 different signatures (452) was six times higher than the number of proteins that had 2 identical motifs (69). Similar relationships were found for other number of repeats.

The Gene Ontology annotation of the Citrus unigenes was performed with BLAST2GO (B2G), a recently developed BLAST-based GO annotation software [38]. The B2G approach uses multiple BLAST hits to search for functional annotations and assigns GO terms to the query sequence applying an annotation algorithm that considers HSP length, e-value, percentage of similarity, Evidence Code of the source annotations and the topology of the Gene Ontology. This is in contrast to most EST projects that perform annotation solely by direct assignment of the GO terms to the best hit of BLAST searches [12, 66, 67]. The B2G method has shown to have a high annotation recall and has been used in other EST projects in eukaryotes [68].

Metabolic pathways responsible of important agronomic traits were further surveyed to determine the extent of representation of these pathways within the Citrus unigene set. In addition to the finding that most enzymatic steps were represented by Citrus homologs, a preliminary estimate of gene duplication based on the occurrence of paralogous sequences was also provided. Defining such relationships and understanding functional diversification of paralogs is an important field of research in genomics-assisted crop improvement. Lignin biosynthetic pathway was the object of a deeper analysis since Citrus fruits are very rich in products with beneficial effects in preventing cancer, diabetes, ...etc such as fiber [42, 69]. Dietary fiber, that consists of non digested structural and storage polysaccharides and lignin, lowers cholesterol levels and helps to normalize blood glucose and insulin levels [70]. The detailed analysis of lignin biosynthesis pathway, carried out in Citrus, indicated that in comparison with Arabidopis, Citrus possessed at least 9 additional enzymatic activities involved in lignin synthesis (Table 5). Furthermore, the results obtained for gene models from Populus trichocarpa [47], also appears to support the idea that the extensive formation of secondary xylem in tree species, requiring high levels of lignin synthesis may have been the origin of the expansion of the genes involved in this pathway.

More than 570 unigenes have been suggested to be possible conserved orthologs of Arabidopsis single copy genes. Recent studies have indicated that ancient polyploidy is common across angiosperm lineages and in fact, the genomes of all angiosperms may have been influenced by at least one genome-wide duplication event [71]. Despite such events, single-copy, apparently orthologous gene sets have been identified in a broad range of angiosperms [72, 73]. Selection against duplicates may be maintaining these genes as single copy, and therefore are precious markers for comparative genetic and physical mapping, and also for phylogenetic analyses [74]. Identification of such genes in Citrus species is mandatory to perform this kind of analyses [75]. The study also revealed a number of genes that might be duplicated in the Citrus genome, while remained as single copy genes in Arabidopsis, although the possibility of finding additional copies of these genes could not be discarded, when the whole transcriptome of Citrus is available. If these duplications are the result of individual events or were caused by a genome-wide duplication cannot be answered with the current information.

Comparative genomics was also used to obtain an overview of conserved gene families in Citrus. All Arabidopsis gene families studied were well represented in the Citrus EST collection, although the number of their members was generally smaller, probably because the unigene set was only a partial representation. For the same reason, the finding of families that in Citrus clearly outnumbered their Arabidopsis counterparts is highly significant. The phylogenomic analysis performed on the gene families of ammonium transporters and glycosyltransferases supported this idea confirming the occurrence of additional members in the Citrus families. Ammonium is one of the prevalent nitrogen sources for growth and development of higher plants including Citrus. The ammonium transporter family is composed in Arabidopsis of 5 AMT1 related genes and AMT2, which is more closely related to ammonium transporters from prokaryotes than to AMT1. AtAMT2 is likely to play a significant role in moving ammonium between the apoplast and symplast of cells throughout the plant [61]. Interestingly, there are three AMT2 like genes in Citrus (Fig 6A). Glycosyltransferases are a ubiquitous group of enzymes that catalyse the transfer of a sugar moiety from an activated sugar donor onto saccharide or non-saccharide acceptors. Although many glycosyltransferases catalyse chemically similar reactions, they display remarkable diversity in their donor, acceptor and product specificity and thereby generate a potentially infinite number of glycoconjugates, oligo- and polysaccharides. [76]. Thus, the additional members found in this family, might be related to the complexity of sugar synthesis that takes place in the Citrus fruits.

The assembly of more than 54000 Citrus ESTs from five cultivars differing in basic fruit developmental aspects, such as major traits for fruit quality and production, and in the responses to environmental conditions, provides an unprecedented insight of the Citrus transcriptome. This study contributes new tools for Citrus genetic and genomic analyses. The unigene set, composed of ~13000 putative different transcripts, including more than 5000 novel Citrus genes, was assigned with putative functions based on similarity, GO annotations and protein domains. In addition, comparative genomics was used to analyze the Citrus transcriptome, and evidences for numerous cases of gene duplication events were presented. The similarity analyses indicated that the sequences of the genes belonging to the varieties and rootstocks studied were essentially identical suggesting that the differential behaviour of these species cannot be attributed to main sequence divergences. This set of processed EST sequences has greatly contributed to the development of a new Citrus microarray.