Background

Grapes species (Vitis spp.) represent the most widely cultivated and economically important fruit crop in the world [2]. The use of grape berries includes the production of juice, fresh and dried fruit, and distilled liquor, although wine produced from cultivars of V. vinifera has the highest economic value of grape products. Grapevine berries are non-climacteric fruits with a characteristic double sigmoid growth curve. The initial phase of exponential berry growth (stage I) is followed by a lag phase (stage II), with growth resuming after the onset of ripening or "veraison" (stage III). Berry development is characterized by changes in numerous biological processes, including cell division and enlargement, primary and secondary metabolism, and resistance or susceptibility to abiotic or biotic stresses [3, 4]. The importance of this plant species to agriculture has made the development of genomic resources a high priority. Among these resources, transcriptional profiling of important grape tissues is a practical option that may reveal transcriptional complexity and changes in this dynamic developmental system.

Massively parallel signature sequencing technology (MPSS) [5, 6] is a sequence-based method for measuring gene expression. The depth of sampling provided by MPSS can identify a nearly complete inventory of transcripts in a given sample. The method is based on a unique process for parallel sequencing, which starts with the cloning of a cDNA library on 5 μm diameter microbeads; one transcript from the original RNA sample is represented on a single bead [5]. From each bead, a sequence of the 'signature' of 17 or more nucleotides is obtained by successive round of sequencing reactions [5–7]. These signatures are derived from and include the most 3' occurrence of a specific restriction enzyme site in a transcript (most often DpnII, producing signatures that start with GATC) [5, 6]. The output of the method is conceptually similar to a possibly more familiar method called Serial Analysis of Gene Expression (SAGE) [8]. However, the MPSS technology permits the simultaneous sequencing of millions of signatures from a given library [5]. By matching these signatures to the genome to identify specific genes, the abundance of each signature represents and measures the gene expression levels in the sample tissue. Among several published applications of this technology, we have previously conducted comprehensive transcriptional analyses of the reference plant species Arabidopsis thaliana and rice [7, 9]. While MPSS, SAGE, and expressed sequence tags (ESTs) are all sequence-based technologies for transcriptional profiling, MPSS provides more thorough qualitative and quantitative description of gene expression due to its tremendous depth. While novel sequencing technologies, such as sequence-by-synthesis (SBS) and 454, offer deeper sequencing and longer read lengths, none have yet demonstrated consistently better results than MPSS for mRNA profiling [10].

In this report, we have measured gene expression in developing grape berries using MPSS, compared this expression profile with that provided by the current Vitis Unigene set [4], and we developed a novel web-based resource for utilization of the grape MPSS data. As a result of this analysis, we were able to annotate thousands of signatures matching predicted genes, quantify the expression level of these genes in the developing berries, compare the expression profiles derived from ESTs and MPSS signature frequencies, and expand the coverage of known transcripts in an important grapevine organ at a specific developmental stage. Because these data are based on sequences, they comprise a resource that will be useful for the annotation of any grape genomic sequence produced in the future.

Results

Analysis of the V. viniferaberry MPSS dataset and signature annotation

An MPSS library was constructed using RNA extracted from stage II berries (green, hard) that were sampled from field-grown V. vinifera cv. Cabernet Sauvignon. After cloning of the cDNA library onto beads, 17-base and 20-base signatures were generated by MPSS sequencing [5, 6]. We note that these are not independent samples, in that 20-base signatures are obtained by extending previously recorded 17-base signatures by three nucleotides; due to a low failure rate at each additional base of sequencing, the raw count of sequences is lower for the 20-base data. A total of 2,635,293 17-base and 2,259,286 20-base signatures were produced that corresponded to 30,737 and 26,878 distinct sequences, respectively (Table 1A–C). This represents a discovery rate or average raw abundance value of approximately one distinctive sequence for every ~49 sequenced cDNA tags.

Table 1 Summary statistics of raw 17- and 20-base MPSS signatures from grape berries.
Full size table

Initially, to link the MPSS signatures to predicted gene annotations, all sites ("GATC") that could potentially produce an MPSS signature were identified from the available Vitis Unigene dataset in public databases. This comprised 14,658 contigs (1,307 from non-vinifera Vitis species) and 14,931 singletons (1,080 from non-vinifera Vitis species). All potential signatures starting with the GATC anchor sequence were extracted from both sense and antisense directions of the grape sequences. A total of 84,834 and 48,490 distinct 17-base potential signatures were identified, respectively, in contigs and singletons of this version of the Vitis cDNA data. When both datasets were combined, the total number of unique genomic signatures equaled 123,563. The total number of in silico-extracted distinct MPSS signatures is approximately six-fold lower than the 753,894 distinct "genomic" MPSS signatures reported for the completed Arabidopsis sequence [11], reflecting the incomplete nature of the grape EST dataset and the lack of intergenic and intron sequences.

Observed MPSS signatures were classified based on the output of "reliability" and "significance" filters [11]. The purpose of these filters is to separate high quality data, which is represented by signatures encountered above specified frequency thresholds, from background signal generated by very low abundance MPSS signatures. As with other MPSS datasets, the grape library was generated from four sequencing runs representing two sequencing frames [11]. There were two runs for each of the "two-step" and "four-step" sequencing frames. The reliability filter asks whether a signature is present in more than one sequencing run (of the four total runs); signatures observed in more than one run are considered "reliable". The significance filter identifies as "significant" only those signatures with a normalized abundance greater than three transcripts per million (TPM). The classifications of 17- and 20-base expressed signatures in terms of reliability and significance are shown in Tables 1A–C and 2; 96.8% of all MPSS signatures corresponded to the "reliable" and "significant" category, consistent with an extremely low abundance for signatures not passing the filters. This value is similar to the 97.5% reported for the Arabidopsis MPSS dataset [11]. Among MPSS signatures with exact sequence matches to EST contigs (Table 2A–B) and singletons (Table 2C–D), unique "reliable" and "significant" signatures represented the largest category (more than 60% of the unique signatures).

Table 2 Distinct MPSS signatures matching EST contigs or singletons classified based on "reliability" and "significance" filters.
Full size table

Expressed signatures were mapped to grape EST contigs and singletons based on exact matches to the in silico extracted "potential signatures" (see above). A total of 5,794 and 5,407 contigs were matched by expressed reliable and significant 17-base and 20-base MPSS signatures, respectively (see Additional file 1A–1B). This represented, on average, more than 40% of all known Vitis sp. genes. On the other hand, only 14% of singletons in the Vitis sp. EST set were matched by MPSS signatures (Table 2C and 2D). The vast majority of the unmatched Vitis sp sequences had in silico potential signatures that were not detected in the MPSS data. It is possible that the corresponding genes were not expressed in this sample; alternatively, unmatched contig and singleton EST sequences may represent 5' reads of cDNA clones, and thus fail to represent 3' regions where the majority of MPSS signatures originate. The disproportionate representation of singleton ESTs among the unmatched set is consistent with this later interpretation, because singleton ESTs in the Vitis dataset are more often the product of 5' sequencing reactions.

Most signatures matched a single contig or singleton, while ~40% matched two or more [see Additional file 1A–1B]. In excess of 70% of matched contigs and singletons showed a one-to-one assignment to a reliable and significant MPSS signature (Figure 1) [see Additional file 2]. The remaining sequences had one-to-many assignments of up to a maximum of 16 different signatures to a single contig [see Additional file 3]. Sequences of 17–20 bp are rarely duplicated by chance in unrelated genes [7] [see Additional file 4]. Instead, biological factors involving gene duplication or transcript processing may complicate the unambiguous assignment of signatures to transcripts. Thus, gene family members with high sequence similarity are likely to yield distinct transcripts containing the same signature, while the use of multiple polyadenylation sites or alternative splice site selection can yield multiple signatures from the same transcription unit. To estimate the frequency of alternative termination, a subset of 5,145 contigs was properly aligned in their 5' to 3' orientation. From this subset, 975 contigs matched by at least two MPSS signatures were identified. The abundance counts of 17-nucleotide significant and reliable MPSS signatures were transformed to relative frequency values and the location of each signature was plotted along the 3'-to-5' axis for each of the 975 contigs (Figure 2). The signature frequency per contig decreased exponentially from the 3'-to-5' direction. On average, ~70% of all signatures originate from the 3' most GATC site, while only ~29% and ~14% of signatures originate from the second and third 3' most positions (further 5'), respectively. Therefore, most of the transcripts matched by MPSS are the product of polyadenylation at the most distal of all recorded 3' sites. It is possible, however, that the MPSS signatures that did not match ESTs (contigs or singletons) are derived from longer 3' ends for which transcript sequence was not available.

Figure 1
figure 1

Frequency distribution of grape ESTs matched by filtered MPSS signatures. Reliable and significant MPSS signatures were matched to EST contigs and EST singletons. Up to 16 and 10 MPSS signatures matched to one EST contig and singleton, respectively. The proportion of the number of MPSS signatures matching to (A) EST contigs and (B) EST singletons are represented by the bar graph.

Full size image
Figure 2
figure 2

Frequency of reliable and significant 17-mer MPSS signatures in a subset of 5'-to-3' oriented contigs. Signatures were mapped based on their location relative to the 3' end of the EST contigs. Most signatures were found at the 3'-most DpnII site, indicated as position #1 on the x-axis. However, expressed MPSS signatures were found as far 5' as eighth DpnII site from the 3' end of the contig.

Full size image

Analysis of sense-antisense expression

Approximately 15% and 11% of the EST contigs and singletons, respectively, were matched by MPSS signatures in both sense and antisense orientations (Tables 3A–B). The MPSS signature frequencies were much higher on the sense strand for some sequences, while other sequences had higher MPSS abundances on the antisense strand [see Additional file 5]. Contigs matched in both orientations represented ~12% of the known berry transcriptome (of a total of 7,828 including contigs derived from EST sequenced and cloned from cDNA libraries other than green stage II), with the 2,891 MPSS signatures matching these contigs representing ~52% of the total MPSS abundance. It is possible that the sense-antisense transcript pairs are an important transcriptional feature which could provide a mechanism for post-transcriptional gene silencing [12] during this dynamic phase of berry development. Functional categorization of these contigs showed no particular overrepresented category (Figure 3). Moreover, none of these contigs had significant identifiable tBLASTx hits in both reading frame orientations, suggesting protein coding is a property of only one strand. It is possible that anti-sense transcripts could result from overlapping 3'UTRs of adjacent genes, or from transcription of an overlapping non-coding RNA.

Table 3 Matched and un-matched Vitis EST contig and singletons.
Full size table
Figure 3
figure 3

Functional categorization of transcripts with both sense and anti-sense MPSS signatures. EST contigs, which have both sense and anti-sense MPSS signatures, were categorized based on GO (Gene Ontology) annotation and the proportion of each category is displayed in pie-chart: (A) Cellular component, (B) Molecular function, and (C) Biological process.

Full size image

Expression profiles determined by EST and MPSS abundances

To quantify gene expression levels, we used the relative abundance of the 7,686 reliable and significant 17-base MPSS signatures from the stage II berry library. These signatures represent the most robust subset of the MPSS expression data. Although the remaining 1,734 reliable but not significant signatures were not considered in this analysis, prior analysis suggests that these signatures are likely to represent genuine transcripts expressed at very low levels [11]. The transcripts represented by these signatures may be expressed at higher levels in different specific cells or tissue layers that were not sampled.

The MPSS sequences provide an inventory of the transcript population in a given organ or tissue that can be sorted based on abundance. This data is particularly powerful when aligned with EST data from related tissues, as it allows sorting based on abundance and predicted gene function. The MPSS-matched set of 5,791 grape EST contigs are derived from a series of cDNA libraries that survey several stages of plant development, as well as responses to biotic and abiotic stress [4]. Off these, 4,753 contigs contained ESTs derived from one or more grape berry tissues, while 1,038 contigs were composed of ESTs from other grape tissues but not from berries (Table 4A). A total of 1,242 EST contigs matched by MPSS signatures were from ESTs found in only a single grape tissue; of these, 555 corresponded to berry-specific EST contigs. The remaining contigs were exclusively derived from leaves, flowers, petioles, stems, buds and even roots. The remaining 4,548 cDNA contigs and sequences were detected in two or more grape organs (Table 4A). Only three MPSS-matched EST contigs were found in all seven of the grape cDNA libraries. In a similar analysis of the EST singletons, the vast majority corresponded to transcripts previously observed exclusively in berry cDNA libraries, but only 207 were stage II berries (Table 4B). Among the contigs and singletons not previously associated with berry libraries were those derived from flower and leaf cDNA libraries. MPSS signatures provided valuable information to confirm the presence and relative transcriptional levels of transcripts. Many of these transcripts may have been previously mistakenly identified as tissue-specific based on EST data only because EST sequencing was not deep enough to detect these low abundance transcripts in different tissues. The MPSS data demonstrate that the inventory of genes in a given tissue is complex and there may be substantially more overlap in diverse tissues than previously characterized, and this can be identified only by sequencing ESTs at a very deep level.

Table 4 Grape ESTs derived from distinct tissue types matched by MPSS signatures (only from Vitis vinifera).
Full size table

One advantage of tag-based transcriptional profiling technologies such as ESTs, SAGE and MPSS is that the targets are not preselected prior to analysis. While the discovery rate of new transcripts using ESTs-based approaches is limited by the extent of sequencing effort and redundancy within a given cDNA library, unmatched or low abundance MPSS signatures could be used as primers for PCR based methods to expand the current set of known genes for Vitis [13]. There were 18,631 distinct 17-base MPSS signatures that did not match known grape EST sequences, of which 5,900 were both significant and reliable; these are most likely to represent novel genes not previously identified as transcribed or transcriptional variants. We tested this hypothesis by using available sequence of the grape genome, composed of 57,662 contigs containing 487,125,096 base pairs [14]. In total, 20,661 17-mer and 17,867 20-mer distinct MPSS signatures matched to genome contig sequences. Among these, there were 9,125 and 7,771 distinct 17-mer and 20-mer MPSS signatures that matched only genomic contigs and not ESTs. Taking the 17-mer signatures as the benchmark, the MPSS data reveal 44% more transcript diversity than recorded in the existing public EST resource.

In silico expression profiles resulting from EST (Table 5) and MPSS signature frequencies (Table 6) showed both differences and commonalities in the relative abundance of the top-ranked genes. For example, a common feature of both datasets is the relative high abundance of several chitinases, metallothionein-like and storage proteins, as well as a putative transcription factor and an elongation factor 1-α. On the other hand, two hexameric polyubiquitins and a plasma membrane aquaporin were among the top ranked genes based on MPSS signatures but not based on EST counts, and the opposite was true (present among top ESTs, not among MPSS signatures) for a non-specific lipid transfer protein A. A similar pattern emerges from the analysis of singleton ESTs that matched abundant MPSS signatures (Table 7). Among such singleton ESTs, there were transcripts related to cell wall modification (xyloglucan-specific fungal endoglucanase inhibitor protein and an extensin-like protein), abiotic/biotic stress factors (catalase and hydroperoxide oxidase), a eukaryotic translation initiation factor and several poorly annotated transcripts.

Table 5 Most highly expressed grape EST contigs in the grape berry stage II libraries, based on MPSS signature abundance.
Full size table
Table 6 Most highly expressed grape EST contigs in the grape berry stage II libraries based on EST frequency.
Full size table
Table 7 Top 20 grape EST singletons based on MPSS signature abundance.
Full size table

Significant differences were observed in the relative abundance of contigs from EST or MPSS signature counts. While a total of 195 contigs accounted for approximately 50% of the ESTs sequenced from the two berry SII libraries, only 10 contigs matched an identical proportion of the filtered MPSS signatures. The top 20 contigs ranked based on MPSS frequency accounted for 410,925 (56.7% of all sequences matching to EST contigs), suggesting a steeper curve and perhaps lower level of diversity in MPSS data. In contrast, the 20 most frequent contigs based on EST counts represented only 29.4% of the total EST for these two libraries.

As might be expected, MPSS signatures sequenced from V. vinifera berries stage II also matched several non-vinifera EST singletons and contigs in the Vitis Unigene set. Although the transcriptome of the non-vinifera species has been minimally characterized, a comparison of the top-ranked transcripts based on MPSS signature frequency (Tables 8 and 9) showed remarkable similarities between the different species.

Table 8 Most highly expressed grape EST contigs from non-vinifera libraries based on MPSS signature abundance.
Full size table
Table 9 Most highly expressed grape EST contigs from non-vinifera libraries based on MPSS signature abundance.
Full size table

A website for access to the grape MPSS data

To facilitate public access and utilization of the MPSS data, we developed a database and web-based interface [15]. The database and interface is a customized version of a previously described website [16]. Unlike the Arabidopsis or rice MPSS sites which utilize the complete genomic sequence of these species, our grape database focuses on EST contigs. This required the development of specialized tools and methods. For example, the incomplete nature of ESTs required a BLAST tool that would allow the user to identify the closest grape sequence to their gene of interest. The MPSS data can be accessed by entering the grape contig identifier or EST code, the MPSS signature sequence, the grape sequence of interest, or a list of contig identifiers. The data on transcriptional activity that this website provides may be used as the starting point for analyses of individual genes or gene families in grape.

Discussion

We have explored expression patterns at a specific stage in grape berry development by comparing and combining two tag-based methods: ESTs and MPSS. Both approaches described similar patterns of transcripts abundances, although there were some clear differences perhaps associated with the methods themselves. In principle, due to deeper sequencing, the MPSS data should provide a more thorough and quantitative representation of the absolute transcript population in terms of representation and relative abundance than that from ESTs [7, 11]. This is particularly true when the number of cDNA clones sequenced from any given library is low or for genes expressed at only low levels in the sampled tissues. For the EST frequency to represent the absolute transcript frequency, sequencing efforts must be large and sampling must be unbiased. The goal of achieving saturation for libraries constructed from a specific tissue may be overcome by combining library information available in public domain databases, if those resources are large enough. However, the different protocols used for library construction and EST sequencing, the lack of complete control of growing conditions, genotype and even standardized guidelines to describe a particular stage in development, makes it difficult to achieve unbiased sampling. On the other hand, MPSS analysis is also subject to bias. For example, some highly transcribed genes (based on EST frequency analysis) were unmatched by any MPSS signatures, possibly due to either the lack of a GATC site in the sequence or a technological artifact. The lack of suitable DpnII sites in some Arabidopsis transcripts is one source of negative results in MPSS transcriptional profiles compared against other high-throughput technologies [17]. In addition, MPSS substantially underestimates expression for signatures either containing the recognition site for the Type IIS restriction endonuclease BbvI (used in MPSS sequencing), or signatures containing certain four-nucleotide words in the sequencing frames [11]. The formerly high cost of tag-based methods limited biological replication as part of the experimental approach; such data would be highly desirable to determine the degree of biological variation and technical noise derived from these technologies [7]. This may be more achievable with the next generation of technologies as costs are reduced. The combined application of multiple approaches for transcriptional profiling is likely to provide the most robust determination of transcript levels.

In the grape MPSS dataset, when multiple signatures matched to one contig, these usually varied significantly in abundance. However, these data were consistent with the most abundant MPSS signature derived from the predominant form of the transcript among the ESTs [1]. An assessment of alternative transcript polyadenylation based on MPSS in diverse tissues and treatments could provide insight into this mechanism of gene regulation by identifying differentially terminated transcripts. The annotation and analysis of signatures matching multiple contigs is a more difficult task, but validation of these data could be performed by using microarrays with specifically designed probes to determine the relative expression of all matched genes, or by repeating the MPSS experiment using a different "anchoring enzyme" such as NlaIII (CATG) instead of DpnII (GATC).

The occurrence of genome-wide duplications may drive genome diversification and speciation in the plant kingdom [18]. Gene- and organ-specific silencing and unequal expression levels have been reported in upland cotton for homeologous genes resulting from whole genome polyploidization [19–21] and a similar phenomenon may be the cause of yellow-seeded commercial soybean cultivars [22]. The extent to which duplication-associated changes in gene expression may be playing a role in grapevine phenotypes is largely unknown. Due to the ancestral polyploid nature of the grape genome [23–25], duplication events leading to interactions or silencing among homeologous genes may have occurred. Evidence of extensive antisense expression was identified by comparing the ESTs and MPSS transcriptional profiling data. Initial whole transcriptome analysis in mammalian systems indicated that up to 20% of all transcripts formed sense-antisense (S/AS) pairs [26–31]. Recent analysis derived from a large scale mouse cDNA sequencing project [32] and a high resolution transcriptional map of human chromosomes [33] revealed that S/AS pairs exists for up to 72% and 50% of all mouse and human transcripts, respectively. S/AS frequencies observed in the berry transcriptome are similar to those reported in Arabidopsis, where approximately 22% of all known genes have tissue specific natural antisense transcript pairs [7]. Considering the unequal contribution of different genes and regions in the genome to the formation of S/AS pairs [32], whole transcriptome analysis would certainly provide a more accurate description of the extent of the phenomena in grapes than the one determined with a limited coverage of the transcriptome in this study.

Two distinct sources of native antisense expression have been identified: cis- and trans-encoded antisense [27–29]. The former correspond to transcripts derived from the opposite strand in the same genetic locus as the sense RNA. Cis-encoded antisense transcripts tend to have complete overlap with the sense strand forming long perfect match RNA duplexes [28]. Approximately 50% of sense-antisense pair categories in humans fell within this category [29]. Trans-encoded antisense transcripts derive from alternative loci and tend to have partial overlap with the sense strand of the original locus [27, 28]. The function of endogenous populations of dsRNA or small RNAs in grape remain to be elucidated with more detailed experiments, and this is best performed using short-read sequencing methods [34].

Tag-based transcriptional profiling approaches provide unique advantages for the discovery of novel expressed sequences. MPSS signatures derived from a specific stage in berry developmental revealed the existence of potentially 6,345 novel transcripts in grapes. These transcripts could be more fully identified to expand the set of known and experimentally verified Vitis genes either by PCR-based approaches [13], or ultimately aligning the signatures with grape genomic sequence. In the absence of full genome sequence information, PCR-based approaches may become particularly important for transcripts that are difficult to identify by means of EST-based approaches due to their low copy number or technical limitations of RNA-dependent cDNA synthesis. Whole genome sequencing of the V. vinifera genome, combined with data-rich tag-based (ESTs and MPSS signature frequencies) and microarray-based transcriptional data will greatly contribute to our understanding of the complex relationships between genome organization, transcriptional activity, and phenotypes. Because automated genome annotation systems are both error-prone and greatly improved with the incorporation of experimental data, the EST and MPSS data will prove invaluable in the coming years for gene discovery and the annotation of genomic sequences.

Conclusion

We have performed a complete transcriptional analysis of V. vinifera berries in transition to the ripening stage using MPSS combined with EST data. Approximately 30,000 distinct signatures, each representing a distinct transcript, were identified from the MPSS data and the signatures were mapped onto EST sequences. The number of MPSS signatures matching to one EST ranged from one to 16 and suggests the existence of numerous alternative transcripts in V. vinifera. In addition, a large set of MPSS signatures that matched to the anti-sense orientation ESTs was identified. Although the existence of antisense transcripts has been reported in many plant species, this is the first data to suggest the existence of antisense transcripts in V. vinifera. In addition to the signatures with EST matches, large numbers of MPSS signatures which do not match to ESTs were identified. While a small proportion could be due to sequencing errors, we believe the majority of these were mainly due to the low depth of sequence coverage in the current EST dataset; support for this interpretation derives from the fact that the proportion of signatures matching V. vinifera sequences was nearly doubled by incorporation of whole genome sequence data. High capacity, short read sequencing technologies, in particular next generation gigabase methods, have potential to contribute an important element to ongoing annotation of the genome sequence of V. vinifera. The grape MPSS data is accessible from University of Delaware MPSS website [1] and the EST data sets are available through UCDavis College of Agricultural and Environmental Sciences Genomics Facility (CGF) website [35].

Methods

Plant material and sampling procedures

The cDNA used for MPSS sequencing was constructed from stage II berries (green hard) sampled from field-grown V. vinifera cv. Cabernet Sauvignon, clone 8 vines located in the Tyree Teaching Vineyard, UC Davis, CA. Berries were sampled from multiple clusters and from different positions in individual clusters in order to ensure a representative sample. A sub-sample of berries at this stage was used to generate a cDNA library and expressed sequence tags (ESTs), as reported previously [4]. For additional details on sample handling and storage, see Goes da Silva et al., 2005.

MPSS data generation and analysis

All MPSS was performed essentially as described previously [5, 6], with the library produced and sequenced at Illumina, Inc. (formerly Solexa, Inc.; Hayward, CA). The raw and normalized MPSS data are available at University of Delaware MPSS website [1]. We compared MPSS signatures to the V. vinifera ESTs available at UC Davis CGF website [35] and assigned signatures to each sequence for which a perfect match was identified. The number of matches of a signature to the EST dataset was recorded as the "hits" for each signature. We merged the sequencing runs and calculate a single normalized abundance as reported earlier [11]. Contig orientation in the 5'-to-3' direction was performed using batch BLASTX search and the analysis of subject indexes of the first EST and last EST for each contig. Data analysis was conducted in MS Excel (Microsoft, Seattle, WA) and SAS V.8 statistical package (The SAS Institute, Cary, NC), or in a customized MySQL database [16] and figures in SigmaPlot version 8.0 (Systat Software Inc., San Jose, CA).