DNA barcoding a unique avifauna: an important tool for evolution, systematics and conservation

Evolutionary and systematic applications

Low levels of genetic divergence, particularly at just one locus, do not invalidate established taxonomy [45]. In cases of recent divergence, phenotypic differentiation can occur more rapidly than the complete sorting of mtDNA [45] while hybridisation and back-crossing can result in genetic introgression from one species to another [46]. Similar genetic patterns can also result from misidentification of specimens, although we made all efforts to minimise this issue. In this study, the majority of the 13 species pairs and triads that could not be distinguished by their COI barcodes, represent well-studied, valid species. For example, the species status of the extant New Zealand snipes (Coenocorypha spp.) are supported by reciprocal monophyly in both nuclear and mitochondrial markers as well as morphometric and plumage data [47]. Divergence is estimated to have occurred only about 96,000 years ago [47] suggesting incomplete lineage sorting as the most likely explanation for COI similarity. This is likely also the case for the masked gulls (Chroicocephalus scopulinus and C. bulleri) which diverged about 240,000 years ago [48]. Occasional hybridisation between these species has also been observed [49] and slow mutation rates have also been implicated [50]. Many gull species within the closely related genus Larus have indistinguishable COI barcodes e.g. [24, 27, 28, 29] which is attributed to recent speciation and hybridisation. The brown and south polar skua (Stercorarius antarcticus and S. maccormicki respectively) diverged only about 200,000 years ago and speciation is considered incomplete with hybridisation common [51, 52]. Originally considered a single species, the northern and southern giant petrels (Macronectes halli and M. giganteus) were split on the basis of morphological and behavioural differences [53]. This taxonomy is supported by nuclear and mitochondrial markers though genetic divergence levels are low [54], a reflection of recent divergence (about 200,000 years ago [54]) and hybridisation [55]. The low divergence between parakeets (Cyanoramphus spp.) is also likely a reflection of recent speciation [56]. Mitochondrial control region sequences were used to measure the divergence of the Antipodes parakeet (C. unicolor) which was estimated to have occurred ~ 270,000 years ago [56]. The sympatric Reischek’s parakeet (C. hochstetteri) colonised the Antipodes Islands much more recently, diverging from the red-crowned parakeet (C. novaezelandiae) ~ 100,000 years ago [56]. This is consistent with our finding that the Antipodes parakeet could be distinguished from the other two species by two diagnostic nucleotides. The introduced mallard and native pacific black duck are known to hybridise extensively with mtDNA introgression being bidirectional [57]. In two mallards, there was evidence of heteroplasmy at two nucleotide sites. Mitochondrial heteroplasmy, the occurrence of more than one haplotype within an individual, can occur as a result of mutation, recombination or paternal leakage [58]. The general assumption that mtDNA is uniparentally inherited and homoplasmic is being questioned by the accumulating evidence of paternal leakage in a variety of taxa (reviewed in Barr et al., [59]).

For other species that could not be identified by their COI barcodes, in-depth studies are lacking and further investigation is required. For example, the royal penguin (Eudyptes schlegeli) and macaroni penguin (E. chrysolophus) are considered conspecific by some [60, 61] and the mitochondrial hypervariable control region and the COI barcoding region show very low levels of divergence [62, 63]. Here we show that the COI sequences of specimens from royal and macaroni penguins generated by Baker et al. [63] could not be distinguished by CAOS. Taxonomic uncertainty also surrounds the New Zealand oystercatchers. A preliminary genetic study using mtDNA found no differences between the mainland species the South Island pied oystercatcher (Haematopus finschi) and the variable oystercatcher (H. unicolor) [64] which occasionally hybridise [65], but vary substantially in morphology. However, the Chatham Island species (H. chathamensis), which is considered by some to be a subspecies of the variable oystercatcher [66], was found to be distinct from the mainland species [64]. While none of the three species could be distinguished by COI barcodes, the Chatham Island oystercatcher showed the highest genetic divergence. Consistent plumage differences are currently the only basis for the separation of the Australasian bittern (Botaurus poiciloptilus) and the great bittern (B. stellaris) [61]. Our results suggest limited genetic divergence at the COI locus between the two species (mean distance of 0.12%) indicating their taxonomy may require further investigation.

Since we only have one Chatham Island pigeon (Hemiphaga chathamensis) specimen we can conclude little from the similarity between it and its sister taxa the New Zealand pigeon (H. novaeseelandiae). The Chatham Island pigeon was only recently elevated to species status on the basis of morphometric differences [67]. In a larger study, Goldberg et al. [68] found no differences in the COI region, and low divergence in cytochrome b and D-loop sequences (1.2 and 2.8% respectively) which was attributed to recent widespread dispersal. In the present study, the fulmar prion (Pachyptila crassirostris) was represented by a single sequence which showed low divergence when compared to the fairy prion (P. turtur). These species are sometimes considered conspecific [69].

Divergent COI lineages were evident within 16 species (Table 3). There is no level of genetic distance that can be used as a cut-off for species status, as speciation results in genetic divergence but is not caused by it [70]. However, divergence in COI barcodes can identify taxa in which further scrutiny may be required [24]. In seven globally distributed species, divergent lineages corresponded to recognised subspecies separated by large geographic distances; gentoo penguin (Pygoscelis papua), Wilson’s storm petrel (Oceanites oceanicus), great egret (Ardea alba), common pheasant (Phasianus colchicus), whimbrel (Numenius phaeopus), little owl (Athene noctua) and Eurasian skylark (Alauda arvensis). Within five of the six species that showed divergent lineages between New Zealand and Australia there are recognised subspecies; spur-winged plover (Vanellus miles), morepork (Ninox novaeseelandiae), little penguin (Eudyptula minor), New Zealand pipit (Anthus novaeseelandiae) and white-faced storm petrel (Pelagodroma marina). However, detailed geographic sampling would be required for each species to determine if COI barcodes could distinguish subspecies. Furthermore, it has been suggested that some of these currently recognised subspecies warrant separate species status. For example, there is evidence in the form of nuclear and mitochondrial markers [70, 71], behavioural [72] and plumage differences [71] that the New Zealand and Australian populations of little penguins (E. minor) should be recognised as separate species. Additionally, a recent study of the genus Ninox recommended that the mainland Australian population be treated as a separate species from Tasmanian and New Zealand populations [73].

The taxonomy of diving petrels remains unresolved and is the subject of debate [74, 75]. Within New Zealand, there was evidence of divergent lineages corresponding to recognised subspecies within four species. Populations of both little shearwater (Puffinus assimilis) and white-faced storm petrel (Pelagodroma marina) were divergent between the Mokohinau and Kermadec Islands. The white-faced storm petrels collected from the Mokohinau Islands and a beach wrecked individual found in the North Island (presumably from the Mokohinau Islands population) were over 5% divergent from the specimens collected in Australia and the Kermadec Islands. The Kermadec Island population is regarded as a distinct species (P. albiclunis) by Birds New Zealand [76]. Though we cannot be certain of which breeding population the Australian sample originated from, given its collection location it was probably a member of the Australian subspecies P. m. dulciae. Genetic comparisons of these populations are lacking, however, Silva et al. [77] found that the Mokohinau Island population was highly differentiated from North and South Atlantic populations using mitochondrial and nuclear markers. Rifleman (Acanthisitta chloris) and New Zealand robin (Petroica australis) showed divergent lineages between the North and South Islands. New Zealand robin lineages showed a divergence of 4.12%, similar to control region sequences which showed 5.9% divergence suggesting long-term isolation [78]. Indeed, Birds New Zealand recognises the North Island robin as a separate species P. longipes [76].

There was no clear pattern in the success rates of DNA barcoding despite the unique composition of the New Zealand avifauna. High levels of endemism had no obvious effect on success rates. For New Zealand species represented by > 1 specimen, 85.5% of endemic, 88.75% of native and 87.1% of introduced species could be identified by DNA barcodes. Divergent lineages were evident in a similar proportion of native and introduced species (6.9 and 9.7% respectively). The high prevalence of seabirds did not appear to influence success rates with 90.4% of seabirds successfully identified compared to 88% of land birds and 7.2% of seabirds showing evidence of divergent lineages compared to 9.2% of land birds. Importantly, this demonstrates that DNA barcoding can be successfully applied to species discrimination in fauna with a wide range of evolutionary patterns and life history traits.

Conservation management applications

The 928 COI sequences from 180 New Zealand bird species generated from this study form a substantial reference database that will be a valuable tool for specimen identification and the conservation of New Zealand birds. DNA barcoding has many advantages over morphological identification when applied to conservation management [14]. DNA barcoding can utilise non-invasive samples such as feathers or faeces [11] which is beneficial when dealing with rare and endangered birds or elusive predators [79]. Invasive mammalian predators are the biggest threat to the survival of New Zealand birds, responsible for the majority of the 26.6 million chick and egg losses of native bird species each year [80]. Diet studies using DNA barcodes can be used to assess predator impact on prey populations and provide superior detection and identification of prey species when compared to morphological analysis [14].

Performance of different methods of analysis

CAOS was found to be the most successful method for identifying specimens to the species level. All species that were distinguishable using neighbour-joining tree or other distance-based methods were also successfully identified by CAOS and an additional 14 species could only be identified using CAOS. While previously the application of CAOS has been limited by scalability issues [24], this has now been overcome and large datasets such as ours can be successfully analysed. We found that there were no differences in output when CAOS was run using smaller datasets consisting of species from one order. When sequences from species not found in our database were queried, CAOS correctly identified these individuals to the genus level. While this issue highlights the importance of thorough sampling in the reference database, genus level identification is more useful than no identification at all.

It is well-established that phylogenetic trees may perform poorly for the purpose of specimen identification [81, 82]. It is not possible to determine if a query sequence belongs to the species which it is topologically closest to unless it is nested within a monophyletic cluster [83]. Additionally, when either speciation is recent and individual genes are still incompletely sorted, or when introgressive hybridisation is occurring, non-monophyly is to be expected [84, 85]. Despite these limitations, quantifying the level of monophyly is still a useful descriptor of the data [4]. In this study, non-monophyly was observed in 14.7% of species, that is similar to values reported in other studies of Aves, between 10.4% [86] and 16.7% [46]. While distance and phylogenetic tree-based methods do not have the same level of success as CAOS, they reveal interesting features of the data which character-based methods do not. For example, evidence of divergent lineages can be quickly observed in a phylogenetic tree while large intraspecific variation may also indicate divergence. For species that show small interspecific distances and/or non-monophyly, we should be more cautious about identifications provided by CAOS as discussed above.

Sampling

We generated COI sequences from 928 specimens representing 180 species from the New Zealand region (Fig. 2a and b). Samples included voucher specimens from the Auckland War Memorial Museum, Museum Victoria, Museum of Natural Sciences at Louisiana State University and the Royal Ontario Museum. Other specimens were collected in the field by a large number of people over the last 35 years. Where possible, individual birds were sampled from across the species’ geographic range in order to determine levels of geographic variation. Additional Genbank sequences were also included in the analysis (see Additional file 4). Taxonomy was based upon Clements [87], including corrections and updates up to 7 March 2017.

Open image in new window

DNA sequencing

For the majority of samples, the DNA extraction protocol, PCR conditions, sequencing methodology and primer details were as previously described by Patel et al. [88]. For the remaining samples the methodology is outlined in Tavares and Baker [89]. Sequences shorter than 519 bp or which contained ten or more ambiguous base calls were excluded from analysis. Specimen information, sequences and trace files can be accessed on the Barcode of Life Data Systems website (BOLD, GenBank accession numbers MK261779 - MK262706) [90].

Additional data

The data gathered in this study were supplemented by 892 sequences from GenBank that fell into two categories. Firstly, 488 sequences from species that occur in the New Zealand region. For non-endemic species, specimens from across their geographic distribution were preferentially included, to capture the most geographic variation. Secondly, in instances where a species’ closest relative did not occur in New Zealand, sequences from the most closely related species available were included to increase within genera sampling. These additional 404 sequences are referred to as related species and while they were included in all analyses, only the success rates and divergence levels of New Zealand species are reported. A number of GenBank sequences followed outdated taxonomic classifications and were renamed to follow Clements [87] (see Additional file 4).

In total 1820 sequences were included in the analysis, of which 1416 were from 211 species that occur in the New Zealand region. A complete list of GenBank accession numbers of the sequences used in this study is available in Additional file 4.

Analysis

Three DNA barcoding analysis methods were used; tree building, distance analysis and diagnostic character assignment. Tree building was conducted in MEGA version 7 [91]. Sequence alignment was performed with MUSCLE [92] and a neighbour-joining tree was produced based on uncorrected p-distances. P-distances have been shown to produce higher or similar levels of correct identification than Kimura-2-Parameter (K2P) distances which are commonly employed in barcoding studies [93, 94]. Support for monophyletic clades was measured using bootstrap values with 1000 replicates. Patterns of divergence were classified as either monophyletic with either greater than or less than 95% bootstrap support, paraphyletic or polyphyletic.

Local barcode gap analysis was conducted by calculating maximum intraspecific and minimum interspecific (nearest neighbour) genetic distances for each species using the Spider package [95] for RStudio [96]. For each species, these values were plotted against each other to visualise ‘local’ barcoding gaps; discontinuity between levels of intraspecific and interspecific distances [4]. Nearest neighbour distances were used in preference to average interspecific distances because species identification is ultimately dependent upon how different a sequence is from its closest allospecific sequence, as opposed to the distance to the “average” sequence [97]. An optimised global distance threshold was also calculated from the data, minimising the cumulative error rate [8].

Character-based identification was implemented in CAOS [42, 43, 44]. CAOS identifies diagnostic characters, termed ‘character attributes’ (CA’s) from a tree of pre-defined species. Single CA’s may be either pure (sPu’s) if they are shared by all members of a clade and are absent from the other clades or private (sPr’s) if they are shared only by some members of a clade [7]. Detailed methodology can be found in the Additional file 5. In brief, CAOS barcoding is comprised of three steps, each performed by a separate program. Firstly, the CAOS-Analyzer extracts CA’s from the input nexus file that consists of the sequence alignment and tree file fused together. Next, the outputs of the CAOS-Analyzer are converted into an easily interpretable character-based barcode matrix using the CAOS-Barcoder. Finally, the CAOS-Classifier tests the efficacy of this matrix by attempting to assign a new query specimen to the correct species in the reference dataset. For species with multiple representatives, the shortest sequence was excluded from the reference database and used as a query sequence.