The complete mitochondrial genome of Asian arowana was sequenced with shotgun sequencing method (min. 6X, average 9X coverage). Its total size was found to be ca. 16,651 bp [GenBank:DQ023143]. Except the mitochondrial control region the size of Asian arowana mitochondrial genome was found to be similar to that of silver arowana, butterfly fish and goldeneye 9 [see Additional file 1 for the exact sizes]. The GC content of Asian arowana mitochondrial genome was 46.1%, the highest among mitochondrial genomes of all Osteoglossiformes available in Genbank (silver arowana – 43%, butterfly fish – 39% and goldeneye – 42%).

On the whole, the structure of the Asian arowana mitochondrial genome is very similar to that of silver arowana, butterfly fish and bichir [see Additional file 2]. The number and order of genes in the Asian arowana mitogenome [see Additional file 3] were found to be the same as common vertebrate form 1. It contains 24 RNA and 13 protein-coding genes: 7 subunits of the NADH ubiquinone oxidoreductase complex (nad1-6 and nad4L), 3 subunits of the cytochrome c oxidase complex (cox1-3), a single subunit of the ubiquinol cytochrome c oxidoreductase complex (cob), 2 subunits of ATPase (atp6 and atp6), 2 ribosomal RNA (rrnL and rrnS) and 22 transfer RNA (trn) genes. The non-coding control regions situated between the trnP and trnF genes contain the heavy strand origin of replication (O_H). A smaller control region containing the putative light strand origin of replication (O_L) was found between trnW and trnY genes.

Eleven potential overlaps between genes have been observed in the Asian arowana mitogenome. The longest one (10 bp) involving the two ATPase genes appears to be common in most vertebrate mitochondrial genome, and its size in fish (7–10 bp 16) is smaller than that in mammals (40–46 bp; 2). The second largest overlap is 7 bp long, (between nad4 and nad4L genes), whereas the remaining nine were in the size range of 1–5 bp.

The Asian arowana mtDNA's heavy strand control region, also known as D-loop, contains O_H and is ca 980 bp long. Similar to typical vertebrate mitogenomes, this non-coding region can be divided into three different domains 1718 (Figure 1A). Domain I which is 400 bp long, consists of a termination associated sequence (TAS: TACATAAATTG) 19 and several copies of a previously described conserved palindromic motif without any known function 20. A 37 bp tandem repeat array, suggested to be involved in the regulation of mitochondrial genome replication by forming a thermostable "hairpin" 21, was also found in this domain (see next section for details). Domain II – commonly known as the central conserved block – covering the 401–641 bp stretch in the control region, showed high similarity to domain II of rainbow trout 22 and sturgeon 21 (data not shown). In domain III, a TA-dinucleotide microsatellite repeat was present in all the six individuals from which the control region was sequenced. Two conserved sequence blocks (CSB1; 724–742 bp and CSB3; 813–839 bp) found in this domain showed high similarity to CSBs detected earlier in other species 23 whereas CSB2 described earlier in teleosts 24 was not found.

A smaller control region (34 bp) for O_L exhibited high sequence similarity to the corresponding region in silver arowana, bichir and butterfly fish (data not shown). The AT content of Asian arowana O_L which was 35.3%, is similar to that of butterfly fish but higher than in silver arowana (31.4%) and much lower than in bichir (44%). The secondary structure of O_L was suggested earlier to regulate light strand replication 25. In Asian arowana this secondary structure consists of a perfect 9 bp stem (CCTCCCGCC/GGAGGGCGG) and loop structure. Despite the fact that the control region is the most variable region in animal mtDNAs, most part of the stem (TCCCGCC and AGGCGGA) was found to be conserved in the mitogenomic O_L of several fish species (including the Asian arowana) and even mammalian ones 26.

The mtDNA of all six Asian arowana individuals tested possess a heteroplasmic tandem repeat array in domain I (Figure 1B). The tandem repeat arrays in the six individual fishes sequenced contained 3 to 6 repeat units, resulting in variable length of the heavy strand control region (976 to 1094 bp long). A partial repeat unit could also be found at the beginning and at the end of the array indicating that it might have been formed by replication slippage 2127.

The tandem repeat units were highly similar with only a few base substitutions (Figure 1B). Each repeat unit in the array was 37 bp long (

TACAT

ATGTA

Another type of repeat – a TA-type dinucleotide microsatellite – was present at the opposite end of the heavy strand control region in domain III (Figure 1B). The number of TA core units ranged from 8 to 10 in the six individuals sequenced. Although both tandem repeats alone (e.g 30313233) or microsatellites in the tandem repeat array 34 has been reported earlier in the heavy strand control regions of some species, to our knowledge no one has reported the existence of both types of repeats on the same heavy strand control region.

The start codon usage in the Asian arowana mitogenome was found to be the same as that of zebrafish 16. All but one of the 13 protein coding genes began with the orthodox ATG start codon, only cox1 used GTG start codon [see Additional file 2]. Ten genes ended in a complete termination codon, either TAA, TAG or AGA. The remaining three genes (cox2, nad4 and cob) did not possess a complete stop codon, but did show a terminal T. This condition is known to be common to vertebrate mitochondrial genomes whereby post-transcriptional polyadenylation provides the two adenosine residues required for generating the TAA stop codon 35.

The total nucleotide length for the 13 coding genes was found to be 11,403 bp, shorter than that of silver arowana and butterfly fish, but longer than that of bichir [see Additional file 2 for the exact sizes]. The coding sequences in Asian arowana consisted of 28.0% A, 25.1% T, 14.8% G and 32.1% C bases. The corresponding ranges for silver arowana, butterfly fish and bichir were 29.2–30.4% (A), 27.9–31.3% (T), 13.4–14.2% (G), and 24.3–28.8% (C). These data further support the observations: i) the GC content of Asian arowana mitogenome is higher than that of other teleost, including other known Osteoglossoidei species; and ii) the frequency of G is the lowest among the four bases in fish mitochondrial genomes 2.

Comparison of amino acid sequences for the 13 proteins among Asian arowana, silver arowana, butterfly fish and bichir confirmed the closer taxonomic relatedness of Asian arowana to silver arowana, than to butterfly fish or bichir (Table 1). In agreement with others' data 236, cox1 was the most conserved gene and atp8 was the most variable.

The pattern of codon usage in Asian arowana mtDNA was also studied. The most frequently used amino acids were leucine (16.9%), followed by threonine (8.9%), alanine (8.4%) and glycine (7.8%) [see Additional file 4]. At the third codon position, the order of nucleotide usage frequency was C > A > T > G (Figure 3), the same order was described earlier for the mitochondrial genome of Japanese fugu 37. The order was somewhat different in the silver arowana, butterfly fish and bichir, where A became the most frequently used base in the third codon position, albeit the frequency of G remained the lowest (Figure 3).

For amino acids with fourfold degenerate third codon position, codons ending with A were the most frequent (42.7%), followed by C (36.5%), T (14.1%) and G (6.2%). Genes located on the heavy strand showed a typical native GC and positive AT skew (Figure 4), whereas the nad6 gene located on the light strand displayed an opposite pattern. With regard to the absolute value, the GC skew was always higher than the AT skew: the former ranged from 0.60 to 1, whereas the latter ranged from 0.33 to 0.72 (Figure 4). Similar patterns were also seen in silver arowana, butterfly fish and bichir (data not shown). The GC and AT skews in Asian arowana were not correlated (R = 0.094, P > 0.05).

The twenty-two tRNA genes typical of vertebrate mitochondrial genomes have all been detected in the Asian arowana mitogenome. All tRNA genes possessed anticodons that match the vertebrate mitochondrial genetic code. The length of tRNA genes ranged from 67 bp to 74 bp [see Additional file 3] with a total length of 1,550 bp, similar to silver arowana and bichir, but shorter than that in butterfly fish [see Additional file 2]. The inferred secondary structure of the 22 tRNA genes had several uniform features: 7 bp in the aminoacyl stem, 5 bp in the TφC and anticodon stem, 4 bp in the DHU stem and 7 bp in the anticodon loop. A "U" residue before the anticodon was found in 19 of the 22 tRNA, whereas a purine was detected in the position immediately 3' to the anticodon. In the stem regions, there were several non-complementary pairings, mainly A-C type. A similar structure has been found in the silver arowana, whereas different kinds of non-standard base pairings have also been described in other fish species 2. The original sequences and the secondary structure of the tRNA genes were quite different in genetically distant related species.

Like the mitochondrial genome of other fishes, the Asian arowana mitogenome was found to possess two ribosomal RNA (rRNA) genes, a small rRNA gene (rrnS) and a large rRNA gene (rrnL), the two being separated by trnV [see Additional file 3]. The length of rrnS and rrnL are 956 and 1,698 bp, respectively [see Additional file 3]. These sizes are similar to those in the other three species used for comparison [see Additional file 2]. Substitution rates of the two rRNAs among Asian arowana, butterfly fish and bichir were lower than those of protein coding genes. Secondary structures found in the four species seemed to be conserved across large evolutionary distances, as described earlier for teleosts 2.

Several studies have been published recently on the phylogeny of Osteoglossoidei suborder using morphological data 11, partial mitochondrial sequences 38, a few nuclear genes 39 or the combination of the latter two 40. On the other hand, there is only a single study that analysed the phylogenetic relationship of the osteoglossids based on all genes present in the mitochondrial genomes 41 but the Asian arowana was not included as its complete mitogenomic sequence was not available.

To determine whether the addition of the complete Asian arowana mitogenome causes any difference in the evolutionary position of the Osteoglossomorpha from the cladograms produced earlier 11383940, we used the complete Asian arowana mitogenome sequence obtained in this study and other osteoglossids' complete mitogenome sequences to carry out phylogenetic analysis. Beside the osteoglossid species our analysis also included mtDNA from three fish species of ancestral lineages: four members of the Chondrostei taxon and representatives for both the Elopomorpha and Clupeocephala taxa (for complete list of species used, refer to Additional file 1). Using both nucleotide and amino acid sequences of different kinds of mitochondrial genes (see Materials and Methods for details) the systematic arrangements were reconstructed as monophyletic which is in agreement with the relationship tree of basal Actinopterygians produced by Inoue and colleagues 41.

Phylogenetic trees constructed with the various data sets using three different methods (i.e. MP, BI and ML) showed little variations within the data set, indicating that variation mainly originated from the type of data and not the methods used (data not shown). On all trees the Asian arowana was clustered into one group with the silver arowana, butterfly fish and goldeneye (all three from the Osteoglossomorpha superorder) with a high bootstrap support value. However, within Teleostei taxon the position of Osteoglossomorpha clade varied in the trees generated using concatenated protein-coding cum tRNA nucleotide sequences and concatenated protein-coding cum tRNA cum rRNA nucleotide sequences data sets. On the other hand, trees generated using concatenated protein-coding nucleotide sequences and concatenated amino acid sequences data sets consistently placed the Osteoglossomorpha clade at the basal level (Figure 5). This is in agreement with trees constructed earlier by others using various molecular 938394041 and morphological data 11. In addition the proximity of Osteoglossomorpha clade to that of basal teleost clades in our study further supports the position of osteoglossids among the early branches of living teleosts' stem lineages (see e.g. 41)

The placement of goldeneye and butterfly fish was different in osteoglossid cladograms produced earlier on the basis of morphological data 414243444546. While most publications predicted that during the evolution of osteoglossids the ancestor of goldeneye split off earlier from the arowanas, than from the butterfly fish 42434445. One study proposed exactly the opposite 46. Our cladogram based on full mtDNA sequences similarly to the data from 41 from four osteoglossids supports the former situation (Figure 5).

Since goldeneye is the only complete mtDNA sequence reported for Notopteroidei suborder, additional full mitogenomic sequences from this taxonomic group will have to be obtained for a more detailed analysis.

Six adult Asian arowana individuals (two green, two red and two golden variety) were obtained from a fish farm in Singapore. A small piece of fin clip (ca. 0.5 cm³) was collected from every individual and kept in absolute ethanol at 4°C. Whole genomic DNA including nuclear and mitochondrial DNA was isolated using a quick method developed previously in our laboratory 47.

Two pairs of primers (Dmt-A1/B1 and Dmt-A2/B2, see Additional file 5) were designed from nucleotide sequences of nad2 gene [GenBank:AB035222] and cob gene [GenBank:AB035234] deposited in Genbank. Long distance PCRs were carried out using Expand Long Template PCR System (Roche) to amplify 2 overlapping fragments of the complete mitochondrial genome. Each 50 μl reaction volume contained 1 × PCR buffer 2 (Roche) with 2.0 mM MgCl₂, 200 nM of each primer, 400 μM dNTP, 50 ng genomic DNA of one green Asian arowana and 3 U Taq polymerase mix (Roche). The following PCR program was used: 10 cycles of 94°C for 10 sec, 63°C for 30 sec and 68°C for 8 min then 19 cycles of 94°C for 10 sec, 63°C for 30 sec and 68°C for 5 min with an addition of 20 sec/cycle, as well as a final extension at 68°C for 10 min. Primer pair Dmt-A1B1 amplified a fragment of ca. 7.3 kb and primer pair Dmt-A2B2 produced ca. 11.5 kb product. The two fragments overlapped at both ends by a total of ca. 2 kb.

PCR products (25 μl) were sonicated using Branson digital sonicator model 450 (Branson) at 20% power for 4 seconds to generate DNA fragments suitable for cloning into a plasmid (400 bp to 2 kb). Sonicated PCR products were treated with T4 DNA polymerase, Klenow DNA polymerase and T4 polynucleotide kinase (all from Stratagene) according to manufacturer's protocol to blunt and phosphorylate the ends. Treated DNA fragments were electrophoresed through a 1% low melting agarose gel (Bio-Rad). Fragments between 500 bp and 1.5 kb size were cut out from the gel and cleaned using self-made glassmilk as described previously 13. The isolated DNA fragments were ligated into pBluescript KS (-) (Stratagene) vector pre-digested with SmaI (Stratagene). Nucleotide sequencing of the cloned inserts was conducted by using BigDye assay kit version 3.1 (Applied Biosystems) and M13F/M13R sequencing primer [see Additional file 5) as described previously 13. One hundred and forty-four clones with average insert length of 1 kb were sequenced; representing at least 6 times coverage of each position in the complete mitochondrial genome. Flanking vector sequences were clipped automatically by using commercially available software Sequencher (GeneCodes) with manual correction. Sequences were assembled by using the same software. The complete sequence of Asian arowana mitochondrial genome was deposited in NCBI's Genbank [GenBank:DQ023143].

tRNA genes were identified as described by Lowe and Eddy 48 with a cove cutoff score of 0.1. Protein and ribosome RNA genes were identified by sequence similarity to their orthologs from other mitochondrial genomes. The 5' ends of protein-coding genes were inferred to be at the first legitimate in-frame start codon (ATN, GTG, TTG and GTT) 49 that did not overlap with the preceding gene, except with an upstream tRNA gene and was limited to the most 3' nucleotide of the tRNA. Protein gene termini were inferred to be at the first in-frame stop codon (TAA, TAG, AGA and AGG). In some genes a T or TA nucleotides adjacent to the beginning of a downstream gene was designated as the truncated codon and assumed to be completed by polyadenylation after transcript cleavage 35.

Editseq and GeneQuest software (both from Dnastar) were used to analyze base composition and codon usage. Compositional skew, which indicates compositional difference between the two strands, was calculated using the following formula proposed by Perna and Kocher 50:

GC skew = (G-C)/(G+C)

and

AT skew = (A-T)/(A+T)

,where C, G, A, and T are the frequencies of the four bases at third codon position of the eight fourfold degenerate codon families.

A pair of primers (Dmt-MS-A/B, see Additional file 5) was designed using PrimerSelect (DnaStar) to flank the microsatellite site in heavy strand control region. One of the primers was labeled with a fluorescent dye 6FAM at the 5' end. The PCR mastermix consisted of 1 × PCR buffer with 1.5 mM MgCl₂ (Finnzyme), 200 nM primer, 400 μM dNTP, 40 ng genomic DNA, and 1 U DyNAzyme polymerase (Finnzymes). The amplification was performed in a PTC-100 PCR machine (MJ Research) using the following program: 94°C for 2 min, 34 cycles of 94°C for 30 sec, 55°C for 30 sec and 72°C for 30 sec followed by a final extension at 72°C for 5 minutes. PCR products were separated on an ABI 377 DNA sequencer (Applied Biosystems) as described previously 13. All six individuals were genotyped to detect possible polymorphism.

For characterization of long tandem repeat in the heavy strand control region, we designed one pair of primers (Dmt-LA/LB, see Additional file 5) flanking the control region using PrimerSelect (DnaStar). Complete heavy strand control region was amplified from total genomic DNA of the six individuals used earlier for microsatellite genotyping under the following PCR conditions: 94°C for 3 min, 30 cycles of 94°C for 30 sec, 55°C for 30 sec and 72°C for 1 min, followed by a final extension at 72°C for 5 minutes. PCR products were cleaned using home-made glassmilk 47 before cloning into pGEM-T cloning vector (Promega). Clones were sequenced from both directions using M13F and M13R sequencing primers and BigDye Assay Kit version 3.1 (Applied Biosystems). Forward and reverse sequences were assembled by using Sequencher (GeneCodes).

For detailed study of the control region tandem repeat array, various Osteoglossiformes species (see Fig 2 for the species used) and Anguilla species (see Fig 2 for the species used) sequences were downloaded from NCBI Genbank. A single unit from the various tandem repeat array were then aligned using ClustalX 52. Hairpin structure of a repeat unit from the tandem repeat array was constructed using Mfold 51.

Phylogenetic analysis was performed using mitochondrial genome of eighteen fish species from representatives of Actinopterygii and Sarcopterygii taxa – among them all four available full mtDNA from the Osteoglossiformes superorder – were used [see Additional file 1]. A shark species, called spiny dogfish (Squalus acanthias, Squaliformes, Chondrichthyes) was used as outgroup. Four different data sets were analysed: i) concatenated protein-coding, tRNA and rRNA nucleotide sequences; ii) concatenated protein-coding and tRNA nucleotide sequences; iii) concatenated protein-coding nucleotide sequences and iv) concatenated protein-coding amino acid sequences. Amino acid sequences were aligned using ClustalX 52 then its nucleotide sequences were aligned with references to the amino acid sequences alignment using CodonAlign 2.0 53) and further edited manually. The full sequence of nad6 encoded by the L strand was excluded from the analysis, due to the deviating nucleotide and amino acid composition of this gene as compared to those encoded by the H strand. Third codons of the 12 heavy strand encoded protein-coding genes were excluded from the analysis together with loops of tRNA. Each of the four datasets were analyzed by maximum parsimony (MP) method in MEGA version 3.1 54, Bayesian inference (BI) method using MRBAYES 3.1.2 5556 and maximum likelihood (ML) using Tree-Puzzle version 5.2 57 for amino acids data set and TreeFinder version of May 2006 58 for nucleotide data sets.

For MP analysis 10 random additions were done using the close-neighbour-interchange option with search level 1. Bootstrap analysis with 1000 replicates was conducted.

To find the best model for the nucleotides and amino acid data sets, we applied Modeltest version 3.7 59 and Prottest version 1.3 60 respectively. The best-fit model was GTR + I + G for all nucleotide data sets and mtRev + I + G for the amino acid data set. BI method was performed for 10⁶ generations using nucleotide data sets and 5 × 10⁵ generations for the amino acid sequences data set. The first 25% of samples were discarded as burn-in.

Quartet-based ML analysis for amino acid data set was performed using TreePuzzle 57. 1000 steps were performed and the mtRev24 substitution model was used. For parameter estimation, quartet sampling and NJ tree option was chosen. For ML analysis of nucleotides data sets, the program TreeFinder 58 was used. GTR substitution model was used and bootstrap analysis was performed with 1000 replicates.