In this study, we constructed non-normalized cDNA libraries for each parental species as this will produce more full length transcripts for significant gene discovery. Each library was sequenced using one lane of a flow cell on the Illumina GAII platform using paired end protocols. We obtained 19,899,637 and 17,859,793 51 bp paired-end raw reads for A. auriculiformis and A. mangium, respectively. Filtering and conversion to FASTQ format resulted in 13,648,154 and 12,621,865 paired-end reads for A. auriculiformis and A. mangium respectively. After filtering of ribosomal RNA sequences, 51-57% of the reads remained with an average Phred score of 34 - 35.

The filtered reads were used to perform de novo assembly using a number of software such as Velvet 32 , SOAPdenovo 33 and Oases 34 , however, we found SOAPdenovo produced the longest assemblies despite using longer k-mers. We assessed different k-mer sizes and chose 29-mer to obtain a good tradeoff between assembly size and accuracy. De novo transcriptome assembly for A. auriculiformis (subsequently referred to as 'Aa') and A. mangium (subsequently referred to as 'Am') produced 42,217 and 35,759 contigs with an N50 contig size of 948 bp and 938 bp, a longest contig of 15,262 bp and 15,220 bp, and an average length of 496 bp and 498 bp respectively (Table 1). The sequencing depth was estimated to be 18.7 × and 18.3 × respectively. Blastx indicated that the longest contig of Aa and Am were homologs of the A. thaliana BIG; binding/ubiquitin-protein ligase/zinc ion binding gene (Figure 1A). This gene which is one of the longest genes in plants, was also reported in de novo transcriptome assembly of lettuce https://atgc-illumina.googlecode.com/files/PAG_2010_AKozik_V09.pdf.

To determine the similarity at the nucleotide level between the transcriptomes, we first mapped filtered reads to their corresponding de novo contigs before mapping each set of reads against the contigs obtained from the other species. To substantially increase the number of mappable reads, we mapped single-end reads using Bowtie -v setting allowing three mismatches. A total of 5,766,757 Aa single-end reads (37.24%) and 4,647,280 Am single-end reads (36.35%) mapped to their corresponding contigs. We observed only a small drop (roughly 15%) in the proportion of mappable reads from one Acacia species to the contigs of the other Acacia species indicating that the two transcriptomes shared a great deal of identity at the nucleotide level and are closely related.

The observation that a large proportion of filtered reads failed to map to the Acacia transcriptomes (> 60%) led us to investigate their origins by mapping to various genomes (Figure 1B). A further 5-6% of the reads mapped to the transcriptome of the other Acacia species probably due to differentially expressed transcripts. We discovered that approximately 14% of the reads mapped to mitochondrial and chloroplast genomes of A. thaliana, suggesting a significant amount of mitochondrial and chloroplast transcripts were sequenced. We suspect that mitochondrion sequences may not be assembled due to the highly heterozygous nature of genomes that were present in high copy number. We tried to map the remaining reads to several model plant genomes but found less than 10% mappable reads and no huge differences between these plant genomes. The number of reads mappable to the model legume, M. truncatula masked genome version Mt3.0 was 6.6-7.6%. The remaining ~36% of filtered reads were unmappable possibly due to several reasons. Some of these reads may be unique Acacia sequences from intergenic and intronic regions based on observation from Wang et al. 35 study that reported 40.75% of RNA-Seq reads from Aspergillus oryzae were located at these regions. Other reasons such as lack of Acacia genome information, poor quality reads and microbial contamination may have contributed to the large number of unmappable reads.

The monolignol biosynthesis pathway consists of several large protein families with members commonly known as isoforms. Isoform identification is challenging due to presence of many closely related superfamily members ("like") in the transcriptome, i.e. 27 "like" proteins of COMT, CCR and 4CL were observed in A. thaliana 36 . In this study, we found a total of 52 contigs in Aa and Am transcriptomes with E-value ≤ 1E-10 corresponding to all ten monolignol genes in A. thaliana. Gene identification using Blast alone often resulted in an overestimation of the total number of genes and isoforms. Shi et al. 37 reported 95 members of phenylpropanoid genes found in P. trichocarpa genome using Blastp (E-value ≤ 1E-3), however, many are proposed to be unrelated to monolignol biosynthesis pathway based on phylogenetic and expression analysis. Therefore, we tried to remove unrelated proteins by checking the conserved motifs which provide important clues in protein function and identity. We excluded contigs with low homology to A. thaliana monolignol genes (less than 55% identity) and we checked the remaining contigs for conserved amino acid motifs identified in previous studies 38 39 40 41 42 43 44 45 46 from the protein alignments (Additional File 1).

We were able to detect all ten genes involved in monolignol biosynthesis pathway, namely phenylalanine ammonia lyase (PAL), cinammate 4-hydroxylase (C4H), 4-coumarate 3-hydroxylase (C3H), caffeic acid O-methyltransferase (COMT), ferulate 5-hydroxylase (F5H), 4-coumarate:CoA ligase (4CL), hydroxycinnamoyl-CoA shikimate/quinatehydroxy-cinnamoyltransferase (HCT), caffeoyl CoA 3-O-methyltransferase (CCoAOMT), cinnamyl alcohol dehydrogenase (CAD), cinnamoyl Co-A reductase (CCR) compared to traditional EST sequencing in A. mangium 29 and A. auriculiformis × A. mangium hybrid 30 . We discovered more than one isoform for half of the genes which failed to be detected by EST sequencing. We identified a total of 18 isoforms for each species whereas 16 orthologous isoforms were shared in both species (Figure 2). All isoforms shared high identities with the corresponding A. thaliana genes where C3H shared the highest identity (68-85%), followed by PAL (71-84%), C4H (64-84%), CCoAOMT (64-83%), HCT (72-78%), CAD (58-76%), COMT (74%), CCR (73%), 4CL (57-71%) and F5H (59-62%). Our observations that orthologous isoforms of Aa and Am shared at least 99% similarity at both nucleotide and protein level while isoforms within the same family usually do not share an exact match of more than 16 nucleotides are important in determining the number of isoforms for both species.

The total assembled sequence lengths of the 36 isoforms ranged from 503 to 2,460 bp and only 14 contained complete open reading frame (ORF). No polyadenylation site was observed as expected because short polyA sequences failed to be assembled. One limitation of our sequence analysis is the presence of gap region in the contigs. Half of the assembled sequences contain gap regions with the total size range of 13 - 403 bp. These regions which were masked by Ns often occur at low coverage area where two contigs or mate pairs are connected during scaffolding. Although most de novo assemblers can estimate the size of the gap region, the predicted size is not always correct and sometimes resulting in inaccurate protein prediction. It is recommended to double-check the protein sequences by translating each fragment in gapped assemblies using other protein prediction software. Missing data poses a challenge to sequence comparison and analysis and therefore, gap filling by resequencing should be done in the future.

The total number of isoforms detected in this study is generally lower than those found in A. thaliana 36 and P. trichocarpa 37 . The identified isoforms possessed 99% DNA sequence identity with previously characterized isoforms from A. auriculiformis × A. mangium hybrid for the five isoforms that we examined, namely PAL, C4H, COMT, CCoAOMT, and CAD 31 . The high sequence similarity between of A. auriculiformis, A. mangium and their hybrids will allow more efficient cross amplification in gene isolation and characterization efforts. Given that several isoforms were only found in one species, greater sequencing depth is required for our analysis to overcome incomplete assemblies and sampling biases, previously observed in genomic sequences of Pseudomonas syringae strains 47 . Nevertheless, transcriptome sequencing of other tissues such as secondary xylem will provide more differentially expressed isoforms which can be new targets for the improvement of wood properties.

We found 1,306 Aa and 1,160 Am contigs with high sequence identity (E-value ≤ 1E-10) corresponding to 72 and 73 families out of 82 A. thaliana transcription factor families downloaded from PlnTFDB 48 . The five most abundant transcriptional gene families were WKRY, Orphans, PHD, HB and the MYB-related group. Several major classes of transcription factors involved in lignin and wood formation were found in both species (Figure 3A), generally in similar numbers of contigs, although the NAC family was substantially more abundant in Aa. Additionally, eight Aa and nine Am contigs were identified as class III HD-ZIP, a member of Homeobox (HB) family.

Some members of the R2R3-MYB family are known to be involved in controlling lignin deposition and secondary wall formation by interacting with other R2R3-MYB genes, activated by NAC transcription factor master switches and binding to AC elements 49 . The AC elements are cis-acting elements found in most promoters of monolignol biosynthetic genes 36 . In this study, we identified five contigs, two in Aa and three in Am, which are homologous to R2R3-MYBs regulating wood-related pathways in other plant species. In addition to R2R3-MYBs, NtLIM1 in tobacco had been proven to bind AC elements and its inhibition reduced lignin content 50 . We found one Am contig which was highly homologous to tobacco NtLIM1 with 86% identity.

Phylogenetic analysis of the Acacia R2R3-MYB proteins with wood-related R2R3-MYBs from A. thaliana and other plant species showed that they fall into three groups (Figure 3B). In group one, three Acacia R2R3-MYB proteins, namely AauMYB1, AmgMYB1 and AauMYB3 are close homologs of Arabidopsis MYB61 and Pine MYB8 while AauMYB1 and AmgMYB1 are orthologs. Pine MYB8 is a close homolog of MYB61 whose overexpression caused ectopic lignin deposition but the exact functions are yet to be known 51 52 . Only one Am R2R3-MYB protein (AmgMYB2) belongs to group two which is a close homolog to Arabidopsis MYB20 and MYB43. MYB20, MYB42 and MYB43 are activated by NAC master switches to regulate downstream MYB proteins in wood-related pathways 53 whereas MYB85 can induce secondary wall biosynthetic genes 54 . Another member of this group, PineMYB1 is able to bind AC elements 55 and is involved in secondary cell wall deposition 51 . AauMYB2 belongs to group three that clustered together with EgMYB1, AmMYB308, ZmMYB31 and ZmMYB42, indicates an important role in regulating the monolignol biosynthesis pathway. EgMYB1 binds AC element and represses the monolignol biosynthesis pathway 56 . AmMYB308, ZmMYB31 and ZmMYB42 have been shown to affect lignin content by regulating the expression of lignin genes 57 58 .

For non-model species like Acacia, microRNAs (miRNAs) can be identified from the transcriptome data based on homology searches against publicly available databases 59 . We searched for miRNAs by comparing our contigs to known plants miRNA stem-loop sequences downloaded from miRbase 60 . We found nine matching sequences from Aa corresponding to eight conserved families (miR319, 396, 162, 160, 168, 166, 172 and 159) and one recently identified family (miR2086). Four of these families (miR319, 396, 2086 and 166) were also found in Am. Most predicted miRNA genes such as miR319, miR396, miR162, miR166, miR168, miR172 are highly conserved in plants. The number of miRNAs detected in this study was lower compared to another study 61 because miRNAs are most abundant in leaves and flowers.

Blastx results showed that all primary transcripts except miR2086 have no significant hits to any protein-coding gene, suggesting that primary transcript sequences are less conserved in plants. Primary transcripts of miR159 and miR166 were removed from further analysis due to incomplete stem-loop structure and missing mature miRNA sequence. The presence of gap region in the stem loop sequences of miR396, miR160 and miR172 in Aa resulted in inaccurate stem-loop structure prediction. Therefore, PCR amplification and sequencing were carried out to fill up the gap. The secondary structures of miR319, miR396, miR2086, miR160, miR162, miR168 and miR172 predicted by Mfold were stable (Figure 4) and all except miR160 have high MFEI values (Table 2). miR2086 is a relatively new family highly expressed in the stem of M. truncatula 62 . Blastx indicated that both primary transcripts of miR2086 code for DNA glycosylase (E-value = 0.0). The predicted target of miR2086 is nodulin-like protein suggesting it might play a role in nitrogen fixing pathway. This family is predicted to be a legume-specific miRNA.

A total of 512 and 442 contigs in Aa and Am were predicted to be the targets for 135 and 134 miRNA families found in plants. Blastx results for the predicted targets of several highly conserved miRNAs are indicated in Table 3. We found known targets such as Auxin Response Factor, APETALA 2, F-box protein, Cc-NBS-LRR disease resistance genes and Heat Shock Protein for miRNA 160, 172, and 396. We predicted three wood-related genes, namely flavonol synthase-like, xyloglucan fucosyltransferase and glucan synthase-like genes to be the targets of miR170, miR172 and miR319, respectively, suggesting that miRNAs might be directly involved in the regulation of phenylpropanoid pathway and hemicellulose biosynthesis pathway. Glucan synthase is involved in the synthesis of xyloglucan which make up the β-1,4-glucan backbone while xyloglucan fucosyltransferase adds fructose sidechains to the backbone. Downregulation of flavonol synthase is predicted to redirect the carbon flux towards lignin biosynthesis as flavonoid biosynthesis uses 4-coumaroyl CoA as precursor. Functional analysis of these putative miRNA targets for potential role in wood formation should be studied in the future.

Single Nucleotide Polymorphisms (SNPs) are abundant markers that are suitable for a species with low genetic diversity such as A. mangium 63 . For a non-model species without genome sequences, we detected SNPs by mapping all the reads to de novo contigs as reference. We used only contigs at least 200 bp long to ensure sufficient flanking region for genotyping purposes. Although paired-end reads provide more accurate alignments, a large fraction of our contigs were too short to effectively utilize the paired end information, so we mapped the reads as single-end data, which substantially increased the number of mappable reads.

By using Bowtie default settings and allowing two mismatches, we detected a total of 30,837 Aa and 19,070 Am putative SNPs. After applying several filtering parameters to remove low coverage, low confidence, low minor frequency allele and multi-allelic SNPs, the putative SNPs number was further reduced to 16,648 and 9,335, respectively (Table 4). As expected, transition SNPs occur almost twice as frequently as transversion SNPs. One SNP was estimated to occur in every 1,123 bp and 1,704 bp in the Aa and Am transcriptomes, respectively. Although these SNPs represent only a portion of the Acacia transcriptome, this study has provided a better SNPs estimation compared to a previous study 31 which was based on the SNPs variation in two lignin genes. Further investigations are being carried to validate these SNPs which are useful for the construction of Acacia hybrid linkage map.

Plant materials were collected from one A. auriculiformis individual (AA6) and one A. mangium individual (AM20) growing in the Forest Research Institute Malaysia (FRIM), Kepong. AA6 and AM20 are parents of an Acacia hybrid mapping population. Both trees were about 5 years old at the time of sampling. The trees were propagated by marcotting the 4-year-old mother trees in FRIM's field station at Bidor, Perak and planting took place at Bukit Hari field plot in FRIM Kepong in 2004. Three different tissues, namely young stem, intermediate inner bark and old inner bark tissues were sampled. Young stem tissues consisted of ~5 cm of non-lignified stem, starting from the shoot tip. Inner bark tissues from intermediate and old developmental stages were sampled by cutting the largest branch on each tree into two halves. The upper half represented the intermediate stage while the lower half represented the old stage. The halves were further cut into disks about 3 cm each. The outer bark tissues were peeled off and we collected the inner bark tissues by separating it from the sapwood. The inner bark tissues are cut into smaller pieces and immediately frozen in liquid nitrogen and stored at -80°C until further use. RNA extraction was carried out using QIAGEN RNeasy Mini Kit for each tissue. A single RNA sample for each individual was generated from 20 μg RNA samples pooled from each of the three tissues. The quality and quantity of the RNA were evaluated using a Nanodrop ND-100 Spectrophotometer and Agilent Bioanalyzer. The RNA Integrity Number (RIN) value given by Agilent Bioanalyzer was greater than 7.5. RNase inhibitor was added to the RNA samples before sending to Canada's Michael Smith Genome Sciences Center where ribosomal RNA depletion using Invitrogen Ribominus Kit, cDNA synthesis and library construction were carried out. Each sample was subjected to one lane sequencing on an Illumina GAII platform.

Raw reads in QSEQ format were filtered and converted to FASTQ format using a AWK command. Ribosomal RNA was removed by mapping to A. thaliana 25S and 18S ribosomal RNA sequences using MUMMER 64 and filtered by a custom Python script (available upon request). The quality of the filtered reads was assessed using Python script htseq-qa from HTSeq package http://www-huber.embl.de/users/anders/HTSeq/doc/overview.html. The filtered reads were used in de novo assembly using SOAPdenovo v1.03 33 with all default settings except -R option was enabled and the insert size of 180-250 bp was used. SOAPdenovo performed scaffolding using paired-end read information and returned the assemblies in contigs and scaffolds. In this paper, we used the term "contigs" to refer to both contigs and scaffolds. The sequencing depth was estimated based on total length of the reads used in the assembly divided by total size of transcriptome assemblies. The contigs were searched against NCBI Non-redundant Database using Blastn and Blastx (E-value ≤ 1e-10). All the contigs were translated into protein sequences using FrameDP 65 . To compare transcriptomes of Aa and Am, we mapped single-end filtered reads from both species to both transcriptome assemblies separately using Bowtie-0.12.3 66 by allowing three mismatches and ignoring quality score. We applied an iterative mapping and filtering approach to the unmappable reads to find out their origins. Using Bowtie and allowing three mismatches, the single-end filtered reads were mapped to the both Acacia transcriptomes and other genomes as reference in the following order: its corresponding de novo contigs, de novo contigs from the other Acacia species, A. thaliana organelles (TAIR8 mitochondrial and chloroplast genomes) and M. truncatula genome (Mt3.0). After each alignment, mapped reads were removed using Bowtie's --un command and mapped to the next reference sequences. The remaining reads were considered as unmapped reads. The raw reads of Aa and Am were deposited on the NCBI Sequence Read Archive (SRA) with accession number SRR098315 and SRR098314.

All monolignol biosynthetic genes and isoforms were downloaded from the Arabidopsis Monolignol Biosynthesis Gene Families Database 67 . We searched the contigs for homologs of A. thaliana genes in monolignol biosynthesis pathway using local NCBI Blast-2.2.23+ blastx algorithm (E-value ≤ 1E-10). The protein sequences of the contigs were double-checked with ExPASy Translate Tool http://expasy.org/tools/dna.html and aligned with the corresponding A. thaliana genes using ClustalW2 68 . NCBI ORF finder 69 was used to search for Open Reading Frame (ORF). The protein sequences were checked for presence of conserved amino acid motifs to distinguish members within the same family. Protein identity shared between the isoforms and the closest A. thaliana isoforms were checked using EMBOSS Matcher 70 available at http://mobyle.pasteur.fr/cgi-bin/portal.py?#forms::matcher. The nucleotide sequences were trimmed and deposited at NCBI Transcriptome Shortgun Assembly (TSA) (Additional File 2). Protein and nucleotide sequences of the monolignol genes of Aa × Am hybrid, namely PAL, C4H, COMT, CCoAOMT and CAD were downloaded from Genbank [Genbank: AAW78382.1, AAY86361.1, ABD42947.1, ABX75853.1 and ABX75854.1] and aligned to the homologs in Aa and Am using ClustalW2.

We downloaded 82 transcription factor and transcriptional regulatory families of A. thaliana from PlnTFDB database 48 . We searched the translated contigs against this database using local NCBI Blast-2.2.23+ blastp algorithm (E-value ≤ 1E-10). We further analyzed several classes of wood-related transcription factors such as MYB, LIM and HD-ZIPIII. Protein sequences of R2R3-MYB [Genbank: CAE09058.1, CAE09057.1, NP_566467.2, NP_172425.2, NP_567390.4, NP_176797.1, NP_197163.1, ACA33851.1, AAQ62540.1, ABD60280.1, NP_196791.1, NP_567664.1, NP_001106009.1, NP_001105949.1, XP_002313303.1, NP_187463.1, XP_002299944.1 Swiss-Prot: P81395.1, P81393.1], LIM [Genbank: AT1G01780.1, AT1G10200, AT1G39900.1, AT2G45800.1, AT3G61230.1, AT3G55770.1] and HD-ZIPIII [Genbank: AY919616.1 -AY919623.1] from other plant species were downloaded from NCBI Protein Database. To generate the phylogenetic tree of R2R3-MYBs family, the full length amino acid sequences of R2R3-MYBs from other plant species and five homologous Acacia R2R3-MYBs, namely AauMYB1, AauMYB2, AauMYB3, AmgMYB1, AmgMYB2 [Genbank: JL052980, JL052981, JL052982, JL053003, JL053004, JL053005] were used. The protein sequences of homologous Acacia R2R3-MYBs were double-checked with ExPASy Translate Tool http://expasy.org/tools/dna.html. All the sequences were aligned using Bioedit ClustalW and the alignments were manually improved (Additional File 3). The unrooted tree was constructed using MEGA 5 71 with the neighbour-joining method and 1,000 bootstraps (Poisson model and pairwise deletion).

Stem-loop sequences of all major plant miRNAs were downloaded from miRbase database. The transcriptomes of Aa and Am were searched for potential stem-loop miRNAs using local NCBI Blast-2.2.23+ Blastn algorithm (E-value ≤ 1e-10). The matching sequences were trimmed to 1,000 bp before submitting to miRbase search tool to find stem-loop sequences and mature miRNAs. For miR396, miR160 and miR172 in Aa, PCR amplification and sequencing were carried out to find the complete stem-loop sequences. Primers flanking the gap region were designed based on primary transcript sequences (Additional file 4). RNA was extracted from inner bark tissues of the Aa individual (AA6) using Qiagen RNeasy Plant Mini kit. The quantity and quality of the total RNA was checked using Nanodrop ND-1000 Spectrophotometer and gel electrophoresis. 5 μg of total RNA were treated with DNase and converted to cDNA using Fermentas RevertAid Premium Reverse Transcriptase. The PCR reaction consists of 300 ng cDNA, 1 × PCR buffer, 2 mM MgCl₂, 0.2 mM dNTP, 0.25 μM of each primer and 1 U Vivantis Taq polymerase. The amplification profile consists of 2 min incubation at 94°C, followed by 35 cycles of 94°C for 30 s, 58°C for 30 s, 72°C for 30 s and a final extension of 72°C for 10 min. The specific PCR products were observed on 1% agarose gel stained with ethidium bromide and purified using Qiagen Gel Extraction kit. The purified PCR products were cloned into Promega pGem-T Easy Vectors and transformed into E. coli strain JM109. The transformed bacteria were spread on a LB plate containing amplicilin, IPTG and X-gal before overnight incubation. Five colonies for each plate were selected and grown overnight in LB broth containing amplicilin. PCR amplification using 1 μl of the culture pellet as DNA template were carried out to select three positive colonies for each primer pair. Plasmid was extracted using Qiagen Qiaprep Spin Miniprep kit and sent to First Base Laboratories Sdn. Bhd. (Malaysia) for forward and reverse sequencing using M13 primers. The sequence data were analyzed using Bioedit and gap regions were identified. Stem-loop sequences were extracted to predict secondary structures using Mfold 3.1 http://http://mfold.rna.albany.edu/?q=mfold/RNA-Folding-Form/. The secondary structures were examined visually and compared to the existing structures in the database. We used modified method from Zhang et al. 72 to identify miRNA genes except lower cutoff value of for Minimal Folding Energy Index (MFEI) was set. miRNA genes with complete stem-loop and mature miRNA sequences are available in miRbase database. We assigned prefixes aau- and amg- to represent A. auriculiformis and A. mangium. The miRNA targets were identified in Aa and Am transcriptomes by allowing 3 mismatches using a custom search in psRNAtarget http://bioinfo3.noble.org/psRNATarget/.

The filtered reads were mapped back to the reference using Bowtie-0.12.3 by allowing two mismatches. Only contigs at least 200 bp were used as reference. The generated SAM files were exported to Samtools 0.1.7 73 and converted to BAM format. We called SNPs using Samtools's Pileup command and removed any SNPs with a SNP score less than 20. The putative SNPs were further filtered using the following criteria: 1) Mapping and SNP score more than 100; 2) SNPs must be covered in at least 10 reads; 3) At least three non-reference alleles are present; 4) SNPs must be bi-allelic; 5) Minor allele frequency must be at least 5%; 6) Total frequency of major and minor allele must be at least 0.95. All filtering was done using Awk and Python scripts (available upon request).