Establishing gene models from the Pinus pinaster genome using gene capture and BAC sequencing

Background In the era of DNA throughput sequencing, assembling and understanding gymnosperm mega-genomes remains a challenge. Although drafts of three conifer genomes have recently been published, this number is too low to understand the full complexity of conifer genomes. Using techniques focused on specific genes, gene models can be established that can aid in the assembly of gene-rich regions, and this information can be used to compare genomes and understand functional evolution. Results In this study, gene capture technology combined with BAC isolation and sequencing was used as an experimental approach to establish de novo gene structures without a reference genome. Probes were designed for 866 maritime pine transcripts to sequence genes captured from genomic DNA. The gene models were constructed using GeneAssembler, a new bioinformatic pipeline, which reconstructed over 82 % of the gene structures, and a high proportion (85 %) of the captured gene models contained sequences from the promoter regulatory region. In a parallel experiment, the P. pinaster BAC library was screened to isolate clones containing genes whose cDNA sequence were already available. BAC clones containing the asparagine synthetase, sucrose synthase and xyloglucan endotransglycosylase gene sequences were isolated and used in this study. The gene models derived from the gene capture approach were compared with the genomic sequences derived from the BAC clones. This combined approach is a particularly efficient way to capture the genomic structures of gene families with a small number of members. Conclusions The experimental approach used in this study is a valuable combined technique to study genomic gene structures in species for which a reference genome is unavailable. It can be used to establish exon/intron boundaries in unknown gene structures, to reconstruct incomplete genes and to obtain promoter sequences that can be used for transcriptional studies. A bioinformatics algorithm (GeneAssembler) is also provided as a Ruby gem for this class of analyses. Electronic supplementary material The online version of this article (doi:10.1186/s12864-016-2490-z) contains supplementary material, which is available to authorized users.


Background
Forests ecosystems play a fundamental role in the regulation of terrestrial carbon sinks and represent nearly 80 % of the world´s total plant biomass [1]. Conifers dominate a large part of the forests in the northern hemisphere, and they are intensively exploited as the primary source of wood for industrial purposes [2]. Conifers also exhibit unique characteristics among vascular plants, including: high genetic variability, long half-lives, seasonal survival, adaptation to secondary growth, and wood deposition among others [3]. Despite their economic and ecological importance, genomic studies of conifers have been hampered by the large size of their genomes, which range from 20 to 40 Gb, approximately 200 times the size of the Arabidopsis genome and approximately seven times the size of human genome [4]. However, recent technical advances in genomic sequencing have enabled the assembly of the Norway spruce [5], white spruce [6] and loblolly pine [7] genomes, and the sequencing of a number of additional species is underway [4,8]. Although these assemblies represent landmark in conifer genomics, technological challenges continue to face the assembly and annotation of conifer genomes; they are characterized by a proliferation of retrotransposons, highly diverged repetitive sequences, accumulation of non-coding regions and extensive gene duplication [4,8]. Also, large families of transposons and retrotransposons have been reported to occupy long stretches of the sequences in Pinus genomes [8,9].
The analysis of BAC clones has been the most common approach used for genome characterization and in hierarchical sequencing projects, such as the human genome [10] or other genomes without available references [11]. The screening of BAC libraries has been used to target gene-rich regions in white spruce, but the approach has proven to be very laborious because most clones contain the non-coding regions of the genes, which is expected due to the large size of conifer genomes [12].
An alternative to obtaining the gene sequences of large and complex genomes is to perform an enrichment step to isolate the genomic DNA sequences of interest that contain the coding regions of genes by massive parallel sequencing and use them for further analysis [13]. This system named "gene capture", uses rapid selective hybridization technique to obtain sequences of interest much more efficiently [14]. "-Gene capture-" has been widely used as a diagnostic tool for human whole exome analyses [15][16][17] but the use of the technique in plants has been much more limited [18,19].
In this work, we used "-gene capture-" to elucidate the target, gene-rich regions in the genome of the maritime pine (Pinus pinaster L. Aiton), a conifer species of great ecological and economic importance in Europe and for which whole-transcriptome resources are available [20,21]. To achieve this goal, 120-mer probes were designed from 866 tentative maritime pine transcripts, which include the probes for three characterized BAC clones as a control. These BAC clones were isolated by screening a maritime pine BAC library using specific cDNA probes [22] and then used as a reference for gene capture assays.
In this approach, megagametophyte calli haploid DNA from maritime pine was isolated, fractioned and bounded by a series of specific adapters for 454 sequencing. The captured genomic sequences were sequenced in an FLX-Titanium platform, and the reads were assembled and analyzed using the GeneAssembler bioinformatic pipeline to recover the gene models. This experimental approach also provided sequences for the proximal promoter region of the targeted genes. This can be used as initial information for genome walking to thoroughly characterize the cis elements contained in the regulatory region of these genes.

BAC clone isolation and characterization
A Pinus pinaster BAC library that had been previously established in pools [22] was used to screen for particular clones containing gene coding sequences for asparagine synthetase 1 (AS1) [23] cDNA [GenBank: HQ625490], sucrose synthase (SuSy) cDNA [GenBank: AJ309093] and xyloglucan endotransglycosylase (XET) cDNA [GenBank: FN824804]. A detailed protocol for the identification and isolation of these BAC clones is described in the Methods section and summarized in the Additional file 1: Figure  S1. As described in the Methods section, DNA was prepared from the purified BAC clones and fully sequenced, and the sequence assemblies for AS1, SuSy and XET were deposited in GenBank [GenBank: KP172187, GenBank: KP172194 and GenBank: KP172185 respectively]. Figure 1 depicts the corresponding BAC clone structures as single scaffolds. The sequences of the BAC clones were annotated and used to visualize the gene structure using GENote v.β. [23], which was used to detect the presence of the gene, its promoter location, the putative intron-exon pattern and the presence of transposable elements. The AS1 sequence in the BAC clone exactly matches the previously characterized maritime pine AS1 cDNA [24], and Fig. 1a shows the pattern of the BAC clone containing the AS1 gene assembled in a single scaffold of 46,111 bp. The AS1 gene is organized into 14 exons spanning a region of 3974 nucleotides, and the processed length without introns corresponds to a gene product with 1782 nucleotides that yields a protein with 593 amino acids. By comparing the Arabidopsis and poplar AS sequences in the databases, we determined the exon and intron structures, which are presented in Tables 1 and 2, respectively. The size and  number of exons contained in the BAC clone are closer to those of the ASN3 (At5g10240) gene from Arabidopsis. A total of 13 introns were identified, and their relative positions with respect to the coding region are well conserved compared to the Arabidopsis ASN3 gene as well as the P. trichocarpa Potri.005G075700 model ( Table 2). A similar analysis of the structure of the SuSy and XET genes was performed. The SuSy BAC clone is represented in a single scaffold of 59,327 bp (Fig. 1b), and the gene sequence is organized into 14 exons encoding a protein with 769 amino acids. The comparison to the SuSy gene models in other plants indicated that the 5é nd of the gene is missing in the pine BAC clone obtained for this study. The lengths and composition of the exon and intron are presented in Additional file 2: Table S1 and Additional file 3: Table S2, respectively, and a comparative study showed a high degree of conservation in the number and size of the exons in the maritime pine BAC model compared with the SuSy2 gene from Arabidopsis and the POPTRDRAFT_830445 gene from P. trichocarpa. In contrast, substantial differences in the number and lengths of the exons were found when compared with SuSy3 from Arabidopsis and POPTRDRAFT_826368 from Populus.
The XET BAC clone is represented in a single, 64,767-bp scaffold (Fig. 1c), and the sequence renders a model organized into 4 exons and 3 introns encoding a protein with 287 amino acids. The lengths and distribution of the exons and introns of the gene contained in the BAC clone are represented in Additional file 4: Table S3 and Additional file 5: Table S4, respectively. Two Arabidopsis gene models, XET 9 (At4g032210) and XET 3 (At3g25050), were closest to the pine gene sequence, but it was impossible to determine which is more similar to the gene contained in the pine BAC.
Considering that conifer genomes are characterized by retrotransposon proliferation and extensive regions containing repetitive elements, we performed a deeper study of the presence of these elements in the BAC clones. Through a comparison with the corresponding BACcontaining-gene available in Picea glauca (no BAC corresponding to XET was available), we analyzed the presence of repetitive elements. The content and type of the repetitions in both conifer BACs are summarized in Table 3. The percentage of bases that are part of the retroelements is quite similar in both SuSy BACs. But this percentage is significantly lower in the AS pine BAC, probably due to the different lengths of both BAC clones. In terms of the types of repeated elements, the retroelements, LTRs and Gypsy/Dir1 are the most represented in both SuSy BACs as well as in the P. glauca AS BAC with similar percentages of the total number of bases. DNA transposons, Simple repeats and low complexity elements contribute less to the size of the BAC.
Genomic DNA capture and gene model generation At the same time as the BAC library screening, we conducted a gene capture procedure using maritime pine haploid DNA [25]. Genomic DNA captured obtained using the SureSelect kit was sequenced in 454/Roche, and a total of 2,036,142 captured raw reads with an average size of 769 nt were cleaned with SeqTrimNext, producing 1,942,057 useful reads. These reads were assembled with MIRA3, yielding a total of 144,707 contigs and 305,396 "debris" reads. The contigs served to reduce the sequence space and to provide longer consensus sequences to facilitate the building of the gene model. The GeneAssembler pipeline was developed to build the most complex gene model that could code for each of the 866 selected full-length cDNAs from the pine transcriptome described in [20]. Starting with 144,707 contigs, the following steps were included: (i) gene assignment and contig filtering (Fig. 2a), (ii) contig clustering by gene (Fig. 2b) and (iii) gene model building to produce a gene model for the selected full-length proteins (Fig. 2c).
The best gene model is as follows: (i) the one that recovers more full-length seeding sequences; (ii) the one with the lowest overlap percentage of contigs at the exon level; and (iii) the one with the lowest fragmentation index, that is, the one obtained from the fewest number of contigs. The GeneAssembler gene model for each fulllength protein was saved in a gff3 file, which included the contigs. Additionally, a FASTA file was generated with the gene model sequences and an index relating each contig with its full-length protein.
Genomic sequences were recovered for the 866 maritime pine genes selected to design the probes, and this recovery was independent of the use of a reference genome for building the gene model. Accordingly, the reconstruction that was done without a reference genome was selected as it represents the highest mean recovery of each protein, approximately 82 %, and the lowest sequence redundancy. Most of the gene models with a high recovery percentage had between one to five exons, and particularly for those with one and two exons, there were many gene models with a recovery rate above 50 %. In addition, there were 20 gene models with a recovery rate over 100 %, but these were considered to be incorrect due to the presence of sequence repeats, which the algorithm was unable to handle. All of the generated contigs and models are available in the Pine Gene Capture database (PGC, http://www.scbi.uma.es/pgc/).
Because there is increasingly more information available about plant genomes, it is possible to make broad comparisons among them. Although it has been suggested that organisms with small genomes have smaller introns, studies of angiosperms indicate that this is not necessarily true in plants [26,27]. We performed an analysis of the size of the introns for our 866 reconstructed maritime pine models and compared the results to homologs in the moss Physcomitrella patens and three well-characterized angiosperm genomes: Arabidopsis thaliana, Oryza sativa and Populus trichocarpa (Fig. 3). The average length of the individual introns (in bp) was 197, 98, 143, 159 and 155 for P. patens, A. thaliana, O. sativa, P. trichocarpa and P. pinaster, respectively. The average intron length did not vary significantly between P. pinaster and the other four models chosen for this study (Fig. 3). However, the intron lengths were more heterogeneous in O. sativa and P. trichocarpa than in the other three species considered, at least for the 866 genes included in this study. Maritime pine did not appear to have a significantly larger intron.
Comparison of BAC and gene capture approaches to defining maritime pine gene models To further explore the quality of the genomic sequences derived from the gene capture approach, the structures of the AS, SuSy and XET genes were compared to those derived by BAC clone sequencing. Tables 1 and 2 show the distribution of exons and introns, respectively, for the AS gene obtained by BAC sequencing and that obtained by gene capture. After the analysis of the gene capture contigs, four different AS gene models were found to be similar to AS1, including itself; the new sequences were named AS3, AS4 and AS5. While cDNA contigs for the AS3 and AS4 genes were found in SustainPineDB (sp_v3.0_unigene3120 and sp_v3.0_unigene97582 respectively) [20], no contigs for AS5 were found in SustainPineDB or Genbank. However, the genomic sequences for the AS5 ortholog are present in the Pinus taeda genome [PtAS5, Congenie:lcl|scaf-fold870050.1]. Figure 4 presents the structures of the five AS genes of P. pinaster, with their relative exon/intron sizes. The AS1 and AS3 gene structures were completed, and the AS5 gene sequence was almost completed except for a part of the third intron. However, the AS4 structure was not completed due to a lack of information for the 3 rd to the 8 th introns. The gene capture assay also included probes for AS2, and the sequence was independently analyzed with respect to AS1. No crossovers were observed between AS1 and AS2 in the gene model building process, but the gene model for AS2 was incomplete as it lacked information for the 5 th to 11 th introns. The gene capture model that was generated fits to the AS3 gene as well as to AS1 and AS5. Although the comparison of the gene capture and BAC data indicates that the exon lengths and distributions from both methods agree well, the structure of the AS gene in the BAC clone was identical to AS1 gene model based on the gene capture. This provides additional confirmation that the isolated BAC contained a genomic fragment encoding the maritime pine AS1 protein [28].
The gene capture model for the SuSy gene agreed well with the structure of the BAC clone. The maritime pine model is closer to the SuSy2 gene from Arabidopsis and Populus with very well conserved number, length and exon position (Additional file 2: Table S1). Only exons 7 and 15 displayed small differences of 6 and 27 nt, respectively. The combination of BAC sequencing and gene capture allowed us to complete the first exon sequence, which can be considered an additional advantage of combining the two strategies.
The BAC clone and the gene capture model displayed very similar intron sizes; even intron 8, which contained a gap in the BAC assembly, was completed using the gene capture data (Additional file 3: Table S2). Although the distribution and length of exons was well conserved, an increase in the variability of the intron size was observed, which was similar to what was found for the AS gene family. In this case, the average size of the introns of the maritime pine gene was higher than for those genes of the other two species included in the comparison.
In the third part of this study, building the XET models, we did not include poplar genes, as we did for the AS1 and SuSy genes. The XET-related gene family is very large in plants, and as the poplar genome recently underwent a whole genome-wide duplication [29], the number of XET genes to compare to find the closest model to the pine gene would be very large. Therefore, considering the larger size of the gene family, we could not obtain a single model through gene capture. Instead, the two potential gene models that were generated are listed in Additional file 4: Table S3.

Gene capture of regulatory regions
An additional aim of this study was to test if the gene capture methodology could be extended to recover unknown sequences of the 5´end of the genes so that these sequences could be used: i) for a genome walking approach to obtain the promoter sequence or ii) directly for functional studies using the proximal promoter region of the genes.
To obtain the 5´end of the captured genes, contigs containing the first exon and having at least 100 nt upstream of the first exon were selected. Based on these standards, we obtained the frequency distribution of the contigs containing a noticeable upstream region of the genes (Fig. 5). The estimation from 866 gene models used in the gene capture approach was that 737 gene models contained a noteworthy 5´region sequence. Of those, 480 had a 5´region ranging 500 to 1000 nucleotides. To validate these results, the 5´upstream sequences of three genes included in our gene capture study were compared to the corresponding gene promoter sequences that had been previously cloned and functionally characterized [30,31]. For the glutamine synthetase (GS1a) gene, 518 nucleotides of the sp_v1.1_unigene26377 gene capture model overlapped with the GenBank promoter sequence [GenBank:AJ225121] (Additional file 6: Figure S2); 907 nucleotides for the phenylalanine ammonia lyase (PAL) gene promoter [GenBank:HE866754] overlapped with the sp_v1.1_unigene15094 model (Additional file 7: Figure  S3); and 630 nucleotides for the prephenate aminotransferase (PAT) gene promoter [GenBank:HE866755] matched the unigene all_rep_c3941_PAT gene capture model (Additional file 8: Figure S4).

Discussion
In this study, we present a strategy to generate maritime pine gene models, confirmed by the structure of available BAC clones, which demonstrates that gene capture is a powerful technology for establishing gene structures in species without reference genomes, such as the maritime pine.

Maritime pine gene structure
Generally, it is well accepted that a positive relationship exits between genome size and intron lengths in eukaryotes. In conifers, which have large genomes ranging from 18 to 35 Gbp, longer introns can contribute to large genome sizes [12]. In this study, we performed a deep comparison of exon/intron distribution for three pine genes against the Arabidopsis and poplar sequences from the databases.
The intron lengths for the AS1 gene (Table 2) are similar to those described for ASN1 and ASN3 in Arabidopsis, with the exception of introns 3 and 9, but none were greater than 426 bp, the length of intron 3 in the AS1 gene. In fact, when the AS1 intron sizes were compared to the poplar gene models, the medium intron size was larger in poplar than in Arabidopsis or pine (Fig. 6a). When intron size is compared in the structure of the SuSy gene, the distribution is also well conserved among the pine BAC, Arabidopsis SuSy2 and poplar POPTR-DRAFT_830445 genes. However, intron length is most variable among the three species, with longer introns for the maritime pine SuSy gene (Fig. 6b). For the XET gene, no significant differences were found in the size, distribution, number or length of the intron when compared to the Arabidopsis gene models. Moreover, a broad comparison of the intron sizes for the 866 gene models generated in this study (Fig. 3) showed that the medium intron size is in the range of that observed for other angiosperms models [18] (for more details, see gene models in the Pine Genome Capture Database PGCD: http://www.scbi.uma.es/pgc/).
The second feature contributing to the large conifer genomes is retrotransposon expansion, which contrasts with what has been described for large angiosperm genomes, where gene duplications and, polyploidization, as well as retrotransposons expansion, are the most common features contributing to genome size [32,33]. Thus, retrotransposon expansion can be of primary importance in explaining genome size in Pinus species [34]. We analyzed the presence of retroelements in the BAC clones included in this study by comparing two available BACs containing the AS1 and SuSy genes in P. glauca.
In terms of number/percentage, the LTR and Gypsy retrotransposons are most abundant in the SuSy and AS1 BACs from P. glauca and P. pinaster (Table 3). These findings suggest that retrotransposon expansion is a reasonable hypothesis to explain the large genome size in P. pinaster, as has been proposed for other conifers [34,35]. However, much more data on the retrotransposon families of maritime pine are needed to confirm this hypothesis.
Gene capture results are consistent with BAC clone structures; this outcome can be extended to other frames The algorithm developed for building gene models in this work has been used with plants as target organisms, but it is not restricted to them, its use with gene capture data derived from other organisms is possible.
A total of 866 gene models were obtained with our inhouse bioinformatics pipeline (GeneAssembler), with a median recovery percentage of 82 % from the original full-length sequence. However, there were several constraints, such as complex gene families, the presence of pseudogenes and the lack of a reference genome, that limited the final results. The algorithm designed in this study can address these problems, correctly resolving the gene models. It is also necessary to note is that the guided recovery using genes from model organisms, such as A. thaliana, P. trichocarpa, O. sativa or P. patens, could be useful when the coding sequence is incomplete. However, if enough transcript sequence information is available, the model can be built without any reference, which can be justified by taking into account that the contigs used to build the gene model come from the pool of single-hsp contigs. This pool has no defined exon-intron contig structure, and the contigs can be either exon fragments or putative pseudogenes.
The pipeline was also able to provide sequences for the 5´upstream region of most genes (85 %), and furthermore, 55 % of the gene models have at least one contig containing 500 to 1000 bp of the regulatory region of the gene. These findings indicate that the pipeline can be tuned to find a substantial portion of the gene promoter regions. These results validate the use of gene capture as a technology to provide information about the promoter region of genes in species where reference genomes are not available. This genomic information can be used in silico studies or to obtain the promoter region for functional studies of a significant number of maritime pine genes.

Exploring the organization of maritime pine gene families
Asparagine synthetase plays a specific role in nitrogen assimilation and storage in pine trees [28,36]. The data derived from gene capture described in this manuscript revealed that this enzyme is encoded by a small gene family in maritime pine with five members: AS1, AS2, AS3, AS4 and AS5. We also found their counterparts in the P. taeda genome (Fig. 7). It is remarkable that this approach allowed us to obtain the sequence of a new gene, AS5, which had not been previously identified by transcriptomic analyses. The nucleotide sequences of the five genes (Additional file 9: Figure S5) and the deduced AS polypeptides (Additional file 10: Figure S6) are well conserved, with the highest variability located at the 3é nd or the carboxy-terminus of the protein. The AS1, AS3, AS4 and AS5 genes/proteins are more closely related to each other than to AS2.
Phylogenetically, the angiosperm AS genes can be grouped into two classes: ASI and ASII [37,38]. The AS genes belonging to ASII class generally have 13 introns while the ASI class lacks intron five [38]. A phylogenetic tree was generated using the deduced amino acid sequences of the five AS pine genes from P. pinaster and P. taeda in addition to the AS sequences from Physcomitrella patens and different angiosperms used in previous AS phylogenetic analysis [38] (Fig. 7). Until now, the gymnosperm AS genes were considered to belong to the AS class II group [39]. But it can be concluded from our phylogenetic analysis that the conifer AS proteins do not belong to either AS class I or class II, which had been previously described for angiosperms. As shown in Fig. 7, the pine AS proteins group together separately from either of the ASI or ASII angiosperm classes. The models that are phylogenetically closer to angiosperms are the class II ASN3 gene from Arabidopsis and the class II Potri.005G075700 gene from poplar, which were included in our comparative study of exon/intron AS gene organization. It has previously been suggested that the angiosperm AS class II genes are closer to the ancestral AS gene [38], and our results support this hypothesis because all conifer AS genes conserve the 13 introns in their structure. Moreover, the conifer AS genes are closer to the bryophyte P. patens at sequence level.
The AS BAC clone sequenced in this work contains the genomic sequence of the AS1 gene, but due to the close similarity among the AS1, AS3 and AS5 pine genes (Tables 1 and 2, Fig. 4, Additional file 9: Figure S5 and Additional file 10: Figure S6), the model we obtained using the designed algorithm was predominantly AS3. A thorough examination of the contigs recovered by gene capture is included in Additional file 11: Table S6, and it illustrates that numerous contigs could be included in the AS3 instead of the AS1 model in this case. There were mainly truncated sequences, or intron-lacking sequences, that can be considered to be pseudogenes of AS3, and this indicates that the presence of numerous pseudogenes derived from a single gene can bias the results of this type of gene capture experiment. In the early stages of our study, we included this gene model because we thought it would be a good example of a pine gene belonging to a family with relatively few members to the XET gene and its family composition. In addition, the AS gene has a relatively large number of exons compared to the structure of the XET gene, which has fewer exons that can be used to detect differences in sequence recovery. However, we have found that this additional information related to the numerous pseudogenes in the AS family highlights the complexity of the maritime pine genome.

Conclusions
The comparison of the gene models produced by both approaches revealed that (i) processing the gene capture data with GeneAssembler was highly successful with 82 % of the gene models recovered even when a reference genome was not available. (ii) Successful and useful gene models can be obtained using probes designed from cDNAs; inference from of a single gene model would be limited in complex gene families with many members for which supplementary information would be required. (iii) Gene capture can serve to fill gaps in the gene structure established by BAC sequencing, as is the case with the SuSy gene model included in this study. (iv) Studying the gene models of different members belonging to the same family (e.g., AS) creates new possibilities for facilitating intra-and inter-comparative studies aimed at understanding function in the light of evolution.

PCR-based screening of a pooled BAC library
A 0.8x coverage pooled BAC library was previously prepared [22] as 83 glycerol stocks of E. coli pools, and each pool contained approximately 4000 distinct clones with an average size of 107 kbp. DNA of BAC pool was prepared by a modified alkaline lysis method using a previously described protocol [40].
For the primary PCR screening, specific genomic sequences of asparagine synthetase (AS), sucrose synthase (SuSy) genes or xyloglucan endotransglycosylase (XET) were amplified using primers designed against the available cDNA sequence (Additional file 12: Table S5). The PCR profile used for screening the pool library was as follows: 20 s denaturing at 94°C, 45 s annealing at 60°C and 45 s elongation at 72°C. This temperature profile was repeated for 35 cycles, and putative positive PCR product pools were on sequenced a CEQ 8000 automated capillary sequencer (Beckman Coulter, Barcelona Spain). After this, the original cell pools were titrated to determine the appropriate dilution to obtain~7000 colonies per Qtrays Genetix plate (24.2 × 24.2 cm), and the different clones were then individualized in 36 × 384-well plates using a QPIX2 ® (Genetix).
For the secondary screening, high-density filters were prepared on 22.2 × 22.2-cm nylon membranes using a 96pin gridding head on the Genetix QPIX2 ® robot. Each BAC filter was gridded to 6 rows × 4 columns in a single spot and distributed in six different fields. Membranes were incubated overnight at 37°C with the colony side facing up on LB/chloramphenicol (12.5 μg/ml) plates, and processing and hybridization of the filters was done as recommended in the QPix manual. [ 32 P]-labeled specific genomic sequences were used as probes. The hybridized membranes were exposed to a phosphorimaging screen (Fuji Imaging plate BAS-MS 2040) for 24 h at room temperature and scanned using an FLA-7000 Imaging System (Fujifilm). Using a grid map, the pattern of hybridization allowed us to identify the 384-well plate(s) and the plate address(es) of the positive clone(s).
For the subsequent steps, the cells of the positive clones were recovered and plated on LB agar with 12.5 μg/ml chloramphenicol, 90 μg/ml IPTG and 90 μg/ ml X-GAL. After a second round of PCR and sequencing, the presence of a specific genomic sequence was confirmed. The detailed workflow protocol is available in Additional file 1: Figure S1.

BAC DNA Sequencing library preparation and 454 parallel sequencing
The BAC insert size was estimated by pulse field gel electrophoresis (FIGE, Bio-rad Lab. Inc.). The 454 sequencing library was prepared with 6 μg of the BAC plasmid DNA using shotgun and standard long, pairedend sequencing kits, according to the specifications of Roche, the manufacturer. Sequencing libraries were quantified with 2100 BioAnalyzer (Agilent), processed by emulsion PCR and sequenced on a 454/Roche GS FLX as described in the manuals (Roche Diagnostics). The 24 L12 (AS), 25 M3 (XET) BACs libraries, were respectively sequenced in paired-end and shotgun pools using MIDs loaded in a 1/2 region, and BAC 26C15 library (SuSy) was sequenced in paired-end pools in a 1/4 region of 70 × 75 Picotiterplate (PTP). The pair-end BAC reads, which had been preprocessed by SeqTrimNext, were assembled using Newbler version 2.3 with the default parameters.

Capture array gene selection
The 91,086 full-length unigenes predicted from the P. pinaster transcriptome assembly from SustainPineDB [20] were used to select the genes for the DNA capture array. The full-length unigenes were filtered based on their GO terms [41], and the KEGG pathways [42] annotations mainly focused in metabolism. Because some unigenes contained annotations from several functions, this filter resulted in 1462 unigenes. Additionally, to reduce redundancy, the unigenes sharing the same ortholog from UniProt [43] were filtered to get only one gene per ortholog, reducing the selected number of unique unigenes to 1026. BLAST [44] best hits were used to assign the orthologous genes.
After the unigene selection was done, a better assembly of P. pinaster transcriptome was performed and published [20], so we searched for the genes from the 1026 unigenes in the SustainPine assembly version 1.1 (http:// www.scbi.uma.es/sustainpinedb/) using BLAST. Finally, 866 unigene sequences were selected to be included in the DNA capture array. As described previously, we added 3 genomic sequences from cloned BACs as control sequences, the asparagine synthetase [BAC at Gen-Bank: KP172187] from 5' UTR to the stop codon, a clone containing the sequence of a gene encoding a sucrose synthase [BAC at GenBank: KP172194] and a clone containing the xyloglucan endotransglycosylase [BAC at GenBank: KP172185]. The distribution of the unigenes used for the capture array in functional categories is shown in Additional file 13: Figure S7.

DNA capture array design
A Ruby custom script was used to design the probes for an Agilent SureSelect DNA Capture Array, and it was optimized to fit the 866 genes selected (see Capture array gene selection). In total, 56,667 probes of 120 nt in length were designed, and the probe density was increased in the first third of the transcript sequences (including the 5'UTR). In this way, all of the genes were represented in the array with 6x coverage in the first third of the gene sequence (using a tilling distance of 19 nt) and 4x coverage in the other two thirds (using a tilling distance of 29 nt). On average, 65 overlapping probes represented each gene, and they were uploaded to the Agilent eArray website (https://earray.chem.agilent.com/earray/) for the production of the array.

Haploid genome DNA preparation
Maritime pine cones were collected from Oria 6, a genotype of P. pinaster Aiton from the natural population Sierra de Oria (Almería, Spain), selected based on its response to extreme drought conditions. Cones were surface sterilized with 96 % ethanol for 20 min and airdried in a laminar flow cabinet before seed isolation. Haploid megagametophytes were isolated from sterilized seeds for tissue culture establishment [25]. For sequence capture, the P. pinaster A5 callus haploid line derived from the megagametophyte tissue was used. The DNA was extracted from the calli using a DNeasy Plant Mini Kit (Qiagen) and it was quantified with a Quant-it Pico-Green dsDNA Assay Kit (Invitrogen). The quality was assessed with an Agilent 2100 Bioanalyzer.
Targeted capture and 454 parallel sequencing Haploid genomic DNA from P. pinaster was captured using the Agilent SureSelect Target Enrichment System following the manufacturer´s protocols with minor modifications. Two micrograms of this DNA were fragmented to 1.5 kb in size and purified by gel extraction using a MinElute Gel extraction kit (Qiagen), and the quality of fragmentation and purification was assessed with an Agilent 2100 Bioanalyzer. Fragment ends were repaired, and RL adaptors (Roche) were ligated to the fragments, and the resulting adapter-ligated sample was purified using Agencourt AMPure XP beads (Beckmann Coulter). The DNA library was amplified by PCR and captured by hybridization at 65 o C for 24 h with the biotinylated RNA library "bait" (Agilent). Bound genomic DNA was purified with streptavidin-coated magnetic Dynabeads (Invitrogen) and re-amplified. Stratagene Herculase II enzyme (Agilent) was used for both PCR reactions, and the resulting captured library was purified using Agencourt AMPure XP beads (Beckmann Coulter) and was assessed with the Agilent 2100 Bioanalyzer. Finally the captured library was sequenced on Roche GS-FLX+ using a two-region gasket according to the manufacturer´s protocols.

Roche 454 data processing
The reads were preprocessed using the SeqTrimNext pipeline (http://www.scbi.uma.es/seqtrimnext) [45], which is available from the Plataforma Andaluza de Bioinformatica (University of Malaga, Spain). Low quality sequences, linkers, adaptors, vector fragments, organelle DNA, and contaminated sequences were removed, and the longest informative part of the read was retained, discarding sequences below 40 bp.

Data assembly and gene recovery criteria
Useful reads were assembled by the MIRA assembler, version 3 [46], to obtain the contigs for building the gene models. Full-length transcripts and the protein sequences were predicted using Full-Lengther Next [47] based on SustainPine 1.1 database [20] information.
The captured fragments were identified, and the gene models were generated using our own bioinformatic pipeline (GeneAssembler), which can be downloaded and installed on any Unix/Linux-based computer as a Ruby gem: https://rubygems.org/gems/gene_assembler (for a detailed description of the GeneAssembler pipeline see Additional file 14).
The gene sequences from Arabidopsis thaliana, Populus trichocarpa, Oryza sativa and Physcomitrella patens were downloaded from Phytozome 9.1 [48] to improve gene recovery, and an ortholog search was performed to enhance the gene model building strategy. A Blast X (by default parameters) [49] with the full-length proteins was performed against each set of genes. All gene matches were considered to be putative orthologs, which means that for each match, the exon-intron coordinate was retrieved.
The CLUSTALW program was used for sequence alignments [50]. The phylogenetic tree was constructed with full-length AS amino acid sequences using the neighbor-joining method [51] with 1000 bootstrap replications. The evolutionary distances were computed using the JTT matrix-based method [52] and are in the units of the number of amino acid substitutions per site. The rate variation among sites was modeled with a gamma distribution (shape parameter = 1). All ambiguous positions were removed for each sequence pair. The positions not presented in all the sequences were eliminated. Finally there were a total of 509 positions in the final dataset. All of these analyses were conducted in MEGA6 [53].