Until recently it was assumed that the flightless moorhen of remote Tristan da Cunha in the southern Atlantic (Fig. 1), Gallinula nesiotis (Sclater, 1861) [1], became extinct by the end of the 19th century [2]. A few decades after its description, a very similar moorhen that was also flightless namely G. comeri (Allen, 1892), was described [3] from the island of Gough, ca. 400 km SE of Tristan. In the period between these descriptions G. nesiotis became rare [4], [5] and by the turn of the century it had probably gone extinct [7], [6]. Authentic remnants are two skins and a skeleton in the Natural History Museum, Tring [8]. Since unequivocal G. nesiotis had been collected only once from Tristan, and because of the presence of a healthy population of similar moorhens on the nearby island of Gough, some authors doubted whether an endemic moorhen had ever existed on Tristan [9]. Eber [10] compared Sclater's description of G. nesiotis from Tristan with her series of G. comeri from Gough and concluded that the differences fall within the range of variation of the latter. In her opinion it was very unlikely that moorhens from two islands in the same region would have independently lost the ability of flight, without differentiating in other characters. She suggested that Sclater's material might have been labelled inaccurately and that his specimens in fact also came from Gough. Consequently, Eber considered G. comeri a junior synonym of G. nesiotis [10] and controversy surrounded future illustrations of both taxa (Fig. 2).

Figure 1 Map showing the location of Tristan da Cunha and Gough in the Mid Atlantic.

Figure 2 Illustrative stamps, issued in 1987 and 2005.

Beintema [2] mentioned that there are old records of moorhens for Tristan, demonstrating that such birds were truly indigenous there. In his view, skeletal measurements differ slightly between G. nesiotis and G. comeri. Furthermore, rails are known to rapidly lose the ability of flight as soon as they arrive on remote islands [11]. When in 1972 live moorhens were discovered on Tristan [12], these birds were regarded as descendants of a small number of individuals brought from Gough [13]. Alternatively, Beintema suggested that G. nesiotis might have been temporarily rare on Tristan, but not extinct, and that the moorhens found there today are descendents of the original island population. Here we address this question, making use of DNA analyses of authentic material of G. nesiotis, recent specimens of the moorhens from both Tristan and Gough and some geographically and taxonomically close other taxa of moorhens.

An overview of the specimens that were used in this study, with taxon names, locality data, and year of acquisition, is given in Table 1. Alignments of all cloned sequences of G. nesiotis (two independent amplifications per target region) are shown in Dataset S1, S2, S3. On the one hand, none of the sequences of genuine, historical G. nesiotis was identical to those of G. comeri and, on the other hand, all sequences of the moorhens collected in Tristan da Cunha in 1993, are identical to those found for specimens of G. comeri from Gough, dated 1960. The genetic distances between G. nesiotis and the other gallinules, are of the same magnitude as the distances between G. comeri and the other gallinules (Table 2). A pairwise relative rates test did not reveal significantly different substitution rates for any of the lineages (P>0.18). Of the selected markers, most variation was detected in the control region (D-loop) sequences, viz. 9.6% TSH for G. nesiotis compared with G. comeri (= Total Sequence Heterogeneity [15], [14]) versus 2.1% and 0.3% TSH for tRNA-Lys/ATP synthase subunit 8 (ATP8) and cytochrome b, respectively.

Table 1 Taxa and collection information.

Table 2 Genetic distances.

The results of phylogenetic analyses (Neighbour-Joining, Maximum Likelihood and Bayes) based on a combined dataset (all taxa, all regions, Dataset S4) are shown in Fig. 3–5. In these cladograms Gallinula nesiotis and G. comeri form a clade with the moorhens of Africa/Eurasia, whereas the other taxa that were investigated are less closely related.

Figure 3 Bootstrap 50% majority rule consesus NJ tree.

Figure 4 Bootstrap 50% majority rule consensus ML tree.

Figure 5 Bayes 50% majority rule consensus (‘halfcompat’) phylogram (branch lengths are proportional to the expected number of substitutions).

Our results show that genuine G. nesiotis, identified on the basis of historical material from the island of Tristan da Cunha, differs genetically from G. comeri, which has been described from the island of Gough. For each marker the sequence of G. nesiotis differs from that of G. comeri, as well as from all other Gallinula taxa that were analysed, but the amount of variation differs strongly between the regions studied. The position of G. nesiotis on the cladograms (Fig. 3–5) makes sense biologically. Apparently, G. nesiotis became extinct on Tristan and G. comeri from Gough was introduced there, resulting in the current situation with G. comeri occurring on both islands. This implies that modern illustrations of so-called G. nesiotis from Tristan (Fig. 2) probably show introduced G. comeri from Gough.

The difference between G. nesiotis and G. comeri is most conspicuous in the D-loop sequence. This is a marker from a non-coding region, which makes it more difficult to exclude it as a potential pseudogene [16]. In some cases preferential amplification of numt (nuclear mitochondrial insertion) sequences has been observed [17], [18], but in most ancient DNA studies only occasional co-amplification of numts has been reported [19], [20]. This is not surprising since (particularly) in ancient DNA samples, mitochondrial DNA will be in excess over nuclear DNA. Consequently the incidence of numts should be reduced in such samples. To minimize the chance of amplifying numts, we did not use blood as a source for DNA-extractions in our recently collected material [17], [21] (in the older specimens it was not possible anyway). Because ‘universal’ primers may also be particularly prone to amplification of numts [17], the primers for the D-loop were made ‘gallinule-specific’. They did not even work for the closely related coot, Fulica atra. Products that were sequenced directly (both strands) showed only one signal, whereas multiple signals can be expected if both the target product and a numt would have been amplified. Interclone variation was low (Dataset S1, S2, S3). Only one sequence (G. nesiotis, marker ATP8) out of 96 clones (Table 3) could clearly be identified as a numt (Dataset S2). No stop-codons or frame-shift mutations were observed for the coding-region datasets (ATP8 and cytochrome b). No obvious deviations in either substitution rate (pairwise relative rate test) or base composition, like a decrease in GC content [22], [23], were observed.

Table 3 Number of colonies sequenced per target region per taxon.

All sequences of recently collected moorhens from Tristan were identical to those of G. comeri from Gough and should be considered conspecific therefore. Cross contamination is very unlikely, since specimens from Tristan and Gough were amplified in different PCR-batches and contamination was not detected in other, partly much older specimens. Most probably the sequences are identical because G. comeri was introduced only recently on Tristan. Genetic variation within island populations is generally small compared to mainland populations [24]–[26]. For example, the giant tortoises (Aldabrachelys) of Mahé (Seychelles) and Mauritius (Mascare islands) still have identical sequences compared to those of Aldabra, from where they were shipped since the 1820s or earlier [27]. G. comeri may have been introduced on Tristan somewhere in the mid 1950s [12]. Assuming a generation time of two years, as known for Gallinula chloropus, G. comeri would only have had about only 20 generations (40 years) to differentiate on Tristan.

Both the genetic distances (Table 2) and the fact that G. nesiotis and G. comeri form a clade with the investigated moorhens of Africa/Eurasia (Fig. 3–5), suggest that the ancestor(s) of these island gallinules originated from Africa and not America, as suggested by Eber [10]. Our data do not allow us to distinguish between a single dispersal event to the archipelago, followed by allopatric differentiation, or two separate introductions from the continent to both Tristan and Gough.

As is inevitably the case with isolated island populations, the question of whether G. nesiotis and G. comeri were reproductively isolated under natural circumstances cannot be answered. Our limited data from a small number of specimens and sequences of only the mitochondrial lineage are insufficient to demonstrate hybridisation. Even though island populations generally show lower genetic variation than related mainland populations [24], the genetic distances between G. nesiotis and G. comeri are of at least the same magnitude as those found between taxa that figure as subspecies of G. chloropus in the literature (Fig. 5, Table 2). Therefore, we propose that the extinct moorhen of Tristan and the moorhens that live on Gough and Tristan today be regarded as subspecies, viz. G. n. nesiotis and G. n. comeri, respectively. This is in conformity with two recent, general checklists of the birds of the world [28], [29] and a detailed monograph of the rails of the world [13], but is different from a morphological study by Eber [10] in which both taxa are considered synonyms.

Table 4 PCR and sequencing primers.

All PCR products from G. nesiotis and a number of PCR product from the D-loop of selected taxa (Table 3) were cloned using either pGEM®-T Easy Vector Sytems from Promega or Topo TA Cloning® from Invitrogen. At least three colonies were picked per plate and used to initiate reamplifications with primers 21M13_F and 21M13_R (Table 4). Reamplified products were cleaned using a Nucleospin® kit (Macherey-Nagel). Subsequently these products were sequenced either in-house on a Megabace™ 1000 DNA Analysis System (Amersham), or on a 3730xl DNA analyzer (Applied Biosystems) at Macrogen Inc. (Korea) using only primer 21M13_F. All other PCR products were cleaned (same procedure) and sequenced directly (both directions) with their respective PCR primers (Table 4). A summary of the specimens and the number of colonies sequenced per target region is given in Table 3. Sequences were assembled using Sequencher version 4.2 (Gene Codes Corporation) and aligned manualy using MacClade version 4.08 [31]. The sequences were deposited in GenBank (accession numbers EF681971-EF682015).

We thank H. van Grouw (National Museum of Natural History Naturalis, Leiden, The Netherlands), Dr R. Prys-Jones (The Natural History Museum, Tring, U.K.) and T. Prins (Zoological Museum, University of Amsterdam, the Netherlands) who provided most valuable material for DNA analyses. We also thank H. Cross (National Herbarium the Netherlands, Leiden) for technical assistance, K. Vrieling (Leiden University, the Netherlands) for putting at our disposal the Topo TA Cloning® System (Invitrogen) and P. Brakefield (Leiden University, the Netherlands) for reviewing our manuscript. Finally we acknowledge the Tristan da Cunha Post Office, who gave permission to illustrate this manuscript with some of their stamps (Fig. 1) and we acknowledge the Crown Agents Stamp Bureau for facilitating correspondence.