Snake-necked turtles (Chelodina, subgenus Chelydera) of northern and northwestern Australia have a long and checkered history of taxonomic unrest operating until today. Snake-necked turtles from northwestern Australia were first mentioned by Gray (1873), who stated that the collection of the British Museum included (as we know today, misidentified) Chelodina oblonga specimens from the Port Essington region in the far north of the present Northern Territory. Gray (1841) had described C. oblonga earlier with the rather vague type locality of “Western Australia”. However, the species actually occurs in southwestern Australia (Fig. 1; see Kehlmaier et al. 2019; TTWG 2021). Gray’s specimens from northwestern Australia represented what was later named C. kurrichalpongo by Joseph-Ouni et al. (2019; see below). Chelodina kurrichalpongo was considered before part of C. rugosa sensu lato, a species originally described by Ogilby (1890) from Cape York (Queensland). Two other early and contentious species descriptions are those of Werner (1901) and Fry (1915). The Austrian herpetologist Werner named C. siebenrocki in 1901 from the former German colony in New Guinea (“Deutsch-Neu-Guinea”), which was located in northeastern New Guinea, remarkably c. 450 km distant from the known distribution range of Chelydera turtles in southern New Guinea (Kehlmaier et al. 2024). Fry described C. intergularis in 1915 without locality data. The resulting taxonomic issues were the focus of a recent study (Kehlmaier et al. 2024). Using historical and genetic evidence, Kehlmaier et al. (2024) reconstructed the provenance of the type specimen of C. intergularis and restricted its type locality to Somerset, Cape York Peninsula (Queensland). Chelodina siebenrocki and C. intergularis are currently regarded as junior synonyms of C. rugosa sensu stricto (TTWG 2021; Kehlmaier et al. 2024). In the early 21^st century, a misidentification of the putative holotype of C. oblonga as representing a northern Australian species (Thomson 2000) caused additional nomenclatural upheavals, resulting in changes of various subgenus and species names (Table 1; see also Georges and Thomson 2010; Kuchling 2010; Cann and Sadlier 2017; TTWG 2017; Kehlmaier et al. 2019; Shea et al. 2020). Kehlmaier et al. (2019) eventually corrected this situation by demonstrating that the name-bearing type of C. oblonga represents the southwestern Australian species and resurrected the name C. rugosa for the northern species (see Shea et al. 2020 and TTWG 2021 for a detailed discussion of these nomenclatural issues).

Figure 1. 

Distribution ranges of Chelodina species (subgenus Chelydera), Chelodina oblonga (subgenus Macrochelodina), and C. canni, one of the species of the subgenus Chelodina sensu stricto hybridizing with Chelydera species. Distribution ranges according to TTWG 2021; the range of C. walloyarrina has been expanded using unpublished data (G. Kuchling). Note overlapping ranges. The grey areas correspond to the exposed continental shelf during the glacial sea level approximately 12,000 to 11,000 years ago. Abbreviations: LB – Lake Bonaparte, LC – Lake Carpentaria.

Using immunoelectrophoresis, Burbidge et al. (1974) compared a putative C. rugosa from Darwin, Northern Territory, and a morphologically different individual (flagged as “C. rugosa?”) from Kalumburu in the northern Kimberley region of Western Australia and found “a strong affinity” between them. The turtle from the Kimberley later became the holotype of C. kuchlingi, but the original description (Cann 1997) neither mentioned its previously reported immunological relationship to turtles from the Northern Territory, nor an additional specimen with similar morphology and the same collection data, nor another specimen from the same collector (Harry Butler) with similar morphology but from a different locality, all in the Western Australian Museum (Cann and Sadlier 2017; Kuchling 2020). Chelodina kuchlingi was listed as a distinct species by Georges and Thomson (2006) and considered as valid by the Turtle Taxonomy Working Group in 2007 (TTWG 2007) but, based on anecdotal information questioning its type locality, later synonymized with C. rugosa sensu lato (Georges and Thomson 2010; TTWG 2010, 2011, 2012). After a four-year hiatus, C. kuchlingi was resurrected by the Turtle Taxonomy Working Group (TTWG 2014). However, a follow-up appraisal of the C. kuchlingi holotype by the Western Australian Museum did not consider the evidence provided by the two additional specimens in the same museum and synonymized C. kuchlingi again under C. rugosa sensu lato (Ellis and Georges 2015). In contrast, Cann and Sadlier (2017) recognized C. kuchlingi as valid and published photographs of all three then known museum specimens collected by Harry Butler in 1965. Eventually, Kehlmaier et al. (2019) confirmed the validity of C. kuchlingi based on its deeply divergent mitogenome, placing its holotype basal to two studied specimens of C. rugosa and the holotype of C. siebenrocki. The geographic provenance (“Kalumburu”) of the holotype of C. kuchlingi has since been emended to Parry Creek on the lower Ord River floodplain (Kuchling 2020). The same study revealed a fourth museum specimen of C. kuchlingi collected by John Legler in 1974. This turtle is the last known unambiguous specimen of C. kuchlingi (Kuchling 2020; but see Smales 2019).

In addition to the mentioned taxa, Thomson et al. (2000) described C. burrungandjii from the Arnhem Land Sandstone Plateau in the Northern Territory and diagnosed it morphologically from its closely related and then undescribed sister taxon “Chelodina sp. (Kimberley)” of the Kimberley Sandstone Plateau. However, no fixed allelic differences between the populations from Arnhem Land and Kimberley were found in an allozyme electrophoretic study (Georges et al. 2002). Subsequently, McCord and Joseph-Ouni (2007) named the population from the Kimberley Sandstone Plateau C. walloyarrina. The Kimberley taxon was not differentiated from C. burrungandjii from Arnhem Land in an R35 phylogeny, but represented another mitochondrial lineage (Alacs 2008). Chelodina walloyarrina was and is listed as a valid species by the Turtle Taxonomy Working Group (Rhodin et al. 2008; TTWG 2010, 2011, 2012, 2014, 2017, 2021), but was synonymized with C. burrungandjii by Georges and Thomson (2010), Thomson et al. (2011), Ellis and Georges (2015), and Petrov et al. (2023), despite morphological and mitochondrial differences. Shea et al. (2020) considered the Kimberley taxon a subspecies of the Arnhem Land taxon using the name combination C. burrungandjii walloyarrina.

Kehlmaier et al. (2019) suggested that the name C. intergularis Fry, 1915 might apply to the western mitochondrial lineage of C. rugosa sensu lato (i.e., from the Northern Territory), but Kehlmaier et al. (2024) established that C. intergularis is in fact a junior synonym of C. rugosa sensu stricto (i.e., from Queensland). Largely based on Kehlmaier et al.’s (2019) mitogenome study that provided evidence that western C. rugosa sensu lato may represent a separate taxon, Joseph-Ouni et al. (2019) described C. kurrichalpongo from the Northern Territory. Chelodina kurrichalpongo was recognized as a valid species by the Turtle Taxonomy Working Group (TTWG 2021), but the nomenclatural availability of its name is still disputed by some, including by the “Official List of Species of Australian Reptiles and Amphibians” of the Australian Society of Herpetologists (ASH 2025) because the original description appeared in a publication not subjected to rigorous scientific peer review.

The taxonomic and nomenclatural conundrum of northern snake-necked turtles had and has implications for conservation assessments and funding of surveys and species conservation. Under the Australian “Environment Protection and Biodiversity Conservation Act 1999” (EPBC Act), for taxa to be considered in environmental impact assessments or for listing in threatened categories (broadly similar to the IUCN Red List), the taxon must be included in the Australian Faunal Directory (AFD (2020)). Neither C. walloyarrina nor C. kurrichalpongo are currently listed there, nor was C. kuchlingi until 2019.

For the purpose of the EPBC Act, both species and subspecies are treated as “species”, so both taxonomic levels are relevant, but the concept of Evolutionarily Significant Units (ESU) is not formally recognized in legislation (Petrov et al. 2023). The EPBC Act focuses on “species” of the threatened categories “Vulnerable,” “Endangered”, and “Critically Endangered,” but listing as “Data Deficient” automatically removes taxa from further consideration; this category has no statutory implications. Petrov et al. (2023) as well as the “Official List of Species of Australian Reptiles and Amphibians” (ASH 2025) assign the conservation status “Possibly Extinct” to C. kuchlingi, even though this category does not exist under Australian legislation. If a species does not qualify for listing as “Extinct,” it falls into the category “Critically Endangered” and, since 30 April 2024, C. kuchlingi has been classified as “Critically Endangered” under Order 2024 of the Biodiversity Conservation Act 2016 of Western Australia (State of Western Australia 2024). However, pursuant to a Memorandum of Understanding between the Commonwealth and the State of Western Australia, adoption of this listing under the federal EPBC Act has not yet been finalized. In any case, this classification would be in line with the conclusion of Garnett et al. (2022), who modeled a 70% likelihood that C. kuchlingi is already extinct and a 45% likelihood of extinction by 2041 if undiscovered populations remain.

What could be the reason that no specimens with the morphology of C. kuchlingi, a species presumably endemic to the Ord River floodplain (Kuchling 2020), have so far been collected after 1974? The Ord River Diversion Dam at Kununurra, Western Australia, was opened in 1963 when large-scale irrigation agriculture commenced on the Ord River floodplain. The first stage of the Ord River Dam was completed in 1973, creating Lake Argyle, Australia’s largest dam reservoir. Biodiversity and ecological values were not considered prior to building the dams and no surveys were commissioned to understand the ecological system (land, water, plants, and animals together) or the impacts of the dams upon that system. Since then, wet season flows of the Ord River below the dams were reduced and dry season flows increased. This effectively transformed the former “dry tropics” river with high seasonal flow variation and strong wet season flood pulses into a “wet tropics” river with a permanent flow regime, minimizing inundation and allowing the now dry floodplains to be converted for year-round irrigation agriculture (Kuchling 2020). Environmental impact assessments were prescribed under the EPBC Act for the Ord River stage 2 developments in the early 21^st century and included aquatic fauna surveys (e.g., Wetland Research and Management 2013), but excluded freshwater turtles, since all species then listed in the Australian Faunal Directory were considered widespread, common, and of no conservation concern. Until 2019, Australian authorities and many Australian researchers considered C. kuchlingi to be synonymous with C. rugosa sensu lato. Thus, at a critical time the opportunity was lost to further the knowledge and to address the conservation needs of C. kuchlingi through the prescription of targeted predevelopment surveys.

It is obvious that a robust taxonomy and nomenclature is urgently needed for solid conservation planning for the turtle fauna of northwestern Australia. The taxonomy of the snake-necked turtles of northern Australia is complicated by their propensity to hybridize with other taxa in geographic contact zones (Georges et al. 2002; Alacs 2008; Thomson et al. 2011). Range-wide phylogeographic and population genetic studies have reportedly been underway at the University of Canberra for well over a decade (see Alacs 2008; Thomson et al. 2011; Kennett et al. 2014), but no published results are so far available. Currently, different classifications compete (Table 1), also compromising the identification of published DNA sequences. We provisionally accept the classification outlined in Table 1 as a starting point for the present study, following the most recent checklist of the Turtle Taxonomy Working Group (TTWG 2021). Figure 1 summarizes the distribution ranges of the taxa relevant to the present investigation.

Our study does not attempt to duplicate any studies underway by others. We focus on the lower Ord River floodplain and its border with the sandstone plateau of northeastern Western Australia, where specimens with C. kuchlingi morphology have been collected over the last century, where recent range expansions of other taxa are suspected (Kuchling 2020), possibly leading to hybridization with C. kuchlingi, where limited sampling by other research groups has occurred, and where the area has been heavily impacted by ongoing water resource and agricultural development for over half a century.

Our aims are (1) to re-examine the genetic distinctiveness of C. kuchlingi using its holotype and the other historical museum specimens, (2) to determine whether C. kuchlingi is genetically distinct from the extant Chelodina population on the Ord River floodplain, and (3) to compare the Ord River population with other Chelodina species using new and previously published DNA sequences. Our datasets are near-complete mitochondrial genomes and sequences of up to 11 nuclear genomic loci generated from historical museum specimens and fresh material combined with previously published data (File S1).

Tissue samples from 21 snake-necked turtles collected between 2006 and 2019 and eight tissues from six museum specimens collected between 1965 and 2004 were studied. In addition, DNA from a historical museum specimen collected in the second half of the 19^th century from a previous study (Kehlmaier et al. 2024) was processed (File S1). One specimen from 1965 is the holotype of Chelodina kuchlingi (Western Australian Museum WAM R29411) and was originally preserved in formalin and later transferred to ethanol, as were four other museum specimens. Another museum specimen (Northern Territory Museum NTM R28391) was most likely ethanol-preserved from the beginning. Tissues from the remaining 21 turtles were immediately preserved in high-quality ethanol. Two of these tissues originate from fresh shells, 19 from live turtles (File S1). Negative controls were processed along with all of these samples and screened for contamination. DNA concentration and quality were assessed using a Qubit 3.0 Fluorometer (Thermo Fisher Scientific) and a 4200 TapeStation system (Agilent).

DNA from the 21 ethanol-preserved tissues was extracted using the innuPREP DNA Mini Kit 2.0 (Analytik Jena) with a final elution of twice 50 µl milliQ water and incubation at room temperature for 5 min. Where necessary, DNA was sheared to approximately 150-bp-long fragments using a Covaris M220 Ultrasonicator before conversion to single-indexed double-stranded Illumina DNA libraries according to Meyer and Kircher (2010).

DNA from the eight tissues from museum specimens was extracted according to Straube et al. (2021) in a dedicated clean room facility, combining a Proteinase K lysis buffer adapted from Sambrook and Russell (2001) with an extraction protocol optimized for very short DNA fragments from Dabney et al. (2013). DNA extracts were eluted in 2x 15 µl TET buffer with incubation at room temperature for 5 min before conversion to double-indexed, single-stranded Illumina DNA libraries according to Gansauge and Meyer (2019). DNA from a historical bone sample from the holotype of C. intergularis (Australian Museum AM R6255) from the second half of the 19^th century originates from a previous study (Kehlmaier et al. 2024) and was processed like the other museum material.

To increase the amount of endogenous library molecules, all DNA libraries were subjected to two rounds of in-solution hybridization capture in a dedicated capture-only workspace using DNA baits generated from PCR products (Maricic et al. 2010; Horn 2012). Details of PCR and bait library preparation are provided in File S2. Sequencing was performed on an Illumina MiSeq platform, generating 75 bp paired-end raw reads (3.5 million for the ethanol-preserved samples and up to 20 million for the degraded samples extracted from museum specimens).

In addition to the mitogenome, the 10 most informative nuclear loci studied by Thomson et al. (2021) were selected, based on their genetic divergence between Chelodina species (File S2): Aryl hydrocarbon receptor 1 (AHR, partial, coding), bone morphogenetic protein 2 (BMP2, partial, coding), high mobility group protein B2 (HMGB2, partial, coding and intron), hepatocyte nuclear factor 1 homeobox A (HNF1A, partial, intron 2), paired box protein (PAX1P1, partial, intron), proteasome 26S subunit (PSMC1, partial, coding and intron), RNA fingerprint protein 35 (R35, partial, intron 1), recombination activating protein 1 (Rag-1, partial, coding), and the non-coding anonymous loci TB01 and TB73. The partial ornithine decarboxylase locus (ODC, coding and introns), known to be useful for resolving turtle relationships (Praschag et al. 2017; Fritz et al. 2023, 2024), was added.

After adapter trimming using Skewer 0.2.2 (Jiang et al. 2014), read merging (minimum length 35 bp), quality filtering (minimum Q score 20), and duplicate removal using BBmap-suite 37.24 (https://sourceforge.net/projects/bbmap; Bushnell et al. 2017), the remaining reads were screened for contamination using FastQ Screen 0.11.4 (Wingett and Andrews 2018) and a set of predefined mitochondrial sequences (Bacillus, Bos, Canis, Cyprinus, Ecoli, Felis, Gallus, Homo, Mus, Penicillium, Sula, Sus, Ursus). Mitogenome and nuclear loci were assembled using MITObim (Hahn et al. 2013) and a two-step baiting and iterative mapping approach with an allowed mismatch value of 2. Selected GenBank sequences (depending on the target locus) were used as starting seeds, e.g., HQ172157 (Chelodina rugosa) for mtDNA. The resulting contigs were visualized and checked for assembly artifacts in Tablet 1.21.02.08 (Milne et al. 2013). Artifacts were manually removed from assembled contigs and all positions with coverage less than 3x were masked as ambiguous (N) using the maskfasta subcommand of BEDTools 2.29.2 (Quinlan and Hall 2010). Sequence length distribution of mapped reads was calculated using a customized awk command and Microsoft Excel. The contigs were aligned to the complete mitogenome of C. oblonga (KY776449) and annotated accordingly. For selected species or sequences, uncorrected p distances were calculated using MEGA X (Kumar et al. 2018).

Mitogenome phylogeny for our sequences and additional GenBank sequences (File S1) was inferred using maximum likelihood and Bayesian inference approaches as implemented in RAxML 8.0.0 (Stamatakis 2014) and MrBayes 3.2.6 (Ronquist et al. 2012). The GenBank data included many mitogenome sequences from type material from two previous studies (Kehlmaier et al. 2019, 2024). The processed mitochondrial alignment (15,893 bp) included 49 Chelodina sequences (26 from GenBank; 23 newly generated) and Elseya flaviventralis (KY776454) as outgroup, i.e., all available mitogenome sequences for the genus Chelodina. In addition, a phylogenetic network was constructed for nuclear loci using SplitsTree 4.19.0 (Bryant and Huson 2023) and the implemented NeighborNet algorithm. For this purpose, all nuclear loci were phased using DnaSP 6.12 (Rozas et al. 2017) and, since the algorithm is sensitive to missing data, the alignment had to be pruned. Two nuclear loci (ODC and HNF1A) were discarded due to their fragmentary coverage across the dataset. In addition, concatenated sequences with more than 5% missing data were removed, resulting in an alignment of nine concatenated nuclear loci and a total length of 6071 bp. It contained 80 phased sequences from 40 Chelodina specimens (20 individuals from GenBank; 20 new). Another alignment included more samples but more missing data (up to 43%) using the same nine loci, with 96 phased sequences from 48 individuals (27 individuals from GenBank; 21 new). The SplitsTree calculations were executed with default parameters, except for turning on the “ignore ambiguous states” option. Details on samples, DNA sequences, and individual analyses are provided in Files S1 and S2. The alignments are available as Files S3–S5.

Three museum specimens and one fresh sample did not yield any DNA of sufficient quality for further processing, among them two specimens of Chelodina kuchlingi (WAM R28116, WAM R28117). The remaining material, including the holotype of C. kuchlingi (WAM R29411) and another museum specimen of C. kuchlingi (WAM R177909), could be used for sequencing mitochondrial and nuclear DNA. A few samples worked only for mtDNA (File S1). For the holotype of C. kuchlingi, an enhanced mitochondrial genome is presented compared to that in Kehlmaier et al. (2019). Details of the revision process are provided in File S2. The DNA extracted from a bone sample of the holotype of C. intergularis (AM R6255), for which the mitochondrial genome was published in Kehlmaier et al. (2024), was also of sufficient quality to sequence some nuclear loci (File S1).

Using mitogenome sequences, the two tree-building methods yielded identical and well-supported branching patterns that conflicted with the three currently recognized subgenera Chelodina, Chelydera, and Macrochelodina (Fig. 2). The mitogenome of C. parkeri (subgenus Chelydera) constituted the sister lineage to all other Chelodina species. One well-supported clade contained all species of the subgenus Chelodina. Within this clade, C. steindachneri was sister to the remaining taxa. However, among the species of the subgenus Chelodina, some representatives of the subgenus Chelydera were interspersed. One C. expansa (subgenus Chelydera, GenBank accession number KY776450) was sister to the mitogenome sequence of the holotype of C. longicollis (subgenus Chelodina), and this maximally supported clade was sister to a well-supported more inclusive clade containing the remaining representatives of Chelodina sensu stricto plus some other representatives of the subgenus Chelydera. Within this more inclusive clade, sequences of C. walloyarrina (subgenus Chelydera) and another sequence of C. expansa (KY705230) occurred together with C. mccordi, another sequence of C. longicollis and a sequence of C. canni (all subgenus Chelodina) in a maximally supported clade that was sister to another maximally supported clade containing the mitogenome sequences of C. gunaleni, two sequences each of C. novaeguineae and C. reimanni, and a sequence of C. pritchardi (all subgenus Chelodina). The two sequences of C. novaeguineae were not monophyletic; one (LR215676) was very similar to two sequences of C. reimanni, with which it was placed in a maximally supported clade. The other C. novaeguineae sequence (KY776446) was with high support sister to the mitogenome sequence of C. pritchardi.

Figure 2. 

Maximum likelihood tree for mitogenomes of Chelodina species, rooted with Elseya flaviventralis. Numbers at nodes are bootstrap support values and posterior probabilities from a Bayesian tree of the same topology; asterisks represent maximum support under both approaches. The three subgenera are indicated by different colors; putative C. kurrichalpongo from the Ord River floodplain, flagged with question marks. For type material, scientific names are followed by abbreviations for the type status (HT – holotype, LT – lectotype, PT – paratype). For all samples, the GenBank/ENA accession numbers or sample codes are shown (see also File S1). For focus taxa of the present study, the geographic origin and, as far as possible, the collection year are given. Abbreviations: NT – Northern Territory, QLD – Queensland, WA – Western Australia. Western Australian turtles color-coded to highlight different taxa. Note that C. expansa and C. walloyarrina (subgenus Chelydera) are placed among species of the subgenus Chelodina sensu stricto.

The mitogenomes of two sequences of C. oblonga, placed in a maximally supported clade, were with high support sister to the mixed Chelodina-Chelydera clade, and another more inclusive and maximally supported clade contained the mitogenome sequences of C. rugosa, C. kuchlingi, C. kurrichalpongo, and putative C. kurrichalpongo from the Ord River floodplain (flagged with question marks in Fig. 2). The sequences of C. kuchlingi, including its holotype, C. kurrichalpongo, and putative C. kurrichalpongo showed little variation (average uncorrected p distance of 0.09%, range: 0–0.43%) and occurred in a maximally supported clade that was sister to another maximally supported clade containing four sequences of C. rugosa. Within the latter clade, the sequence of the holotype of C. siebenrocki from New Guinea was sister to three very similar sequences, one each from Papua New Guinea (KY705234), the holotype of C. intergularis from the Cape York Peninsula, Queensland (OY859728), and a pet trade turtle (HQ172157).

The SplitsTree analysis using the alignment with up to 5% missing data reflected the three currently recognized subgenera of Chelodina (Fig. 3). All included species of the subgenus Chelodina clustered together on one side of the NeighborNet, as did the species of the subgenus Chelydera on the other. Chelodina oblonga, the only representative of the subgenus Macrochelodina had an intermediate position on a long branch. Within the subgenus Chelodina, the two individuals each of C. longicollis and C. steindachneri represented distinct long branches with a few internal reticulations; C. longicollis was placed farther from the remaining species of Chelodina sensu stricto than C. steindachneri. Also, the two alleles of a C. pritchardi represented a relatively long branch that was, however, very close to a reticulated and little differentiated terminal cluster that included the sequences of C. canni, C. gunaleni, C. mccordi, C. novaeguineae, and C. reimanni. The latter five species were weakly differentiated and C. gunaleni, C. novaeguineae, and C. reimanni were not differentiated at all. Within the subgenus Chelydera, C. expansa was the most distant species, followed by C. parkeri, each representing a very long branch that resembled the length of that of C. oblonga (subgenus Macrochelodina). However, the remaining samples and species of the subgenus Chelydera (C. kuchlingi, C. kurrichalpongo and putative C. kurrichalpongo, C. walloyarrina) constituted a weakly resolved terminal cluster with five indistinct subclusters. All four alleles of two C. kurrichalpongo sensu stricto (Northern Territory) formed a subcluster on their own, but none of the four remaining subclusters representing Western Australian sequences corresponded to a single taxon, and the alleles of some individuals occurred in two distinct subclusters (see enlarged inset of Fig. 3).

Figure 3. 

NeighborNet using concatenated sequences of nine nuclear loci (6071 bp; 40 individuals) with up to 5% missing data. The three subgenera are indicated by different colors. The terminal clusters containing species of the subgenus Chelodina sensu stricto (left) and Chelydera (right) are enlarged, showing the individual alleles (1, 2 following the sample codes; File S1). Note that alleles of the same individual may occur in different subclusters. Putative Chelodina kurrichalpongo (with question marks) from the Ord River floodplain are only indicated by their GK codes. Abbreviations: NT – Northern Territory, WA – Western Australia. Alleles from Western Australian turtles color-coded to highlight different taxa. Inset right shows putative C. kurrichalpongo (GK37 = WAM R150821, Parry Creek, WA).

The NeighborNet with more samples but up to 43% missing data showed the same general topology (Fig. 4). As a consequence of more missing data, some reticulations increased, in particular in the two terminal clusters corresponding to species of the subgenera Chelodina sensu stricto and Chelydera. The terminal cluster with species of Chelodina sensu stricto showed four subclusters. One relatively distinct subcluster each corresponded to ten allele sequences of C. mccordi (five individuals), two allele sequences (same individual) of C. novaeguineae, and two sequences of C. pritchardi (also from the same individual). The fourth and largest subcluster contained allele sequences of C. canni (one individual), C. gunaleni (one individual), C. novaeguineae (two individuals), and C. reimanni (three individuals), which were all little differentiated. In the terminal cluster containing Chelydera species, the additional allele sequences of C. rugosa sensu stricto (Cape York Peninsula) were placed on long branches in a relatively distinct basal subcluster that contained, however, also one sequence of C. kurrichalpongo. The long branches of C. rugosa could be caused, at least partially, by the missing data in these sequences. The remaining sequences of C. kurrichalpongo, one of them from the same individual as the allele clustering with C. rugosa, were placed in another basal subcluster, whereas all C. kurrichalpongo alleles clustered together in the analysis including only more complete sequences but fewer taxa (Fig. 3; without C. rugosa!). The remaining weakly distinct subclusters in the expanded analysis (Fig. 4) contained sequences of C. kuchlingi, putative C. kurrichalpongo from the Ord River floodplain, and C. walloyarrina, without any clear pattern.

Figure 4. 

NeighborNet using concatenated sequences of nine nuclear loci (6071 bp; 48 individuals) with up to 43% missing data. The three subgenera are indicated by different colors. The terminal clusters containing species of the subgenus Chelodina sensu stricto (left) and Chelydera (right) are enlarged, showing the individual alleles (1, 2 following the sample codes; File S1). Note that alleles of the same individual may occur in different subclusters. Putative Chelodina kurrichalpongo from the Ord River floodplain are only indicated by their GK codes. Abbreviations: NT – Northern Territory, QLD – Queensland, WA – Western Australia. Alleles from Western Australian turtles color-coded to highlight different taxa.

Morphological differences among the northern Australian taxa in the subgenus Chelydera are impressively illustrated by a large number of photographs in Cann and Sadlier (2017). Solely morphological characters were used for the original descriptions of C. rugosa, C. kuchlingi, C. burrungandjii, C. walloyarrina, and C. kurrichalpongo (Ogilby 1890; Cann 1997; Thomson et al. 2000; McCord and Joseph-Ouni 2007; Joseph-Ouni et al. 2019).

Chelodina rugosa was described based on a single shell (Ogilby 1890). In the description of C. kuchlingi, Cann (1997) used its oval carapace and round head profile to distinguish it from the then undescribed sandstone plateau forms (C. burrungandjii and C. walloyarrina) and the radiating rugosities of the carapace scutes to diagnose it from C. rugosa (including the later described C. kurrichalpongo). In their description of C. burrungandjii, Thomson et al. (2000) provided for all five, then partly still undescribed, species the most detailed morphological analyses based on head and shell characters. Calculating multivariate distances, the distinction was 100% accurate between the groups [C. rugosa + C. kurrichalpongo + C. kuchlingi, i.e., the coastal plain taxa] and [C. burrungandjii + C. walloyarrina, i.e., the sandstone plateau taxa]. Using a continuous series of 3–5 exposed neurals in C. burrungandjii, they could reliably separate this species from C. walloyarrina. The females, but not the males, of the two taxa were also distinct in the canonical discriminant space. McCord and Joseph-Ouni (2007) described C. walloyarrina largely by referring to the morphological analysis in Thomson et al. (2000). In their description of C. kurrichalpongo, Joseph-Ouni et al. (2019) referred to differences in head and shell measurements and coloration compared to other taxa (without providing details of data and material) as well as to the deeply divergent mitogenome of a Northern Territory turtle published by Kehlmaier et al. (2019).

Apart from a broad shell, the main diagnostic trait for C. kuchlingi is its radiating rugosities on all carapace shields. This character is present in all four putative C. kuchlingi specimens in the Western Australian Museum collected between 1965 and 1974 (Burbidge et al. 1974; Cann 1997, 1998; Cann and Sadlier 2017; our Fig. 5). Radiating rugosities are also frequently found in juvenile C. canni (see Cann and Sadlier 2017), a representative of the subgenus Chelodina sensu stricto, and also rarely in juveniles of the subgenus Chelydera. Photos in Cann and Sadlier (2017) show radiating rugosities in a “week-old” hatchling of C. kurrichalpongo (p. 59), a juvenile C. burrungandjii of approximately 15 cm shell length (p. 64), and a juvenile C. expansa (p. 101). We examined 36 C. walloyarrina in the Western Australian Museum and found that seven juveniles and subadults also show this trait. What appears exceptional in C. kuchlingi is that the two adults collected half a century ago maintained these radiating rugosities.

Figure 5. 

Western Australian snake-necked turtles. All specimens were genetically studied. A Chelodina kuchlingi, holotype, Parry Creek (GK9, Western Australian Museum WAM R29411, collected 1965); B C. kuchlingi, Parry Creek (GK8, Western Australian Museum WAM R177909, collected 1974); C putative C. kurrichalpongo (in life), Parry Creek (GK38, Western Australian Museum WAM R150885, collected 2019); D C. walloyarrina, Cockburn Creek (GK21, Northern Territory Museum NTM R34788, collected 2006; this specimen was identified as C. kuchlingi by Smales (2019). Scale bars = 5 cm.

How consistent are the morphological characters among all these taxa? Figures 5 and 6 show photos of the carapace of five specimens for which DNA was analyzed in this study plus three additional specimens representing one C. burrungandjii and two C. walloyarrina from the western area of its distribution. While the morphology of a C. walloyarrina from the Mitchell River (Fig. 6C) is in agreement with the description of the species, the specimen from Kimbolton (southwestern Kimberley) shows morphological characters considered typical for C. kuchlingi (radiating rugosities of carapace scutes and brown color; Fig. 6B). Specimens from the single locality Parry Creek (Fig. 5A–C) are morphologically different and a specimen from Cockburn Creek (Fig. 5D) representing the mitochondrial lineage of C. walloyarrina had previously been identified as “C. kuchlingi?” by Smales (2019) based on this trait.

Figure 6. 

Snake-necked turtles from Western Australia and the Northern Territory. Specimen bearing GK code was genetically studied. A Chelodina kurrichalpongo, Bullo River, Northern Territory (GK2, Northern Territory Museum NTM R28391, collected 2004); B C. walloyarrina, Stewart River, Kimbolton, Western Australia (Western Australian Museum WAM R51829, collected 1975); C C. walloyarrina, Mitchell River, Western Australia (Western Australian Museum WAM R141069, collected 1998); D C. burrungandjii, Katherine, Northern Territory (Western Australian Museum WAM R26838, collected 1966). Scale bars = 5 cm.

In addition, the presence or absence of a series of 3–5 neurals was considered another diagnostic trait to distinguish C. burrungandjii from C. walloyarrina (Thomson et al. 2000; McCord and Joseph-Ouni 2007). However, Smales (2019), while finding neurals to be present at the type locality of C. burrungandjii (Koolpin Gorge, South Alligator River) and routinely absent in C. walloyarrina, found neurals may be absent, or discontinuous and isolated in C. burrungandjii from other Arnhem Land locations than the type locality (e.g., Maningrida and Mann River). A complete series of 5–8 neurals is typical in C. oblonga and appears for Chelodina to be a phylogenetically uninformative plesiomorphic trait (compare also Burbidge et al. 1974; Pritchard 1988). According to Smales (2019), one or two isolated neurals may occur in occasional individuals of C. rugosa (including C. kurrichalpongo) and he also reported one isolated neural in one of three “C. kuchlingi (?)” specimens he identified based on external morphology, revealing limited taxonomic utility of these traits.

Based on genetics, Alacs (2008) found that the mitochondrial lineage of C. burrungandjii corresponded both to morphotypes of C. rugosa sensu lato (including C. kurrichalpongo) and of what is currently known under C. burrungandjii. Moreover, the mitochondrial lineage of C. burrungandjii was closely related to that of the “C. rugosa West lineage” (i.e., C. kurrichalpongo) and paraphyletic with respect to the eastern lineage, i.e., C. rugosa sensu stricto. According to her results, unidirectional hybridization and introgression occurs, with C. burrungandjii males producing fertile hybrids with C. kurrichalpongo females, leading to hybrid swamping of C. burrungandjii. This situation suggests that the two taxa involved qualify as distinct species with incipient reproductive barrier. However, in the light of our SplitsTree analyses, it remains questionable whether the Northern Territory C. kurrichalpongo represent the same taxon as the morphologically similar turtles from the Ord River floodplain and whether C. walloyarrina is conspecific with C. burrungandjii, as suggested by Georges et al. (2002), Georges and Thomson (2010), Thomson et al. (2011), Ellis and Georges (2015), and Petrov et al. (2023), in particular because their ranges are separated by that of C. kurrichalpongo (Fig. 1).

Northern Australia experienced pronounced sea level changes during the glacial cycles of the late Pliocene and Pleistocene, and the rapid diversification of Kimberley fish species was triggered by associated spatio-temporal vicariant events (Shelley et al. 2020). The Pleistocene Last Glacial Maximum exposed a wide continental shelf, reaching its maximum about 12,000 to 11,000 years ago and connecting northeastern Australia and New Guinea (Fig. 1). This landmass surrounded two large lakes, Lake Carpentaria in the east and Lake Bonaparte in the west, both with brackish or fresh water, depending on the sea level stands (Yokoyama et al. 2001).

The fluctuating cycles of land emergence and submergence in northern and northwestern Australia likely corresponded to intermittent connections among turtle populations, but given the long generation times of snake-necked turtles compared to many fish species, the time scale of these events may have been too short for speciation. It may, however, have facilitated the evolution of local morphological differences during allopatric episodes. This could explain the morphological diversity in today’s snake-necked turtle populations of the Kimberley together with the lack of taxonomic resolution in the Western Australian terminal subcluster of our NeighborNets, where even the alleles of several individuals occur in distinct subclusters (Figs 3 and 4). The same scenario may have operated in southern New Guinea and northeastern Australia with respect to the terminal subcluster of our NeighborNets containing C. canni, C. gunaleni, C. novaeguineae, and C. reimanni.

Considering all evidence, we conclude that the morphological characters used in the past to diagnose northern Australian Chelydera species are less important than previously thought. According to our genetic data, only one species of the subgenus occurs in the Kimberley of Western Australia. Chelodina kuchlingi is not extinct, but widely distributed across the Ord River floodplain, and C. walloyarrina represents the same species, but with introgressed mitochondria. We therefore synonymize C. walloyarrina (McCord & Joseph-Ouni, 2007) with C. kuchlingi Cann, 1997. Further research is needed to clarify the status of C. kurrichalpongo sensu stricto, C. burrungandjii, and C. rugosa. The hybridization and incipient reproductive isolation reported by Alacs (2008) suggest that these northern Australian taxa represent a complicated mosaic within a speciation continuum that can only be incompletely reflected by Linnean nomenclature.