DNA barcoding of billfishes

Abstract

DNA barcoding is a method promising fast and accurate identification of animal species based on the sequencing of the mitochondrial c oxidase subunit (COI) gene. In this study, we explore the prospects for DNA barcoding in one particular fish group, the billfishes (suborder Xiphioidei—swordfish, marlins, spearfishes, and sailfish). We sequenced the mitochondrial COI gene from 296 individuals from the 10 currently recognized species of billfishes, and combined these data with a further 57 sequences from previously published projects. We also sequenced the rhodopsin gene from a subset of 72 individuals to allow comparison of mitochondrial results against a nuclear marker. Five of the 10 species are readily distinguishable by COI barcodes. Of the rest, the striped marlin (Kajikia audax) and white marlin (K. albida) show highly similar sequences and are not unambiguously distinguishable by barcodes alone, likewise are the three spearfishes Tetrapturus angustirostris, T. belone, and T. pfluegeri. We discuss the taxonomic status of these species groups in light of our and other data, molecular and morphological.

Keywords::

• COI
• rhodopsin
• mitochondrial DNA
• species delimitation

Introduction

In the continuing quest to catalog and monitor the biodiversity of the world's oceans, molecular methods are assuming an increasing prominence. Fishes constitute the largest vertebrate group on the planet, with approximately 30,000 species (Nelson 2006), and fisheries and related industries are vital to the economies of many nations providing a significant fraction of global human food needs. Accurate and reliable species identifications of fish and fish tissues are important in many situations, including the detection of market substitutions (in which fish is sold under an incorrect species name, either to increase profit or to conceal illegal catches) (Wong and Hanner 2008; GAO 2009), monitoring of fisheries landings for stock assessments and management, early life stage assessments (Victor et al. 2009), and scientific research generally (Yancy et al. 2008). Fish species identification has traditionally been carried out based on the examination of specimen morphology; however, in many cases in which species identifications are needed, morphological features may not be available (e.g. when food inspectors wish to determine whether a piece of fillet sold at a market is correctly named), or not yet developed (e.g. in which fish eggs or larvae need to be assigned to species). In such cases in which morphological information is lacking, molecules (DNA, RNA, and proteins) may serve as unique identifiers to discriminate among species (Arnot et al. 1993; Floyd et al. 2002).

In the past, a variety of molecular techniques have been utilized to properly discriminate among species, including protein electrophoresis, nucleic acid ‘fingerprinting’ methods such as restriction fragment length polymorphism (RFLP) and Amplification fragment length polymorphism (AFLP), and direct sequencing of specific genomic regions (Hsieh et al. 2005, 2007). These methods have generally been of small scale, developed on a case-by-case basis by different research groups and applicable only to small groups of species. In contrast, DNA barcoding is a sequencing-based approach that attempts to establish a single global system of both laboratory protocols and data management, which should ultimately be applicable to all animals (Hebert et al. 2003a).

In animals, DNA barcoding is based on sequencing a ∼658 base-pair fragment of the mitochondrial gene, cytochrome c oxidase subunit 1 (COI; Hebert et al. 2003a,b). This signature sequence can be compared against a database of known sequences from reference specimens to obtain a species identification (Ekrem et al. 2007). Because this method depends on molecular approaches rather than on morphology, it may be applied to organisms of any life stage from egg to adult. It also aims to employ standardized protocols that may be applied to a wide range of organisms by individuals possessing a minimum amount of technical expertise and without the need for extensive knowledge of traditional morphological taxonomy. Sequence and specimen data are stored and made available in the Barcode of Life Data System (BOLD) database (Ratnasingham and Hebert 2007), an online collaborative workbench used to construct and curate a global barcode reference sequence library. Under the organizing framework of the International Barcode of Life (iBOL) project, global campaigns are under way to barcode the world's biodiversity, including Fish Barcode of Life (FISH-BOL), the campaign to barcode fishes (Costa and Carvalho 2007; Ward et al. 2009). Here, we discuss the prospects for DNA barcoding in one particular fish group, the billfishes (suborder Xiphioidei).

The billfishes are a group of ray-finned fishes (class Actinopterygii) comprising two extant families: the monotypic Xiphiidae (swordfish, Xiphias gladius) and Istiophoridae (marlins, spearfishes, and sailfish) (Collette et al. 2006). All are large, active apex predators characterized by an elongate, sword-like, or spear-like snout. All species are dioecious (having separate sexes) and females attain larger sizes than males, but otherwise do not display strong sexual dimorphism (Collette, 2010). They are among the largest and fastest swimming bony fishes, and are capable of trans-oceanic movements (Sedberry and Loefer 2001). Billfishes are wide ranging, inhabiting both tropical and temperate waters, and, seasonally, also cold waters (to which they migrate during summer months for feeding); spawning of most species occurs in tropical and subtropical waters (Nakamura 1985).

All billfish species are of commercial value with particular importance in some Asian markets (Hsieh et al. 2005; Gentner 2007). Furthermore, all species are also highly desirable targets in recreational fisheries; just in the USA, for example, expenditures by recreational billfish anglers are estimated to exceed $2 billion annually (SEFSC/NMFS 2004). Commercial long-line fisheries cover nearly the entire natural distributions of the species, whereas sport fishing grounds are more restricted. Areas where the two zones overlap have been subject to conflicts over fishing rights (Nakamura 1985). Populations of all billfish species investigated have declined in recent years due to high levels of human exploitation, and international fishery management organizations have highlighted the need for more fisheries and basic biological information, including better tracking of landings by species and identification of spawning habitats, to aid global billfish management and conservation (Gentner 2007; Mora et al. 2009).

Although billfishes are both targeted in specific international fisheries and caught extensively as bycatch in pelagic fisheries, few countries keep accurate track of their landings, and even when they do, often group species together (e.g. sailfish and spearfishes) in their landings records (SEFSC/NMFS 2004; Gentner 2007). Monitoring landings by species even for adult fish is further complicated by the difficulties in identifying processed fishes that end up as similar looking carcasses or smaller body parts in the trade pipeline (McDowell and Graves 2002; SEFSC/NMFS 2004; Hsieh et al. 2005). Delineation of billfish spawning habitat is typically accomplished by conducting oceanographic surveys to find and monitor the temporal and spatial distributions of individuals across life-stages, including eggs, larvae, and adults (Kawakami et al. 2010; Richardson et al. 2010). Yet identification of billfish early life stages (eggs and larvae) based on morphological characters is often problematic (SEFSC/NMFS 2004; Hyde et al. 2005).

These species identification difficulties in the context of the ecological and fishery importance of billfishes have led to the development of a variety of genetic approaches to facilitate the accurate identification of their body parts and early life stages. Approaches developed to date include restriction fragment length polymorphism analysis of polymerase chain reaction (PCR) amplified loci, multiplex PCR, and direct sequencing of either the mitochondrial 16S rRNA or cytochrome b loci (McDowell and Graves 2002; Hsieh et al. 2005; Hyde et al. 2005; Luthy et al. 2005; Richardson et al. 2007; Kawakami et al. 2010). These approaches successfully discriminate among the subset of billfish species tested in each study; however, none of these approaches have been applied to all known billfish species. In particular, the white marlin and striped marlin (Kajikia albida/K. audax), which are believed (based on both morphological and molecular comparisons) to have diverged very recently in evolutionary time (Collette et al. 2006), have proven problematic for molecular-based discrimination. Previous studies (e.g. Graves 1998) have failed to resolve this species pair utilizing genetic sequence data derived from multiple loci, including the mitochondrial genes ND4 and cytochrome b (Shivji et al. 2006), and the mitochondrial control region (mtCR) and ND2 and 12S rRNA genes (Collette et al. 2006).

Given the goal of establishing a global, standard genetic marker and method for all fishes under the organizing framework of the iBOL initiative, we aim to calibrate the utility of DNA barcoding (using the COI marker) in this important group, evaluating it against a nuclear marker (Rhodopsin) that has also been recommended for large-scale use (Sevilla et al. 2007).

Materials and methods

Samples

Specimens were obtained from a variety of sources. Because of the challenges associated with the retention of large-bodied animals and due to the fact that many of the samples derived from fisheries and sport fishing initiatives, there are no morphological voucher specimens associated with these samples. Species identifications were made on the basis of morphology by the collectors at the time of collection. All fish tissue voucher samples (muscle or fin clips) are deposited in the Guy Harvey Research Institute (Dania Beach, FL, USA).

DNA extraction, PCR amplification, and sequencing: Mitochondrial COI gene

DNA was extracted from these samples using an automated glass fiber protocol (Ivanova et al. 2006). The DNA barcode region of COI (652 bp) was then amplified by PCR using M13-tailed fish primer cocktails (Ivanova et al. 2007). Each 12.5 μl PCR comprised 6.25 μl of 10% trehalose; 2 μl of ultrapure water; 1.25 μl of Invitrogen 10 × PCR buffer [200 mM Tris–HCl (pH 8.4), 500 mM KCl]; 0.625 μl MgCl₂ (50 mM); 0.125 μl of each primer cocktail (C_FishF1t1 and C_FishR1t1 from Ivanova et al. 2007); 0.062 μl of each dNTP (10 mM); 0.060 μl of Platinum Taq DNA Polymerase (Invitrogen, USA); and 2.0 μl of extracted fish DNA template. Thermocycling conditions were as follows: 95°C for 2 min; 35 cycles [94°C for 30 s, 52°C for 30 s, 72°C for 1 min]; and 72°C for 10 min (hold at 4°C).

PCR products were visualized on pre-cast 1.2% Agarose E-gels (Invitrogen) to check results, and successful amplicons were sequenced bidirectionally using the primers M13F and M13R (Ivanova et al. 2007) with BigDye 3.1 Cycle Sequencing mix (Applied Biosystems, USA), and run on an ABI 3730 capillary sequencer to generate electropherogram trace files. Sequences were quality checked, edited, and assembled using SeqScape software (Applied Biosystems), before being uploaded to the Barcode of Life Data Systems (Ratnasingham and Hebert 2007). All sequences, as well as trace files and associated specimen data, are available in the project ‘Billfish and Swordfish COI identification’ (EBFSF) at http://www.boldsystems.org; sequences are also deposited in GenBank with the accession numbers (pending).

Previously published COI barcode sequences

Using the BOLD's Taxonomy Search function, all COI sequences in public projects deriving from the members of the families Xiphiidae and Istiophoridae were extracted and added to our analysis. Data were drawn from the following projects on BOLD: Fishes of Australia Part I (FOA); Marine Fishes of California (MFC); Marine Fish of Mexico II (MXII); DNA Barcoding the Indian Marine Fishes (WLIND); GenBank Fish (ANGBF); and Overlooked Fishes in Marine Settings (TZSAA).

DNA extraction, PCR amplification, and sequencing: Nuclear rhodopsin gene

Genomic DNA was extracted from all samples using DNeasy Kits (QIAGEN, Inc., Valencia, USA) from approximately 25 mg of tissue. Amplification via PCR targeted a 460 bp segment of the rhodopsin gene (Rho), suggested by Sevilla et al. (2007) as a potential nuclear barcode for teleost fishes. The primer pair Rod-F2w (5′-AGCAACTTCCGCTTCGGTGAGAA-3′) and Rod-R4n (5′-GGAACTGCTTGTTCATGCAGATGTAGAT) (Sevilla et al. 2007) were utilized to both amplify and sequence the DNA segment. Total PCR volumes were 50 μl and contained 33.3 μl of high purity water (OmniSolv®, VWR, Radnor, USA), 5 μl 10 × PCR buffer [15 mM MgCl₂], 8 μl dNTPs [1.25 mM of each dNTP], 1.25 μl of each primer (10 pmol), 1 unit of HotStar Taq DNA Polymerase (QIAGEN, Inc.), and 1 μl of template genomic DNA. PCR was carried out in a Mastercycler Gradient (Eppendorf, Inc., New York, USA) thermal cycler as follows: 95°C initial heating for 15 min; 35 cycles [94°C for 1 min, 50°C for 1 min, and 72°C for 2 min]; 72°C for 20 min.

A negative control (no genomic DNA) was included in each set of reactions to check for reagent contamination. PCR products were visualized on a 1.2% agarose gel to check for successful amplification, and were subsequently purified using the QIAquick PCR Purification Kit (QIAGEN, Inc.). Sequencing was carried out using BigDye® Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems), and all samples were run on an automated ABI 3130 genetic analyzer. Sequences were aligned manually using the program Genedoc 2.6.002 (Nicholas and Nicholas 1997), and were subsequently uploaded to the BOLD database. As indicated above, all sequences (GenBank Accession nos. pending), trace files, and specimen data are available in BOLD.

Phylogenetic reconstruction

The program MEGA 4.0.2 (Kumar et al. 2008) was used to generate a neighbor-joining (NJ) phenogram using the Kimura 2-parameter (K2P) distance metric to visually compare sequence similarities. To examine statistical support for branches, we reduced the dataset to a single representative for each unique haplotype (47 sequences), and a maximum parsimony analysis was carried out in MEGA, including bootstrap resampling with 100 replicates.

Sequence groups that were not clearly separated on the NJ tree were analyzed using haplotype parsimony networks, generated using the program TCS (Clement et al. 2000) at the 99% connection limit.

Reanalysis of previous data

mtCR sequences from two previously published studies (Graves and McDowell 2006; McDowell and Graves 2008) were downloaded from GenBank. These comprised 79 sequences from striped marlin (K. audax) (Accession Nos DQ199950–DQ200028) and 91 from white marlin (K. albida) (Accession Nos DQ835191–DQ835281). These previous papers only examined genetic variation within these two species separately, and did not combine all sequences into a single analysis.

Results

DNA barcodes (mitochondrial COI)

COI sequences were obtained from 296 individuals from the 10 currently recognized species of billfish: black marlin (Istiompax indica)—12 specimens; blue marlin (Makaira nigricans)—49 specimens; sailfish (Istiophorus platypterus)—34 specimens; white marlin (K. albida)—44 specimens; striped marlin (K. audax)—28 specimens; shortbill spearfish (Tetrapturus angustirostris)—11 specimens; Mediterranean spearfish (T. belone)—15 specimens; roundscale spearfish (T. georgii)—51 specimens; longbill spearfish (T. pfluegeri)—43 specimens; swordfish (X. gladius)—12 specimens. Mean sequence length was 647 bp (range: 500–652 bp). No evidence of indels or stop codons was seen (which if present would suggest amplification of a NUMT rather than the functional mitochondrial COI). Overall, nucleotide frequencies in this dataset were G = 18.82%; C = 29.51%; A = 22.71%; and T = 28.95%.

In addition, a further 57 billfish COI sequences from the publically accessible section of BOLD (from previously published barcoding projects or gathered from GenBank) were included in our analysis. A NJ tree of all 353 sequences (K2P distance) showed that 5 of the 10 species form distinct, cohesive clusters (Figure 1). In contrast, the white and striped marlin pair (K. albida and K. audax) appear mixed in the same cluster, and the same is true for three of the four known spearfishes: T. angustirostris, T. belone, and T. pfluegeri (in contrast, T. georgii was well separated from its congenerics). The maximum parsimony analysis (included in supplementary online material) shows low bootstrap support for the deeper nodes of the tree; therefore, these internal branches were reduced to polytomies on the main NJ tree in Figure 1. It should be noted that this tree is not intended as a hypothesis of phylogenetic relationships but simply as a visual representation of haplotype groupings.

Figure 1.  NJ tree of COI sequences. NJ analysis, carried out in MEGA using K2P distance, of all billfish COI barcode sequences included this study. The full tree of 353 sequences is included in Supplementary Online Material. Multiple sequences are collapsed to triangles in which vertical distance corresponds to the number of sequences, and horizontal distance is proportional to sequence diversity. Numbers indicate the number of individual specimens in a species. Several internal branches were collapsed to polytomies to indicate low bootstrap support for the same nodes in a maximum parsimony analysis—see supplementary online material. Images are taken from the FAO ‘Billfishes of the World’ catalog (Nakamura 1985) and are used with the permission of the copyright holder.

To examine sequence variation more closely within each of these two multi-species groups (white/striped marlins and spearfishes), we constructed haplotype parsimony networks using TCS. For the white and striped marlin (Figure 2), the 91 sequences formed 21 distinct haplotypes; whereas sequences between the two species were often highly similar (differing by only 1 bp in some cases), only one exact haplotype was shared between K. albida and K. audax; this was also determined in TCS to be the likely root (ancestral) haplotype. The individuals carrying this haplotype included six specimens of K. albida and five of K. audax (Figure 2). The between-group mean K2P distance between K. albida and K. audax for all COI sequences was 0.004.

Figure 2.  Haplotype parsimony network for white and striped marlin. Haplotype parsimony network, generated using the program TCS at the 99% connection limit, of 91 COI sequences belonging to the white marlin (K. albida) and striped marlin (K. audax). Individuals carrying the shared haplotype (square box) included six specimens of K. albida (EBFSF021-07, EBFSF025-07, EBFSF103-09, EBFSF115-09, MXII141-07, MXII138-07) and five of K. audax (EBFSF278-10, EBFSF280-10, EBFSF284-10, EBFSF289-10, EBFSF291-10).

For the spearfishes (Figure 3), the 64 sequences fell into 16 haplotypes, of which just three were shared by two of the three species (none were shared by all three species). The remaining haplotypes were private to a single species.

Figure 3.  Haplotype parsimony network for spearfishes. Haplotype parsimony network, generated using the program TCS at the 99% connection limit, of 64 COI sequences belonging to the spearfish species T. angustirostris, T. belone, and T. pfluegeri.

Reanalysis of mtCR sequences

A new NJ tree (Figure 4) of 170 control region (mtCR) sequences from striped marlin (K. audax) and white marlin (K. albida) (Graves and McDowell 2006; McDowell and Graves 2008) showed that, although most of the sequences clustered into two separate clades by species, a small group (n = 5) of striped marlin sequences shows greater similarity to the white marlin clade than to the rest of the striped marlin sequences.

Figure 4.  NJ tree of mtCR sequences. NJ analysis, carried out in MEGA using K2P distance, of mtCR sequences, extracted from GenBank and originally from the studies by Graves and McDowell (2006) and McDowell and Graves (2008).

Nuclear rhodopsin

Sequences of the nuclear Rho were obtained from 72 specimens (all but two also represented by a COI sequence). All sequences were 460 bp in length. Overall nucleotide frequencies in this dataset were G = 25.4%; C = 29.48%; A = 17.82%; and T = 27.22%.

An NJ tree of these 72 sequences (Figure 5) showed that rhodopsin also failed to resolve the albida-audax and the angustirostris-belone-pfluegerii species clusters, and that furthermore, T. georgii is not cleanly separated from the other spearfishes. Although a single fixed nucleotide difference discriminated all nine K. albida specimens from the majority of the K. audax (position 199/460; ‘A’ in albida, ‘G’ in audax; a synonymous substitution), a small number—four individuals—of K. audax appear to be heterozygous at this position (represented ‘R’—the IUPAC code for A or G—in our alignment)—determined by two overlapping peaks at this position in the sequence electropherogram. These specimens were EBFSF035-07, EBFSF129-09, EBFSF130-09, and EBFSF137-09 (Figure 5). Notably, however, these do not include the same individuals that share the identical COI haplotype, as seen in the TCS analysis (see above).

Figure 5.  NJ tree of rhodopsin sequences. NJ analysis of 72 nuclear rhodopsin (Rho) sequences, carried out using the BOLD Taxon ID Tree function, using K2P distance. Arrow indicates the four apparently heterozygous K. audax individuals.

Discussion

These results suggest that at least 5 of the 10 recognized species of billfish are readily distinguishable by standard COI barcodes: I. indica, I. platypterus, M. nigricans, T. georgii, and X. gladius. The remaining five species fall into two complexes within which species may not be unambiguously resolved by barcode sequences: the striped marlin (K. audax) and white marlin (K. albida) are one complex, whereas the other consists of the three spearfishes T. angustirostris, T. belone, and T. pfluegeri. Nevertheless, the presence of private haplotypes (Figures 2 and 3) still offers the potential for assigning an unknown specimen to species (though given the relatively small sample size examined thus far, the possibility that more shared haplotypes might be seen with more sequences cannot be ruled out). It is interesting to note that, for the white and striped marlin, the single COI haplotype that is shared by the members of both species was also determined by TCS to be the most likely ancestral haplotype for the whole group. Sampling locations in which this haplotype was recovered included both the Pacific and Atlantic coastal areas of Mexico. Interestingly, when examining the same individuals for both mitochondrial COI and nuclear rhodopsin sequences, it is not the same groups of individuals carrying the identical sequences—those which are indistinguishable by COI are distinguishable by Rho, and vice versa (in which both loci are available from the same individuals).

Our analysis of COI and rhodopsin further underscores the difficulty associated with resolving this species-complex, failing to detect a clear genetic distinction between the two species. Thus, far, all genetic analyses conducted have failed to clearly distinguish the two species as reciprocally monophyletic groups, regardless of the locus examined (Collette et al. 2006; this study). Although some genetic differentiation exists among most examined specimens of white and striped marlin, this differentiation is thus far small (mean between-group K2P distance of 0.004 for COI sequences), raising the issue of whether this is sufficient to diagnose the two as separate species. Alternatively, the detection of such limited genetic differentiation may in fact suggest that these two ‘species’ actually comprise two geographically separated subspecies with gene flow still occurring at points of contact, or indeed might simply be different populations of the same species. Indeed, morphologically the white marlin and striped marlin also show a high degree of similarity. In his key, Nakamura (1985) separates audax from albidus and georgii just based on the tips of the fins being pointed in audax versus rounded in the other two species. The existence of four genetically discrete populations of striped marlin within the Pacific (McDowell and Graves 2008) further complicates the issue of species delineation, as some of these populations may also merit recognition as subspecies. On the basis of the available genetic and morphological evidence, the whole question of the validity of continued recognition of K. albida and K. audax as distinct species, as opposed to subspecies or even populations, requires reexamination with additional morphological and/or molecular data.

Given the strong morphological similarity of the billfishes, the ability to unambiguously resolve species is impeded by a lack of vouchers or associated metadata (especially for sequences gathered from GenBank) for these species. Consequently, there is often no way to follow-up/recheck on questionable identifications (Floyd et al. 2010). The adoption of barcoding data standards for future collections (Hanner 2005), with at least the recording of photographic images and Global Positioning System (GPS) coordinates of capture for all reference specimens, would enhance the value of sequence data subsequently generated and this is a major objective of Fish Barcode of Life (FISH-BOL). Given the lack of vouchers for the billfishes, pooling all available samples and analyzing them with a common set of markers is a logical first step. However, we cannot rule out the possibility that some of the sequences, particularly those gleaned from GenBank, might derive from misidentifications.

The lack of resolution of COI to discriminate half of the billfish species is notable given the relative utility of DNA barcoding based on this locus in other fish groups (Ward et al. 2008; Wong and Hanner 2008; Steinke et al. 2009a,b). Presumably the explanation lies in the fact that all of these species are relatively young in evolutionary time, and consequently there has been insufficient time for the accumulation of mutations. Speciation is a phenomenon which occurs at the interface between population genetic (reticulate) and phylogenetic (bifurcating) processes; species are diagnosable to the extent that different character states (alleles) become fixed in reproductively isolated populations, but the fixation of neutral alleles is a probabilistic process occurring over many generations (Padial et al. 2010), at heterogeneous rates for different genetic loci. This is particularly important in which genetic variation is neutrally evolving, and uncoupled to whatever phenotypic changes are responsible for species differences. In such cases, haplotypes deriving from the common ancestor population may persist for an indeterminate period of time in descendent populations, even after they have become fully reproductively isolated. As Padial et al. (2010) emphasize, modern taxonomy must be integrated with many new approaches for species delimitation, and in such cases, species should be diagnosed by differing allele frequencies rather than by distinct allele sequences, which the generation of broader-coverage barcode data could help to reveal (Hajibabaei et al. 2007).

The three spearfishes T. angustirostris, T. belone, and T. pfluegeri could not be resolved based on our data (COI and Rho), even though all of these species were distinguishable based on the dataset employed by Collette et al. (2006); this study used a concatenated suite of mitochondrial genes (ND2, CR, and 12S) and one nuclear gene (MN32), and found that these three species all formed reciprocally monophyletic groups with >70% bootstrap support. This is most likely a data sufficiency problem, i.e. simply the amount of genomic data which needs to be read to capture enough variation; the total length of concatenated sequence used by Collette et al. (2006) was 3787 bp for each specimen, compared with 1112 bp for our study (length of COI+Rho sequences combined). Hence, some examples of barcoding missing ‘young’ species (Moritz and Cicero 2004) are expected, but in general, this had not proven to be an issue for the vast majority of fishes, both freshwater and marine, examined to date (Ward et al. 2009).

As an identification tool for an unknown tissue sample, barcoding using COI alone may be expected to provide at least a genus-level identification of billfish in all cases. On the data in this study, for half of the possible billfish species this will also yield an unambiguous species identification, whereas for the other half it will be necessary to sequence alternative genes to unambiguously determine the species. Thus, barcoding can provide a routine and effective ‘first pass’ for analysis of unknown samples, which can be refined by further sequencing as required.

Declaration of interest: This work was supported through funding to the Canadian Barcode of Life Network from Genome Canada (through the Ontario Genomics Institute), NSERC, and other sponsors listed at http://www.BOLNET.ca; and additionally by operational funds to the Guy Harvey Research Institute, AFTCO, Inc., and an NSERC postgraduate fellowship (AB). We thank L. Beerkircher, J. Graves, J. McDowell, E. Prince, J. Serafy, and D. Snodgrass for providing tissue samples and Eugene Wong for assistance in the laboratory. We are grateful to Johanne Fischer of the FAO for granting permission to use images from their catalog (Nakamura 1985) for the first figure in our study. The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.