Analysis of 4,664 high-quality sequence-finished poplar full-length cDNA clones and their utility for the discovery of genes responding to insect feeding

Background The genus Populus includes poplars, aspens and cottonwoods, which will be collectively referred to as poplars hereafter unless otherwise specified. Poplars are the dominant tree species in many forest ecosystems in the Northern Hemisphere and are of substantial economic value in plantation forestry. Poplar has been established as a model system for genomics studies of growth, development, and adaptation of woody perennial plants including secondary xylem formation, dormancy, adaptation to local environments, and biotic interactions. Results As part of the poplar genome sequencing project and the development of genomic resources for poplar, we have generated a full-length (FL)-cDNA collection using the biotinylated CAP trapper method. We constructed four FLcDNA libraries using RNA from xylem, phloem and cambium, and green shoot tips and leaves from the P. trichocarpa Nisqually-1 genotype, as well as insect-attacked leaves of the P. trichocarpa × P. deltoides hybrid. Following careful selection of candidate cDNA clones, we used a combined strategy of paired end reads and primer walking to generate a set of 4,664 high-accuracy, sequence-verified FLcDNAs, which clustered into 3,990 putative unique genes. Mapping FLcDNAs to the poplar genome sequence combined with BLAST comparisons to previously predicted protein coding sequences in the poplar genome identified 39 FLcDNAs that likely localize to gaps in the current genome sequence assembly. Another 173 FLcDNAs mapped to the genome sequence but were not included among the previously predicted genes in the poplar genome. Comparative sequence analysis against Arabidopsis thaliana and other species in the non-redundant database of GenBank revealed that 11.5% of the poplar FLcDNAs display no significant sequence similarity to other plant proteins. By mapping the poplar FLcDNAs against transcriptome data previously obtained with a 15.5 K cDNA microarray, we identified 153 FLcDNA clones for genes that were differentially expressed in poplar leaves attacked by forest tent caterpillars. Conclusion This study has generated a high-quality FLcDNA resource for poplar and the third largest FLcDNA collection published to date for any plant species. We successfully used the FLcDNA sequences to reassess gene prediction in the poplar genome sequence, perform comparative sequence annotation, and identify differentially expressed transcripts associated with defense against insects. The FLcDNA sequences will be essential to the ongoing curation and annotation of the poplar genome, in particular for targeting gaps in the current genome assembly and further improvement of gene predictions. The physical FLcDNA clones will serve as useful reagents for functional genomics research in areas such as analysis of gene functions in defense against insects and perennial growth. Sequences from this study have been deposited in NCBI GenBank under the accession numbers EF144175 to EF148838.

FLcDNA clones for genes that were differentially expressed in poplar leaves attacked by forest tent caterpillars.

Conclusion:
This study has generated a high-quality FLcDNA resource for poplar and the third largest FLcDNA collection published to date for any plant species. We successfully used the FLcDNA sequences to reassess gene prediction in the poplar genome sequence, perform comparative sequence annotation, and identify differentially expressed transcripts associated with defense against insects. The FLcDNA sequences will be essential to the ongoing curation and annotation of the poplar genome, in particular for targeting gaps in the current genome assembly and further improvement of gene predictions. The physical FLcDNA clones will serve as useful reagents for functional genomics research in areas such as analysis of gene functions in defense against insects and perennial growth. Sequences from this study have been deposited in NCBI GenBank under the accession numbers EF144175 to EF148838.

Background
Poplars are keystone tree species in several temperate forest ecosystems in the Northern Hemisphere. Poplars are also intensively cultivated in plantation forestry for the production of wood, pulp, and paper. Fast growing poplars can serve functions in phytoremediation, as a sink for carbon sequestration, and as a feedstock for biofuel production. Poplar has also been firmly established as a model research system for long-lived woody perennials (reviewed in [1]). Advances in functional genomics of poplar have been greatly enhanced by the availability of a high-quality genome sequence from P. trichocarpa (Nisqually-1; [2]), combined with comprehensive genetic [3][4][5][6] and physical genome [7] maps, as well as the availability of several platforms for transcriptome analysis [8][9][10][11] and genetic transformation. Large collections of expressed sequence tags (ESTs) have also been developed from a variety of poplar species and hybrids focussing on gene discovery in wood formation, dormancy, floral development and stress response [9,[11][12][13][14][15][16][17][18][19][20]. These short, singlepass EST reads have been a critical resource for gene discovery, genome annotation, and the construction of microarray platforms.
High-accuracy, sequence-verified FLcDNA sequences that span the entire protein-coding region of a given gene can advance comparative, functional, and structural genome analysis. For example, the accuracy of ab initio prediction of protein-coding regions in genome sequences is limited by the difficulty of finding islands of coding sequences within an ocean of non-coding DNA, and by the complexity of individual genes that may code for multiple peptides through alternative splicing. More robust approaches that unambiguously identify protein-coding regions in a genome sequence have used FLcDNA data, as demonstrated for example in Arabidopsis thaliana [21][22][23]. Despite their immense value, sequence-verified FLcDNA clones, where multiple passes verify the authenticity of reads, have not been generated in most plant species subjected to genomic analysis. Only a few large FLcDNA data sets have been generated for plants; namely for rice [24], Arabidopsis [25], and maize [26,27]. In contrast, as of September 2007, there were only 1,409 complete sequences from individual poplar FLcDNA clones in the non-redundant (NR) division of GenBank, in addition to a larger number of putative full-length sequences assembled from EST reads of multiple cDNA clones.
Our poplar FLcDNA program in the areas of forest health genomics and wood formation has focused on mechanisms of defense and resistance against insects and genes associated with xylem development. The forest tent caterpillar (Malacosoma disstria; FTC; [28]) is a major insect pest that threatens the productivity of natural and plantation forests. Poplars deploy an array of combined defense strategies against herbivores that can be grouped as chemical and physical defenses, direct and indirect defenses, constitutive and induced defenses, as well as local and systemic defenses (reviewed in [29]). Several recent studies have been conducted on the molecular mechanisms underlying inducible defenses against herbivores in poplar [11,18,[30][31][32][33][34][35][36][37].
In this paper, we report on the development of four FLcDNA libraries from poplar that served as the starting template for creating a substantial genomic resource of 4,664 sequence-verified FLcDNAs. We describe the overall structural features of these FLcDNA clones, annotation based on comparisons with other species, and the identification of 536 putative poplar-specific transcripts. Mapping the FLcDNA collection to the poplar genome sequence confirmed the overall high quality of the assembled genome sequence as well as the high quality of the FLcDNA resource, while also identifying 39 expressed poplar transcripts that appear to be derived from gap regions of the current genome sequence assembly and 173 new poplar genes that have not previously been identified in the genome assembly. By mapping 3,854 FLcDNAs to a poplar 15.5 K cDNA microarray platform and performing a comparison with existing transcriptome data, we identi-fied 153 FLcDNAs that match transcripts differentially expressed following insect attack by FTC on poplar leaves.

Selection and sequence finishing of FLcDNAs
FLcDNAs are defined as individual cDNA clones that contain the complete protein-coding sequence and at least partial 5' and 3' untranslated regions (UTRs) for a given transcript. This definition distinguishes bona fide FLcDNAs from in silico assembled EST sequences derived from multiple cDNA clones. In the latter case, it is possible that multiple, closely related genes or allelic variants of the same gene are assembled into a single consensus sequence. This problem is avoided when only sequences derived from the same physical FLcDNA clone are assembled. We prepared four FLcDNA libraries using the biotinylated CAP trapper method [38]. Three libraries constructed from xylem, phloem and cambium, and green shoot tips and leaves were derived from the P. trichocarpa Nisqually-1 genotype, for which the genome sequence has been reported [2]. An additional library was developed from the P. trichocarpa × P. deltoides hybrid H11-11 genotype using leaves subjected to FTC herbivory (Table 1).
To select candidate FLcDNAs for complete insert sequencing, we used a previously described bioinformatic pipeline for EST processing [11]. An initial set of 26,112 3' ESTs derived from FLcDNA libraries was combined with 81,407 3' ESTs from standard EST libraries [11] to generate a starting set of 107,519 3'-end ESTs, which resulted in 90,368 high-quality ESTs after filtering to remove sequences of low quality and contaminant sequences from yeast, bacteria and fungi. These sequences were then clustered using the CAP3 assembly program ([39]; assembly criteria: 95% identity, 40 bp window) to identify a set of 35,011 putative unique transcripts (PUTs; Figure 1). To maximize the capture of complete open reading frames (ORFs) and UTRs, only clones from full-length libraries were considered further. Using this strategy, we identified 5,926 cDNA candidate clones for full insert sequencing, which resulted in 4,664 sequence-verified poplar FLcDNA clones (see Additional file 1 and Figure 2). Inserts of 2,672 clones were completely sequenced using end reads only, with an average sequenced insert size of 735 ± 434 bp (average ± SD) and required an average of 4.5 ± 1.3 end reads to finish to high sequence quality. Using a combination of end reads and primer walking, inserts of an additional 1,992 clones were completely sequenced, with an average insert size of 1,308 ± 567 bp requiring 5.9 ± 2.8 end reads and 3.4 ± 1.8 internal primer reads per clone.
Analysis of the 4,664 FLcDNA sequences using the CAP3 clustering and assembly program ([39]; assembly criteria: 95% identity, 40 bp window) identified 3,505 FLcDNAs as unique singletons, with the remaining 1,159 grouping into 485 contigs, suggesting a total of 3,990 unique genes represented with finished FLcDNA sequences. The high percentage of unique transcripts (85.5%) within this set confirms the successful clone selection strategy (Figure 1) for establishing a low-redundancy clone set prior to sequence finishing.

Sequence quality and "full-length" assessment of poplar FLcDNAs
All 4,664 finished FLcDNAs achieved a minimum of Phred30 (i.e., one error in 10 3 bases) sequence quality at every base. The majority of FLcDNAs were of even higher quality with the minimum and average Phred values exceeding Phred45 (i.e., one error in 3 × 10 4 bases) and Phred80 (i.e., one error in 10 8 bases), respectively ( Figure  3). We predicted the complete protein-coding ORFs for all 4,664 FLcDNAs. The distribution of 5' UTR, ORF and 3' UTR lengths is illustrated in Figure 2 [also see Additional file 1]. The average sequenced FLcDNA length (from the beginning of the 5' UTR to the end of the polyA tail) was 1,045 ± 475 bp (mean ± SD), and ranged from 147 to 3,342 bp, whereas the average predicted ORF was 649 ± 429 bp and ranged from 33 to 2,935 bp. ORFs could not be detected (i.e., 30 [24]). Similarly, the aver-  [2], in close agreement with our results obtained with FLcDNAs.
To further assess the quality of the 4,664 poplar FLcDNAs, we performed reciprocal BLAST analysis against peptide sequences in The Arabidopsis Information Resource (TAIR) and against a set of 1,409 poplar sequences previously identified to be full-length (collected from the NR division of GenBank). Reciprocal BLAST analysis was performed with a stringent similarity threshold [% identity ≥ 50%; expect (E) value ≤ 1e -20 ] and identified 2,774 and 288 pairs, respectively, with Arabidopsis and previously published poplar FLcDNAs ( Figure 4). Of the 288 homologous poplar transcript pairs (i.e., previously published poplar sequences with high sequence similarity to FLcD-NAs reported in this study), 228 (79.2%) agreed well with regard to their ORF lengths and position of their start and stop codons (± ten amino acids; Figure 4). For the remaining pairs, the predicted 5' and/or 3' ORF ends did not match suggesting alternative start or stop codons, splice variants, or the possibility that one of the pair members was either truncated or had an incorrectly predicted ORF. When comparing the poplar FLcDNA collection to reciprocal matches from TAIR Arabidopsis peptides, we observed a similar number of 2,151 (77.5%) pairs with similar ORF lengths and positions of their starting methionine and stop codons (± ten amino acids; Figure  4). These results indicate the majority of the 4,664 poplar FLcDNAs represent true full-length transcripts with complete ORFs and correctly annotated start and stop codons.
In addition to BLAT analysis, we also compared the FLcD-NAs with the 45,555 predicted protein-coding gene loci identified in the genome sequence using BLASTN and observed 4,452 (95.5%) matched at an E value < 1e -50 (see Additional file 1). In order to identify possible sequence gaps in the 7.5× coverage genome, we searched for FLcD-NAs lacking a stringent BLAT to the genome match and a BLASTN match (E value ≥ 1e -50 ) to the predicted gene models. This approach identified only 39 candidates, of which 20 (0.4%) FLcDNAs also had a strong match by BLASTN (E value < 1e -50 ) to one or more poplar ESTs in the public domain, excluding ESTs reported in this study ( Table 2 and see Additional file 1), suggesting that these FLcDNAs represent expressed poplar genes that likely map to gap regions within the current genome draft. We cannot exclude the possibility that the remaining 19 FLcDNAs represent sequences from bacterial, fungal or insect species present on poplar tissues harvested for cDNA library construction, which were not filtered as contaminant sequences in our EST and FLcDNA processing procedures.
To identify expressed genes that were not predicted in the original genome annotation [2], we searched among the set of 4,487 FLcDNAs with a stringent BLAT match to the genome that did not match to any of the 45,555 predicted gene models (E value ≥ 1e -50 ). This analysis revealed 173 FLcDNAs, 79 of which also showed strong similarity (E value < 1e -50 ) to one or more poplar ESTs in the public domain (see Additional file 1), suggesting that these 79 FLcDNAs represent expressed genes and possibly noncoding RNAs, that were missed by gene prediction software during the annotation of the poplar genome. The fact that these poplar transcripts had been missed could be due in part to the relatively short lengths of these 79 FLcDNAs (average FLcDNA and predicted ORF length of 555 bp and 67 bp, respectively; see Additional file 1). 100% 99.9-95.0% 94.9-90.0% < 90%

Validation of sequence quality of FLcDNAs
Proportion of FLcDNA sequence length

Comparative sequence annotation of poplar FLcDNAs against Arabidopsis and other plants identifies proteins unique to poplar
Despite the growing research interest in poplar as a model angiosperm tree species and the recent completion of the poplar genome sequence, poplar still represents a difficult experimental system with relatively few functionally characterized proteins, compared to other established model systems such as Arabidopsis. Therefore, our effort of in silico annotation of poplar FLcDNAs was largely based on comparison with Arabidopsis together with the NR database of GenBank containing sequences from all plants, among other species. Using BLASTX, we found that the proportion of FLcDNAs with similarity to TAIR Arabidopsis proteins was 87.5% (4,081) at E value < 1e -05 and 55.5% (2,590) at E value < 1e -50 ( Figure 6A). Similar values were obtained when using BLASTX to compare against peptides from other species in the NR division of Gen-Bank (88.0% matches at E value < 1e -05 and 56.9% matches at E value < 1e -50 ) ( Figure 6A). As expected, the proportion of poplar FLcDNAs with sequence similarity to previously published poplar ESTs (i.e., ESTs available in the dbEST division of GenBank, excluding ESTs from this study) by BLASTN was very high, with 96.3% (4,496) and 94.3% (4,401) of FLcDNAs having matches with E values < 1e -05 and < 1e -50 , respectively ( Figure 6A).
To identify genes that are potentially unique to poplar, we next examined the relationship of sequence similarity among the poplar FLcDNAs and best matching sequences in the TAIR Arabidopsis proteins, other NR database proteins (which includes all plant species), and previously published poplar EST datasets. Of the 4,664 poplar FLcD-NAs, 3,994 (85.6%) had at least low sequence similarity to sequences in all three databases (E values < 1e -05 ; Figure  6B). Only 95 FLcDNAs had no similarity (E values ≥ 1e -05 ) to sequences in any of these databases; however, 87 of these strongly matched to the poplar genome using BLAT  (2) the reciprocal TBLASTN analysis identified the same poplar FLcDNA with a score value equal to or within 10% of the top match. ORF lengths for Arabidopsis and public poplar sequences were extracted from the TAIR and NR records, respectively, and poplar ORF lengths from this study were predicted using either the EMBOSS getorf or in-house BLAST-aided programs (see Figure 2 legend). The greyscale shading of each hexagon represents poplar FLcDNA abundance. ORF lengths for three Arabidopsis-poplar pairs and eight homologous poplar transcript pairs differed by more than 500 aa and are not included in the figure.  Figure 6B and see Additional file 1). These poplar FLcDNAs could represent genes that were gained and then rapidly diverged in sequence since the recent whole genome duplication in poplar, or they may also represent non-coding RNAs or small peptides in poplar that share limited sequence similarity with other plants. The fact that these putative poplar-specific FLcDNAs do not share similarity with existing plant sequence data may also reflect the limited availability of sequence data from Salicaceae species closely related to poplar in the current NR database.  Figure 5 Mapping FLcDNAs to the poplar genome. 4,664 poplar FLcDNAs were aligned to the genome using BLAT with default parameters (match length ≥ 11 bp, BLAT score ≥ 30, sequence identity ≥ 90%). Prior to alignment, the 5' second strand primer adaptor sequences (SSPA) and polyA tails were removed. Among 4,642 poplar FLcDNAs that exceeded the minimal criteria for a match to the genome, 3,847 mapped to chromosomes whereas the remainder mapped to scaffold segments. Colored bars indicate the cDNA library of origin for those FLcDNAs mapping to one of the 19 poplar chromosomes. Applying a higher stringency threshold (sequence identity ≥ 95%, alignment coverage ≥ 95%), 4,487 or 96.2% of poplar FLcDNAs could be mapped to the genome.

Annotation of poplar FLcDNA transcripts affected by FTC herbivory
A major emphasis of the program that motivated the development and analysis of poplar FLcDNAs is the discovery of genes affected by insect attack. To identify herbivore-responsive genes among the poplar FLcDNAs, we first mapped the FLcDNA set onto a poplar 15.5 K microarray based on BLASTN comparison to ESTs spotted on the array. This microarray platform was previously used for profiling of the poplar leaf transcriptome affected by FTC larvae feeding [11]. Using a stringent similarity threshold of ≥ 95% identity over ≥ 95% alignment coverage, we identified 3,854 FLcDNAs that matched with 3,974 EST elements on the array (see Additional file 2). Although we did observe some cases of individual FLcD-NAs mapping to multiple array elements, as well as mul-tiple FLcDNAs mapping to the same array element, it should be noted that the in silico match stringency applied here is likely higher than the capability of cDNA microarrays to discriminate among highly similar transcripts by actual DNA hybridization. Next, we identified poplar FLcDNAs with a role in the response to insect attack by screening the 3,854 FLcDNAs against existing transcriptome data of differentially expressed (DE) genes in leaves that were exposed for 24 hours to FTC feeding [11]. This approach resulted in the identification of 129 and 24 FLcDNAs that were induced or repressed, respectively, in FTC-treated leaves compared to untreated control leaves (Tables 3 and 4) using the DE criteria of fold-change ≥ 2.0fold, P value < 0.05 and Q value < 0.05. A complete list of expression data is provided [see Additional file 2]. Each of the 153 FLcDNAs was translated and evaluated for the presence of ORFs, and annotation was assigned based on manual examination of the highest scoring and most informative BLASTX matches in NR. Among FTC-induced transcripts represented with FLcD-NAs, we identified a large number of defense-related and stress response proteins such as chitinases, Kunitz protease inhibitors, dehydrins, beta-1,3-glucanases, pathogenesis related protein PR-1, and glutathione-Stransferase (Table 3). Several classes of transcription factors (TFs) were also strongly affected by FTC feeding such as bZIP domain TFs, NAC domain TFs, NAM domain TFs and ethylene response factor TFs. A number of genes asso-ciated with signaling were also strongly affected by FTC feeding, including allene oxide cyclase involved in jasmonate formation and calreticulin associated with calcium signaling. We also observed a substantial number of FLcDNAs annotated as involved in phenolic metabolism, particularly flavonoid biosynthesis, including isoflavone reductase, EPSP synthase, flavonoid 3-O-glycosyl transferase and flavanone 3-hydroxylase, along with several cytochrome P450s of unknown function (Table 3). Among the FTC-repressed transcripts represented with FLcDNAs, we observed photosystem II proteins associated with photosynthesis, malate dehydrogenase and thiamine biosynthesis enzyme associated with primary metabolism, several zinc finger TFs, and stress-responsive proteins such as small heat shock and universal stress proteins (Table 4). Twenty two of the 153 FTC-responsive genes represented with FLcDNAs matched to hypothetical proteins of unknown function and nine have no obvious similarity to any proteins in the NR database.

Discussion
Previous studies using the biotinylated CAP trapper method for FLcDNA library construction have demonstrated this technique to be highly effective for capturing predominantly true full-length clones in large-scale projects [24,25,27]. In this study, we generated a set of 4,664 FLcDNAs, which represents the third largest plant FLcDNA resource published to date, behind only Arabidopsis and rice. CAP3 clustering and assembly indicates that more than 85% of the FLcDNAs are non-redundant within this collection. The average sequence length, ORF and UTR sizes of the poplar FLcDNAs were comparable to those observed with the CAP trapper-derived FLcDNA collections for maize [27], Arabidopsis [40] and rice [24], and were also very similar to the ab initio predicted reference genes in the poplar genome sequence [2]. Applying a reciprocal BLAST strategy, we demonstrated that among FLcDNAs with high sequence similarity to known Arabidopsis peptides and/or previously published poplar FLcDNAs, nearly 80% had similar ORF lengths and starting methionine and stop codon positions. Collectively, these data show that the poplar FLcDNA libraries are of high quality and that our clone selection strategy combined with the CAP trapper method was effective in capturing bona fide FLcDNAs from poplar.
Comparison of poplar FLcDNAs and the poplar genome sequence assembly confirmed both the overall high accuracy of the current genome assembly, as well as the quality of the FLcDNA resource described here. However, as has been previously demonstrated with efforts to identify the complete catalogue of genes in Arabidopsis and rice, gene prediction and genome assembly is an iterative process. The results reported here for the mapping of FLcDNAs to the poplar genome sequence reveal opportunities for    improvement of the genome sequence assembly (i.e., targeting apparent gaps for re-sequencing), as well as opportunities to further improve tools for the in silico prediction of genes. To address the discovery of apparent gaps in the genome assembly, the availability of 39 FLcDNAs that are not covered in the current assembly could be used to target BAC clones for re-sequencing and filling of gap regions. Similarly, the discovery of 173 FLcDNAs that do not have corresponding gene predictions in the current genome annotation may provide an opportunity to further improve gene prediction tools for poplar. Algorithms used for gene prediction in the poplar genome sequence assembly could be tested with these 173 FLcDNAs to find out why they may have initially been missed. If this leads to an improvement of prediction tools, the assembled genome sequence could be tested with the modified tools to identify additional genes.
The comparative sequence annotation of poplar FLcDNAs against Arabidopsis, the NR database, and previously pub-lished poplar ESTs revealed that ca. 88% of poplar FLcD-NAs showed similarity to sequences in Arabidopsis or other plants. Many of the ca. 11.5% of poplar FLcDNAs without significant sequence similarity in Arabidopsis or other plants are supported with evidence of gene expression in the form of previously published poplar ESTs and matching the poplar genome sequence, thus excluding the possibility that they are artifacts of cDNA library construction. The discovery of poplar FLcDNAs without matches in other plant species is also in agreement with previous analysis of the poplar genome sequence where 11% of predicted proteins had no similarity to proteins in the NR database and 12% had no similarity to Arabidopsis proteins [2]. For comparison, only 64% of the 28,444 ORFs derived from rice FLcDNAs showed significant similarity to coding sequences predicted from the Arabidopsis genome and conversely, only 75% of Arabidopsis coding sequences had similarity to rice FLcDNAs [24]. These findings suggest that a substantial proportion of protein-coding sequences are not conserved among all plant species. The putative poplar-specific genes could be the product of past local or whole genome duplications in the lineage that led to extant poplar species [2,43] followed by sequence divergence [44,45]. Furthermore, ca. 2% of poplar FLcDNAs did not contain a predicted ORF suggesting these putative poplar-specific genes likely encode noncoding RNAs (i.e., rRNAs, tRNAs, snoRNAs etc.).

Conclusion
We developed a large FLcDNA resource of high sequence quality and low-level redundancy that facilitated the discovery of a substantial number of genes not present among the published sequences of other plant species, and that also facilitated the discovery of several hundred insect-affected genes in the poplar leaf transcriptome that were represented by FLcDNAs. The newly established poplar FLcDNA resource will be valuable for further improvement of the poplar genome assembly, annotation of protein-coding regions, and for functional and comparative analysis of poplar genes. Specifically, the identification of FLcDNAs that are not covered in the current genome assembly or that were not predicted during the genome annotation provides opportunities to further refine the current genome assembly. The availability of a large collection of FLcDNAs that show altered gene expression following insect herbivory affords more rapid characterization of the role of these genes in poplar biotic interactions.

Full-length cDNA libraries
Plant materials used in the construction of cDNA libraries are described in Table 1 Sequences were also compared to the GenBank NR database using BLASTX. Top ranked BLAST hits involving other non-plant species and with E values < 1e -10 were classified as contaminants and removed prior to EST assembly.

Selection of candidate FLcDNA clones and sequencing strategy
All 3'-end ESTs remaining after filtering were clustered and assembled using CAP3 [39] (assembly criteria: 95% identity, 40 bp window). The resulting contigs and singletons were defined as the PUT set. PUTs with a cDNA clone from a FLcDNA library were selected as candidates for complete insert sequencing ( Figure 1). Candidate clones from FLcDNA libraries were single-pass sequenced from both 3'-and 5'-ends and both sequences were used for subsequent clone selection. Next, clones were screened for the presence of a polyA tail (3'-end EST) and the secondstrand primer adaptor (SSPA; 5'-ACTAGTTTAATTAAAT-TAATCCCCCCCCCCC-3'; 5'-end EST). Clones lacking either of these features were eliminated. A polyA tail was defined as at least 12 consecutive, or 14 of 15 "A" residues within the last 30 nt of the 3'-end EST (5' to 3'). The presence of the SSPA was detected using the Needleman-Wunsch algorithm limiting the search to the first 30 nt of the 5'-end EST (5' to 3'). The SSPA was defined as eight consecutive "C" residues and a > 80% match to the remaining sequence (5'-ACTAGTTTAATTAAATTAAT-3'). In each case, the algorithms used to detect the 5' and 3' clone features were set to produce maximal sensitivity while maintaining a 0% false positive rate, as determined using test data sets. Candidate clones for which either of the initial 5'-end or 3'-end EST reads had a Phred20 quality length of < 100 nt were also excluded. Finally, candidate clones were compared to poplar ESTs in the public domain (excluding ESTs from this collection; BLASTN match E < 1e -80 ) to identify candidate FLcDNAs potentially truncated at the 5' end of the transcript relative to a matching EST. Any clone with a 5' end that was > 100 nt shorter than the matching public EST was excluded. For each PUT represented by multiple candidate clones after filtering, the clone with the longest 5' sequence was selected for complete insert sequencing. Insert sizing performed on 4,848 of 5,926 candidate clones using colony PCR with vector primers and standard gel electrophoresis revealed an average insert size of ca. 1,085 bp. Based on this information, a sequencing strategy emphasizing the use of end reads was chosen.

Sequence finishing of FLcDNA clones
FLcDNA clones selected for complete sequence finishing were rearrayed into 384-well plates, followed by an additional round of 5'-end and 3'-end sequencing using vector primers. All end reads from an individual clone were then assembled using PHRAP (version 0990329) [48-50]. To meet our sequence quality criteria, the resulting clone consensus sequence was required to achieve a minimum average score of Phred35, with each base position having a minimum score of Phred30. Each base position also required at least two sequence reads, of minimum Phred20, that were in agreement with the consensus sequence (i.e., no high-quality discrepancies). Clones that did not meet these finishing criteria after two rounds of end read sequencing were then subjected to successive rounds of sequencing using custom primers designed using the Consed graphical tool version 14 [54] until the required quality levels were achieved. Regardless of the finishing strategy, all clones that did not meet the minimum finishing criteria according to an automated pipeline were flagged for manual examination. Clones were aborted if they were manually verified to lack the minimum finishing criteria after three rounds of custom primer design, were identified as chimeric sequences, or were refractory to sequence finishing due to the presence of a "hard-stop". FLcDNA sequences have been deposited in the NR division of GenBank [EF144175 to EF148838].

Gene expression meta-analysis of FLcDNAs
Poplar FLcDNA sequences were mapped to a cDNA microarray containing 15,496 poplar ESTs [ [11]; Gene Expression Omnibus (GEO) platform number GPL5921] using BLASTN with a stringent threshold of ≥ 95% identity over ≥ 95% of alignment coverage. To identify FLcD-NAs that were DE following FTC feeding, FLcDNAs mapping to the microarray were matched to an existing microarray dataset that examined gene expression in hybrid poplar leaves 24 hours after continuous FTC feeding ( [11]; GEO series number GSE9522).