Neurals occur as bone elements of the turtle carapace dorsal to the vertebrae and in many turtles they form part of the skeletal surface of the carapace. Based on the presence and morphology of neurals in a wide variety of recent forms, Pritchard (1988) described the plesiomorphic chelonian condition as having a series of 8 neurals (1–8) that extends continuously from anterior contact with the nuchal to posterior contact with the suprapygal. In this condition each neural is roughly hexagonal, is widest anteriorly, and is longer than wide. He also described a variety of derived neural conditions across the extant turtle radiation, including patterns for differences in the shapes of individual neurals and for the number of neurals that are retained, from the original 8 through to none. Bona and Alcalde (2009) described an early ontogenetic process in the chelid Phrynops hilarii in which there were 8 neurals in all embryos examined but fusion of some neurals occurred subsequently, so that hatchlings and older animals had fewer than 8.

At the outset, it is worthwhile to briefly consider the origin and basic anatomical context of neurals in the shell of chelids. As detailed by Rieppel (2012) and MacCord et al. (2015), current understanding of the origin of neurals within the turtle shell has a background of 2 centuries of debate about whether neurals, like costals, ultimately originate from ectodermal, endodermal, or a combination of ectodermal and endodermal bony elements. Scheyer et al. (2008) and Hirasawa et al. (2013) have provided embryological evidence pointing strongly to a purely endodermal origin for neurals. They have shown that in extant turtles neurals do not develop as independent ossification centers but as initial outgrowths of the embryonic periosteal collars of the carapacial neural arches. As development progresses the ossification of the neural element continues without a distinct periosteum but by metaplastically ossifying precondensed soft tissue integumentary structures. Thus, as noted by Cherepanov (2016), neurals “are formed as a result of expansion of perichondral bone of the vertebral neural arches . . . neural plates are modified neural arches”. In simple terms, while the turtle literature often refers to the presence or absence of “neural bones,” a “neural” is not a separate element, but is a dorsal bony process of the neural arch in the position of the dorsal spine in other vertebrates. Bojanus (1819) correctly figured neurals as single bones, composed of a neural arch and a dorsal process, for the cryptodire Emys orbicularis in his seminal work on turtle anatomy.

The distinction between the neural arch and the dorsal “neural” process of the same bone has been made explicit by Scheyer et al. (2008). A neural is generally considered to be “present” when the process is exposed on the external skeletal surface of the carapace and I use the term “neural” in that sense. As a neural is part of the same bone as an underlying neural arch, the anatomy of the dorsal or thoracic vertebral column is important to understanding of the morphology of neurals. Vertebra 9, situated below costal pair 1 (C1), is the anteriormost of 10 dorsal vertebrae (V9–18) that are sutured to the carapace (Fig. 1). A significant aspect of the turtle bauplan is displacement between the vertebral centra and their corresponding neural arches so that a number of the thoracic neural arches each span parts of 2 vertebral centra (Goette 1899) and often suture to 2 pairs of overlying costals. The displacement commences anteriorly with a small anterior portion of neural arch 10 displaced forward to suture with the posterior of vertebral centrum 9 (Vc9). The subsequent neural arches are similarly displaced so that they also each suture with 2 vertebral centra. The degree of displacement diminishes toward the posterior of the carapace so that by V17 and caudad from it, the displacement between centra and neural arches no longer occurs and the neural arches and centra correspond to form entire vertebrae. In chelids that possess neurals, the anterior neurals may extend craniad to their underlying neural arches. In some taxa this anterior extension permits neural 1 to suture with the nuchal as it does in many cryptodires in what is presumed to be the plesiomorphic condition (Pritchard 1988).

View Figure

• Download as Powerpoint
• Download Figure

With this anatomy in mind, it can be seen that the position of any neural (1–8) is fixed, relative to adjacent bones on the carapace surface, by the position of its underlying neural arch and vertebral centra. Neural 1 is located above Vc9 and Vc10, neural 2 is located above Vc10 and Vc11, and so forth. Figure 1 illustrates the skeletal anatomy of chelid turtles with and without neurals. Neurals are routinely described on the basis of their shape on the dorsal skeletal carapace surface, which does not necessarily relate to the shape of the bone on the thoracic carapace surface.

Pritchard (1988) described the conditions of neurals in recent chelonians and provided a typology to indicate the number and dorsal shape of neurals. Neurals, especially those in contiguous series, are generally approximately symmetrical on the longitudinal axis (Karl and Tichy 2004). The greatest width of a neural usually coincides with the sutures of 2 adjacent pairs of costals between which it sits. Some longitudinally parallel-sided neurals do occur in a few chelid taxa such as Hydromedusa, where the dorsal surface of the bone is situated entirely between a single pair of costals. Isolated neurals on the carapace surface most often occur at the midline suture between 2 adjacent pairs of costals. In such cases, the neurals are also widest at the intersection of the costal pairs and may thus be roughly “kite-shaped” or somewhat irregular. Karl and Tichy (2004) discussed various polygons that are known for turtle neurals and the manner in which both the possible shapes and their configuration in a series are constrained by the shapes of adjacent neurals.

The condition of neurals varies considerably among the Chelidae. In some—exemplified by Chelus—neurals extend laterally to form a substantial central portion of the skeletal carapace, while in some other taxa the lateral expansion of neurals is limited and/or their number is reduced, and in many chelids the costals meet above the neural arches so that no neurals exist.

Within pleurodires, the Chelidae is considered to have diverged from the Pelomedusidae by the Late Jurassic at the latest (Lapparent de Broin and de la Fuente 2001). The earliest known chelids, Prochelidella from Patagonia, Argentina (de la Fuente et al. 2011), and at least 2 indeterminate chelid turtles from the Griman Creek Formation of New South Wales, Australia (Smith 2010), date from the early Cretaceous. All known chelid turtles are from South America and Australasia and the family is of undoubted Gondwanan origin. The earliest known chelid fossils indicate that by the early Cretaceous the family had a wide distribution across much of Gondwanaland (Ferreira et al. 2016). In a number of lineages within the extant Chelidae the series of neurals is reduced, discontinuous, or completely absent from the carapace surface (Pritchard 1988). Zangerl (1969) and Pritchard (2008) have noted that simplification of the shell by reduction of the number bones is a widespread trend in turtles.

The presence or absence and morphology of neurals in Australasian chelid turtles has been the subject of attention since Burbidge (1967) first reported the presence of neurals in Chelodina (Macrodiremys. colliei (as Chelodina oblonga). While they are present in a number of South American chelids (Pritchard 1988), prior to Burbidge (1967) it had been thought that neurals were absent from all Australasian forms (Boulenger 1889; Waite 1929; Zangerl 1948; Williams 1953). Subsequently, Rhodin and Mittermeier (1977) reported the occurrence of occasional small, isolated neurals in a number of both short- and long-necked Australasian chelid taxa.

Gaffney (1977) could not resolve the presence or absence of neurals in his consideration of hypotheses for phylogeny for the Chelidae. McDowell (1983) proposed a hypothesis, followed by Manning and Kofron (1996), for the evolutionary radiation of the Chelidae with emphasis on the condition of neurals as a primary character. Thomson and Georges (1996) discussed the presence and absence of neurals relative to a phylogeny for the majority of Australian taxa then recognized. They investigated the presence of neurals in Australasian chelids and reported their discovery in 3 taxa, of small isolated neurals as distinct bones embedded under the surface of the vertebral costals suture and thus not exposed on the skeletal carapace surface.

The fossil record for neurals has not previously been comprehensively reviewed for the Chelidae. Pritchard (1988) provided a detailed review of neurals in all extant chelonians recognized at that time, while reviews by Rhodin and Mittermeier (1977) and Thomson and Georges (1996) surveyed selections of extant chelid taxa. A further review is warranted because a number of new extant chelid taxa, principally Australasian forms, have been described since those reviews were published and the fossil record has improved, substantially so for South American forms. Considerable further efforts have also been made to resolve the phylogeny of the Chelidae since previous considerations of neurals were published. While consensus about some fundamental aspects of phylogeny of the family has not been reached, our understanding of relationships between many taxa is substantially improved since earlier investigations of neurals were made. In light of this improved knowledge it is worthwhile to reevaluate the extent to which the condition of neurals either improves our understanding of evolution within the family, or can be explained by it.

The purposes of this investigation were to

• further investigate the carapace structure of some chelids in order to describe and compare the skeletal anatomy of taxa that do and do not have neurals;

• provide a comprehensive review of neurals across the Chelidae, including all extinct and extant taxa; and

• consider the evolution of variable possession of neurals in the Chelidae in relation to available phylogenies for the family.

METHODS

Taxonomy of all fossil chelids known until 2016 follows the review of Maniel and de la Fuente (2016), with the addition of a further fossil chelid described by de la Fuente et al. (2017). Taxonomy for extant forms follows Turtle Taxonomy Working Group (2017) of the IUCN Tortoise and Freshwater Turtle Specialist Group. Citations for taxa not encompassed by that taxonomy are included for their descriptive content and are not an endorsement of the taxonomic or nomenclatural validity of any of such taxa.

Nomenclature for shell bones follows Zangerl (1969). Pritchard's (1988) typology provides a shorthand description for the dorsal shape of neurals and was expanded by Karl and Tichy (2004). The system describes the shape of each neural by a numeral representing the number of sides of the neural polygon in dorsal view. Neurals are frequently wider at one end than the other and this is denoted by an “A” or “P” indicating that the greater width is in the anterior or posterior part of the neural. I use “K” for a neural that is kite-shaped (“bigonal”, Karl and Tichy 2004). It may or may not be a rhombus but is roughly 4-sided and often with its greatest length on the midline axis of the carapace. K-type neurals occur as isolates only. A numeral without a letter indicates a neural that is either roughly parallel-sided on the longitudinal axis, or whose widest point is midway along its length.

I have added to Pritchard's system to provide a notation describing the entire configuration of neurals from the nuchal to the suprapygal. It is based on the concept that the primitive condition in chelids is a complement of 8 neurals. I use “Nu” for nuchal and “S” for suprapygal. A hyphen (-) indicates that 2 bones in a sequence are sutured to each other, a missing neural is denoted by an asterisk (*), 2 neurals existing in sequence but isolated from each other by a pair of costals with some medial contact is denoted by 2 back slashes (\\), and missing information (mainly where fossil remains are fragmentary) is shown by a question mark (?). Figure 2 illustrates a variety of neural shapes on a hypothetical carapace and provides an example of the descriptive typology for a neural series.

View Figure

• Download as Powerpoint
• Download Figure

Abbreviations of institution and other sources of specimens include the following: AC: author's collection; AM: Australian Museum, Sydney; AMNH: American Museum of Natural History, New York; BMNH: The Natural History Museum, London; FMNH: The Field Museum of Natural History, Chicago; JC: John Cann collection, Sydney; MAGNT: Museum and Art Gallery of the Northern Territory, Darwin; MCZ: Museum of Comparative Zoology, Boston; MV: Museum of Victoria, Melbourne; MZUFU: Museu de Zoologia João Moojen da Universidade Federal de Viçosa; QM: Queensland Museum, Brisbane; SMNS Museum fuer Naturkunde, Stuttgart, UFRJ: Museu Nacional Rio de Janeiro; UTG: Department of Geology, University of Tasmania, Hobart; WAM: Western Australian Museum, Perth; WM: William McCord collection, Hopewell Junction; YPM: Yale Peabody Museum of Natural History, New Haven; ZSM: Zoologische Staatssammlung, Munich; ZV: Zoos Victoria, Melbourne.

Anatomical Investigation of Neurals. — In order to obtain information about the anatomy of chelids with and without neurals, whole and disarticulated shells of chelid turtles in the collections of the following museums were examined: AC, AMNH, AM, JC, MCZ, MAGNT, MV, QM, BMNH, WAM, and WM. Selected fully ossified specimens were sectioned through the carapace, in the manner of Thomson and Georges (1996), to permit microscopic examination of the vertebrae and carapace bones in cross section. These shells were vertically sectioned on the transverse axial plane at a line passing through the posterior portion of costals pair 2. This position was chosen because it is in the portion of the carapace in which a neural tends to be retained even in chelid taxa with a substantially reduced neural series. The following specimens were examined in this way: Chelodina colliei (WAM R828, Lake Preston, Western Australia [W.A.]); Chelodina expansa (MV D75943 Gunbower Creek [Ck.], Victoria, Australia [Vic.]); Chelodina longicollis (MV D75945, Kerang Lakes, Vic.; AC, Lake Colac, Vic.; AC Gunbower Ck., Vic.); Emydura macquarii macquarii (MV D75944, Gunbower Ck., Vic.; D75946, Shepparton, Vic.). For comparative purposes, a specimen of the cryptodire Pseudemys concinna (ZV, unknown provenance) was sectioned in the same manner. A shell of C. longicollis (AC, Gunbower Ck., Vic.) was vertically sectioned on the midline sagittal plane.

Dissection was not permissible for specimens of a number of taxa and radiography and computed tomography (“CT scanning”) were used to examine relevant bone structures in these. Dorso-ventral radiographs were made of 2 specimens each of Acanthochelys pallidopectoris. Acanthochelys radiolata, and Acanthochelys spixii (WM, no locality data); 3 specimens of C. colliei (BMNH 40.12.9.81; WM, no locality data); and one specimen each of C. expansa (AC, Gunbower Ck., Vic.), Chelodina gunaleni (WM, southeastern Papua), C. longicollis (AC, Gunbower Ck., Vic.), C. oblonga (AM R175590, Cobourg Peninsula, Northern Territory, Australia [N.T.]), Chelodina reimanni (WM, southeastern Papua), Elseya branderhorsti (southeastern Papua), Emydura m. macquarii (AC, Kyalite River, New South Wales, Australia [N.S.W.]); Emydura m. binjing (JC, Clarence River, N.S.W.), and Emydura m. nigra (AC, Fraser Island, Queensland, Australia [Qld.]). CT scans in the sagittal and transverse planes were made of entire shells at 0.4-mm increments. The entire sequence of scans for each specimen thus entailed several hundred images, depending on the length and width of the carapace. Detailed examination of the entire sequence of CT scans along the length of the carapacial vertebrae was undertaken to ascertain the location and size of neurals where these were present. Individual images from CT scan sequences (Figs. 3 and 4) are provided as examples from the sequences. CT scans were made of the following specimens: Chelodina burrungandjii (MAGNT R13525, Mann River, N.T.; R16008, Koolpin Gorge, N.T.; R22582, Sleisbeck, N.T.; R36032, Maningrida, N.T.), Chelodina canni (JC, Limmen Bight River, N.T.), C. expansa (JC, Fraser Island, Qld.), Chelodina kuchlingi (?) (MAGNT R34788, Cockburn Ck., W.A.; MAGNT R34786, R34789, Parry Ck., W.A.), C. longicollis (AC, Gunbower Ck., Vic.), C. oblonga (AM R175587, R175589, Cobourg Peninsula, N.T.), Chelodina walloyarrina (AM R136148, Mitchell River, W.A.; R143558, Manning Gorge, W.A.; MAGNT R34787, Minah Ck., W.A.), Emydura m. kreffti (JC, Burnett River, Qld.), Emydura m. signata (JC, Brisbane River, Qld.), Emydura subglobosa worrelli (JC, Limmen Bight River, N.T.), Elseya irwini (JC, Broken River, Qld.), Elseya sp. (JC, Johnstone River, Qld.), and Myuchelys belli (JC, Macdonald River, N.S.W.).

View Figure

• Download as Powerpoint
• Download Figure

View Figure

• Download as Powerpoint
• Download Figure

Review of Neurals in the Chelidae. — The review entailed collation of information from the published literature for fossil and extant chelids. Additional information was obtained from my own examination of material as detailed in the Appendix. I sought and examined high-resolution photographs of some taxa that were not physically accessible and for which published information about neurals was not available.

Results of Anatomical Investigation of Neurals in Australian Chelids

The presence, absence, and other details of neurals, including neural shape and configuration formulae, for all taxa are summarized in the review portion of this article and are detailed in the Appendix.

In all taxa and specimens I examined that had no neurals on the carapace surface, the dorsal portion of neural arches formed a simple suture with the thoracic surfaces of costals spanning the costal midline suture (Fig. 3). In each of such cases the neural arch had minor lateral processes that were slightly wider than the narrow “waist” of the arch and that sutured to small medial portions of 1 or 2 pairs of costals (e.g., Fig. 3B).

In all examined specimens that possessed neurals, these were simple dorsal and dorso-lateral processes of neural arches (Fig. 4). For comparative purposes, a sectioned shell of the cryptodire Pseudemys concinna (ZV) is shown in Fig. 4C. This specimen has a relatively wide neural, typical of many nonchelid turtles, but as with the chelids examined, each neural is a process of a neural arch bone.

Review of Neurals in the Family Chelidae

The review here encompasses 85 taxa of the Chelidae. A table with information and references for conditions of neurals in every recognised chelid taxon is provided in the Appendix. For taxa that have a neural series, a representative sample of the neural formula is provided there. All specimens examined in my own investigation are also listed there.

The fossil record for chelids of South America offers substantial insight into the condition of neurals within the family. Relationships between some extinct fossil forms and extant taxa of the continent have been postulated by various authors and in a number of those cases the presence or number of neurals on the carapace surface has been an important factor used in analyses. For that reason, I mention those possible relationships, while noting that Maniel and de la Fuente (2016) have been cautious pending further analyses and have placed the extinct short-necked taxa Bonapartemys. Linderochelys. Lomalatachelys. Paleophrynops. Prochelidella, and Salamanchelys within a broadly encompassing Pan-Chelidae rather than in crown Chelidae. The authors of the recently described Mendozachelys (de la Fuente et al. 2016) and Rionegrochelys caldieroi (de la Fuente et al. 2017) have ascribed these also to Pan-Chelidae.

The earliest known South American chelid, Prochelidella cerrobarcinae (Cretaceous, Aptian-Albian of Patagonia) (de la Fuente et al. 2011), had a contiguous series of 8 neurals extending from contact with the nuchal to contact with the suprapygal. On the dorsal surface, costal pair 8 meets behind neural 8 but on the visceral surface the neural is in contact with the suprapygal (de la Fuente et al. 2011). Prochelidella argentinae (Cretaceous, Turonien-Campanian, possibly even Albian) (Lapparent de Broin and de la Fuente 2001) and Prochelidella portezuelae (Upper Cretaceous, late Turonian-early Coniacian of Patagonia, Argentina) (de la Fuente 2003) are each known from fossil anterior portions of the carapace only. Each of these fossil fragments has 2 or 3 contiguous anterior neurals intact. Lapparent de Broin and de la Fuente (2001) mention the existence of a neural 8 in remains of “small forms belonging to or very close to Prochelidella argentinae”. It thus appears likely that 8 neurals was the uniform condition in Prochelidella. Although Prochelidella had neurals, it has been considered to be part of a group including Acanothochelys, which usually lacks neurals (Lapparent de Broin and de la Fuente 2001; de la Fuente 2003, 2007).

Linderochelys rinconensis (Cretaceous, late Turonian-Coniacian of Neuquén Province, Patagonia, Argentina) is known only from fragments and the presence or extent of neurals in that taxon is unknown (de la Fuente et al. 2007). A series of 8 neurals was present in Bonapartemys bajobarrealis (Cretaceous, Turonien-Campanian of Chubut, Argentina) (Lapparent de Broin and de la Fuente 2001). A contiguous series of 7 neurals is found in Lomalatachelys neuquina, which may be a member of the Chelus group, of the Upper Cretaceous, Santonian, of Argentina (Lapparent de Broin and de la Fuente 2001).

Paleophrynops patagonicus is from the Cretaceous, Upper Campanian–Lower Maastrichtian of Patagonia, Argentina (Lapparent de Broin and de la Fuente 2001). Lapparent de Broin and de la Fuente (2001) considered it to be a member of the Phrynops group. The anterior portion of the carapace of the known specimen is preserved. Neural 1 is in contact with the nuchal and neural 2 is present and contiguous with neural 1, which may indicate that this taxon possessed a full neural series.

Mendozachelys wichmanni is a large, short-necked chelid described from a substantially complete fossil skeleton specimen (de la Fuente et al. 2016) from Mendoza Province, Argentina, dated from the Cretaceous, Upper Campanian–Lower Maastrichtian. It has a series of 5 or 6 neurals with the first in contact with the nuchal and the last separated from the suprapygal by midline contact of costal pairs 7 and 8. Investigation of the relationships of Mendozachelys by de la Fuente et al. (2016) returned different possibilities depending on the use of morphological evidence or a combination of morphological and molecular data in the analysis. They found that Mendozachelys may be a sister of Chelus, or it may have a more basal relationship within the family.

The holotype of Salamanchelys palaeocenica, described by Bona (2006) from the Paleocene, Danian of Argentina, has a series of 5 neurals on the preserved portion of the carapace, with the first in contact with the nuchal. Bona (2006) said that referred fragmentary material indicates that the neural series is interrupted posteriorly by medial contact of the last pair of costal bones. She considered the taxon was probably closely related to the Phrynops group.

Within long-necked chelids, a complete and contiguous series of 8 neurals was present in the earliest known forms, Yaminuechelys gasparinii (late Cretaceous–early Campanian–early Maastrichtian of Rio Negro Province, Argentina) (de la Fuente et al. 2001, 2010; Bona and de la Fuente 2005; current stratigraphy provided by M. de la Fuente, pers. comm., November 2016) and Yaminuechelys maior (Lower Paleocene) (Bona and de la Fuente 2005). Posterior neurals are also present in the preserved rear portion of a carapace of a possible third taxon, Yaminuechelys aff. maior (Lower–Middle Campanian) (de la Fuente et al. 2015), which suggests it too had a complete neural series. Yaminuechelys is considered to be the sister to Hydromedusa (Bona and de la Fuente 2005; de la Fuente et al. 2015). A series of 8 neurals is present in the Lower Eocene Hydromedusa casamayorensis (de la Fuente and Bona 2002; Maniel et al. 2012, 2018). Hydromedusa cf. casamayorensis (Bona 2006) or cf. Hydromedusa Maniel and de la Fuente (2016), from the Danian, Paleocene, is known only from peripheral bones. Neurals number between 6 and 9 in extant Hydromedusa maximiliani and Hydromedusa tectifera (Baur 1893; Wood and Moody 1976). Pritchard (1988) reported for these species that neural 1 is routinely in narrow contact with the nuchal, but the neural series may or may not reach the suprapygal and neural 8 may be isolated from the otherwise contiguous series. Neural 1 is a long, almost triangular-P element and, as noted by Pritchard (1988), the subsequent neurals are generally hexagonal A-shaped, although unlike most other chelids, it is quite usual for a number of them to be quadrangular as their anterior and posterior extremities coincide with the sutures between adjacent pair of costals. A posterior neural, which may be a ninth, is often pentagonal (Wood and Moody 1976).

Two extinct species of Chelus have been described from northern South America, Chelus colombiana from early–late Miocene (Wood 1976; Cadena et al. 2008) and late Miocene Chelus lewisi (Wood 1976; Bocquentin 1988). It is possible that these are synonymous with each other or with extant Chelus fimbriata (Ferreira et al. 2016). The shape of neural 1 has been suggested to be a character distinguishing the 3 species (Wood 1976; Ferreira et al. 2016). All Chelus have a contiguous series of between 6 and 8 (usually 7) neurals, with neural 1 in contact with the nuchal and the last neural usually separating costal pair 7 but not reaching the suprapygal (Fig. 5) (Pritchard and Trebbau 1984; Pritchard 1988). Neural 1 is often a 4P or 6P element and it frequently has a constriction coinciding with the sulcus between the cervical and vertebral 1 scutes. Although the neurals in Chelus are much wider than in any other chelid, the majority of them are roughly 6A-shaped, although variants are frequent. The extent of the neural series within the carapace appears to vary little and individual differences in the numbers of neurals are primarily due to fusion of 2 neurals into a single one, or divison of 1 into 2 (Pritchard 1988). The neurals in Chelus constitute a major portion of a raised central keel above the thoracic vertebrae and rib heads.

View Figure

• Download as Powerpoint
• Download Figure

The late Miocene–Pliocene Parahydraspis paranensis (Wieland 1923) is now considered to be an extinct member of Phrynops (de la Fuente 1992; Kischlat 1993; Maniel and de la Fuente 2016). It appears that the nuchal and neural 1 were in sutural contact although a very small fragment broken from the holotype fossil prevents confirmation of that. The preserved anterior portion of the carapace has a contiguous series of 6 neurals (de la Fuente 1992), most of which are type 6A. In extant Phrynops (Phrynops geoffroanus. Phrynops hilarii. Phrynops tuberosus, and Phrynops williamsi) the neurals form a contiguous series of between 2 and 7, and usually between 5 and 7. Neural 1 is usually in contact with the nuchal but the series does not routinely extend to the posteriormost costals (Rhodin and Mittermeier 1983; Pritchard and Trebbau 1984; Pritchard 1988; McCord et al. 2001). Neural 1 in these 4 species varies between quadrangular, pentagonal, and hexagonal and is often a P-shaped element. The second to penultimate neurals are generally type 6A and the posterior neural is often type 5A or 7A.

The condition of neurals is highly individually variable in the 10 extant species of Mesoclemmys, but there is no nuchal/neural contact and neurals do not exist between the posteriormost costals in any of them (Pritchard and Trebbau 1984; Pritchard 1988; Bour and Zaher 2005). McCord et al. (2001) provide information about the number of neurals in all species except Mesoclemmys hogei while other authors do so for a few of them (Pritchard and Trebbau 1984; Pritchard 1988; de la Fuente 1992; Friol 2014). Published information suggests that many individuals of most Mesoclemmys possess 3 or 4 neurals, but their complete absence is not unusual and occasional individuals with up to 5 are documented for Mesoclemmys gibbus and Mesoclemmys nasuta. McCord et al. (2001) noted that in Mesoclemmys vanderhaegei the neurals are discontiguous. de la Fuente (1992) says that total absence of neurals in Mesoclemmys dahli and Mesoclemmys zuliae has been established. The anteriormost neural in Mesoclemmys is often 5A or 6A and subsequent contiguous neurals are usually 6A, with the posteriormost neural being 5A or 7A. Friol (2014) found that a single neural may be present in some individuals of M. hogei. Of 6 specimens I examined, 4 (UFRJ RRS001, MZUFU 0037, USP 96, and 0120-C) had no neurals, a fifth (MZUFU 0051) had a single isolated type 6 neural (not wider at one end than the other) between the posterior portion of costal pair 2 and a sixth specimen (MZUF 0050-C) has 2 isolated neurals, a type 6 between the anterior portion of costal pair 2 and a small irregular neural between the anterior of costal pair 4 (Fig. 5).

Rhinemys rufipes usually has between 5 and 8 contiguous neurals, with the first always in contact with the nuchal (McCord et al. 2001). The holotype (ZSM 3006/0) has the formula Nu-6P-4-6A-6A-6A-6A-6A-6A\\S. Pritchard (1988) says that Siebenrock (1904) illustrated a specimen with neurals 2 to 6 being type 9A neurals; however, that appears to be a typographic error as Siebenrock's figure clearly shows the formula as Nu-4P6A-6A-6A-6A-6A-7A-*-S. Acanthochelys is composed of the Mid–Late Pliocene Acanthochelys cosquinensis (de la Fuente 1992) and 4 extant species, Acanthochelys macrocephala. Acanthochelys pallidipectoris. A. radiolata, and A. spixii. de la Fuente (1992) notes that a total lack of neurals is a characteristic of the genus, although Friol (2014) says that between 0 and 6 neurals may be found in A. spixii. No neurals were present in 2 specimens of A. pallidipectoris, 3 specimens of A. radiolata, or 5 of A. spixii in my study (Fig. 5). Platemys platycephala, extant across a wide South American range, usually has no neurals, but Pritchard (1988) noted that occasional individuals may have one or more “rudimentary” neurals.

A relatively small amount of fossil material for Australasian chelids has been documented but virtually all of the known forms are clearly referable to extant genera or are undoubtedly within crown-Chelidae (Maniel and de la Fuente 2016).

Among fossil short-necked Australasian forms, Pseudemydura from Riversleigh, Queensland, is known only from a skull fragment (Gaffney et al. 1989). Neurals do not occur in extant Pseudemydura umbrina (Burbidge et al. 1974). Fossil material from the Darling Downs in Queensland of Pliocene or Pleistocene Rheodytes devisi (Thomson 2000a) is fragmentary and insufficient to determine the presence or absence of neurals. Neurals do not occur in extant Rheodytes leukops (Legler and Cann 1980) or Elusor macrurus (Cann and Legler 1994).

Megirian and Murray (1999) provide a comprehensive assessment of Miocene chelids from the Camfield Beds of Australia's Northern Territory. These include the diagnosable Birlimarr gaffneyi and another short-necked form, both of which are considered to be quite closely related to Emydura or the Elseya dentata “generic group,” which comforms with the subgenus Elseya (Elseya) as delimited by more recent diagnoses (Thomson and Georges 2009; Le et al. 2013; Thomson et al. 2015). Megirian and Murray (1999) also discuss and figure a form from the Miocene Carl Creek Limestone of Riversleigh, Queensland, first detailed by White (1997), and which they designated as an indeterminate species of Birlimarr. From the virtually entire skeleton of Birlimarr gaffneyi and the substantially complete carapace of the Carl Creek Birlimarr, it is apparent that the genus lacked neurals. However, Megirian and Murray (1999) describe the disarticulated and/or fragmentary nature of the majority of the other Camfield Beds material from multiple individual short-necked chelids. Evidence for the presence or otherwise of neurals in these taxa is inconclusive.

Gaffney (1979) specifically noted the lack of neurals in his description of 2 complete shells of Emydura sp. from the Miocene Wipariji Formation of South Australia. Warren (1969) reported on fossils of “sp. aff. E. macquari” dating from the Oligocene or Miocene found at Taroona, Tasmania. One specimen has a single 4K neural at the suture of costal pairs 3 and 4, while another from the same deposit has none. He suggested that the single neural “may represent a normal variation as yet undetected in populations of Australian turtles or it may represent a remnant of an ancestral stage prior to the loss of neurals. Its absence in one of the Taroona specimens (UTG 59374) and its presence in another indicate that as early as mid-Tertiary times it was not a stable element”.

Neurals are not routinely present in any living taxon of Emydura. Scheyer et al. (2008) found no indication of any neurals in histological investigation of embryonic and hatchling stages of Emydura s. subglobosa and Scheyer (2009) reports that extant Emydura spp. with carapace lengths up to 300 mm (i.e., mature adults), “lack neurals so the costals meet medially above the vertebral column”. Thomson and Georges (1996) found no neurals on the skeletal carapace surface of 2 specimens of Emydura macquarii nigra or in 5 specimens of Emydura subglobosa worrelli. Rhodin and Mittermeier (1977) reported a “second small suprapygal between the 8 costals” in 5 of 14 specimens of Emydura subglobosa from southern New Guinea, but their designation clearly shows they did not consider these to be neurals. Neurals were not present in specimens of any Emydura I examined.

Three clades within extant Elseya have recently been recognized as subgenera (Thomson et al. 2015). The subgenus Elseya contains E. dentata, E. branderhorsti, and Elseya flaviventralis. New Guinean Elseya novaeguineae, Elseya schultzei, and Elseya rhodini belong to the subgenus Hanwarachelys. Extant species of northern and eastern Queensland, Elseya albagula. Elseya irwini, and Elseya lavarackorum, and the fossil taxa Elseya uberrima and Elseya nadibajagu are assigned to the subgenus Pelocomastes. While Thomson et al. (2015) report that absence of neurals is common to the entire genus and Thomson and Georges (1996) found no neurals in 8 specimens of E. dentata, it is apparent that rare isolated neurals occur in some species within at least 2 subgenera because those authors reported a single neural from 1 of 6 specimens of E. flaviventralis (as “Elseya sp. aff. dentata [South Alligator River, Northern Territory]”) and also from 1 of 2 specimens of E. schultzei. No neurals were present in a single mature specimen of E. irwini (Fig. 5), nor were they present in a radiograph of E. branderhorsti that I examined.

Fossil material from the Darling Downs in Queensland of Pliocene or Pleistocene Elseya uberima (Thomson 2000a) is fragmentary and insufficient to determine the presence or absence of neurals in that species. Pliocene Elseya nadibajagu (Thomson and Mackness 1999) from Bluff Downs in Queensland is represented by fragments, but they appear to be sufficient to indicate the lack of a contiguous series of neurals. The preserved portion of the carapace of the Pleistocene fossil of Elseya lavarackoram from Riversleigh in Queensland (White and Archer 1994; Thomson et al. 1997) has dermal scutes and the condition of neurals is not visible. Neurals are not present in the subsequently discovered extant population of that species.

Rarely, up to 4 small neurals have been reported from Myuchelys latisternum (Rhodin and Mittermeier 1977). Thomson and Georges (1996) found none in 6 specimens of that species or in 4 specimens of Myuchelys georgesi. A single specimen of Myuchelys bellii I examined had no neurals.

Myuchelys purvisi is the only Australasian short-necked turtle that consistently possesses a series of between 3 and 5 neurals (Thomson and Georges 1996). Neural 1 is absent, as is neural 2 in some specimens, and there is no nuchal/neural contact. The anteriormost neural is pentagonal and the subsequent few neurals are type 5A or 6A. The posteriormost neural in 4 specimens was type 5A.

Rhodin and Mittermeier (1977) reported an individual of Elseya (AMNH R76199) of unknown origin with 3 neurals. At that time Myuchelys had not been distinguished from Elseya. I examined the specimen at the AMNH. It has the neural formula N**4A-5A\\4K***S. D. Dickey (Dept. of Herpetology, AMNH, pers. comm., September 2015) provided information that the specimen originated from a dealer in Sydney, Australia, and was in the live collection of the New York Zoological Society prior to its deposition in the AMNH collection. The overall carapace size and shape; the presence of 3 neurals, 2 of which are contiguous; a cervical scute; the morphology of the axillary bridge sutures with costals; the position of the ileum suture with costals; and the shape of parietals demonstrate the specimen is Myuchelys purvisi.

The fossil record for the Australasian long-necked Chelodina is also relatively poor. However, the holotype of the Early Cenozoic (at least Eocene and possibly Palaeocene) Chelodina alanrixi had a contiguous series of 8 neurals (Lapparent de Broin and Molnar 2001). The preserved portion of the carapace includes the posterior portion of neural 1, neurals 3 to 5, the posterior portion of neural 7, and the whole neural 8. The extant neurals are contiguous. Neurals 3 to 5 are type 6A and neural 8 is type 5. Costal pair 8 has a midline suture between neural 8 and the suprapygal. Georges and Thomson (2010) recognize 3 subgenera: Chelodina. Macrodiremys, and Macrochelodina. Lapparent de Broin and Molnar (2001) suggest that C. alanrixi, with its broad carapace, shows some similarities to C. expansa and it appears to be most similar to extant taxa of Chelodina (Macrochelodina). Other fossil taxa from the same deposits are not identifiable beyond recognition that they are Chelodina and at least one has some carapace features similar to Chelodina (Chelodina. steindachneri (Lapparent de Broin and Molnar 2001). Those authors consider a partial plaston of the Pliocene from north Queensland, and Miocene to Pleistocene carapace fragments from “Darling Downs: Eight-mile Plains, near Brisbane; Warburton River” that were referred to Chelodina insculpta by De Vis (1897), to be similar to C. longicollis in a number of aspects. Gaffney (1981) considered that among these specimens, only a partial plastron was attributable to C. insculpta. Thomson (2000b) provided evidence for recognition of C. insculpta and he determined that the fossils are from the Darling Downs alone. He also noted some characters that indicate the taxon may have been more closely allied to extant Chelodina (Macrochelodina. expansa than to species of Chelodina (Chelodina). Nonetheless the carapace fragments are insufficient to indicate whether or not neurals were present. Chelodina from the Miocene Camfield Beds of Northern Territory designated “sp. A” and “sp. B” by Megirian and Murray (1999) are represented by plastal bones only. Gaffney et al. (1989) discussed a Miocene specimen fragment of Chelodina from Riversleigh in Queensland that includes “two medial fragments of costals demonstrating the absence of neurals”. Given the potential for neural series of variable numbers (see below), it would seem prudent to consider that this fragmentary specimen demonstrates the lack of a full series of neurals rather than a complete lack of any neurals. In describing the Miocene Chelodina (Chelodina. murrayi, Yates (2013) noted that while its remains are fragmentary, there is sufficient evidence to determine that it either lacked neurals or, at most, there could have been 1 or 2 discontinuous neurals.

While neurals are most usually absent from extant taxa within the subgenus Chelodina, small isolated neurals have been previously reported from occasional mature individuals of C. longicollis and Chelodina novaeguineae (Rhodin and Mittermeier 1977; Thomson and Georges 1996). I found between 1 and 4 small neurals in 7 of a series of 31 C. longicollis shells (Fig. 5) and 1 small neural in 1 of 13 shells of C. canni. These were almost all small and isolated and, in each case, were located at the midline juncture of 2 pairs of costals. Most frequently they were irregular, but roughly type K. No neurals were present in 7 specimens of C. steindachneri or in small samples of Chelodina mccordi (Rhodin 1994a) and Chelodina pritchardi (Rhodin 1994b). No information about neurals is included in descriptions of C. reimanni (Philippen and Grossman 1990), Chelodina rankini (Wells 2007), C. gunaleni (McCord and Joseph-Ouni 2007a), C. mccordi timorensis (McCord et al. 2007a; Kuchling et al. 2007), or C. m. roteensis (McCord et al. 2007b). No neurals were observed in radiographs I examined of single specimens of C. gunaleni and C. reimanni.

Chelodina (Macrodiremys. colliei consistently has a series of between 5 and 8 neurals (Burbidge 1967; Burbidge et al. 1974) (Fig. 5). There is no nuchal/neural contact in this species, either because neural 1 is absent or because it is relatively small and does not extend forward sufficiently to reach the nuchal. The series is routinely contiguous from the anteriormost neural situated between the posterior of costal pair 1, back to at least costal pair 5. Posterior to that, neurals are sometimes missing or discontiguous. In 9 of 19 specimens examined, the short neural 1 was type 5 (usually 5A) and in most of the remaining specimens it was type 4. Most of the neurals in the subsequent contiguous series were type 6A or 4A. However, many of them are modified from the 6A “coffin” shape (Pritchard 1988) common to many other taxa. The neurals are narrow and elongated, as is the entire carapace, and their wide anterior portion is often foreshortened and its anterior suture is concave, corresponding with a convex posterior edge of the preceding neural. The length of these neurals often corresponds closely with the width of the pair of costals between which they are situated.

Among species of the subgenus Macrochelodina, neurals have never been found in C. expansa. Chelodina parkeri (Rhodin and Mittermeier 1976; Pritchard 1988), or C. walloyarrina (McCord and Joseph-Ouni 2007b). Rhodin and Mittermeier (1977) reported 2 neurals in 1 of a combined total of 12 specimens of C. oblonga (as Chelodina rugosa and Chelodina seibenrocki). They also examined the type specimen of C. oblonga (BM 40.12.9.81) radiographically and determined that it has at least “two large contiguous neurals between the second and third costals”.

In their description of C. (Macrochelodina. burrungandjii, Thomson et al. (2000) stated it has a contiguous row of between 3 and 5 neurals. I examined 2 paratype specimens (MAGNT R16008, Koolpin Gorge, N.T., and R22582, Sleisbeck, N.T.) using CT scanning and, respectively, they had 4 and 6 neurals. However, the presence of a neural series in this taxon may vary geographically because 3 specimens from the Maningrida and Mann River area did not have series of neurals. Two of them (MAGNT R13525, R35010) each had 2 very small isolated neurals, while the third (R36032) had no neurals. As pointed out by S. Thomson (pers. comm., May 2017), there is significant evidence of hybridization between C. burrungandjii and C. oblonga in various localities where they co-occur and variability in neurals may be influenced accordingly.

Chelodina kuchlingi was described by Cann (1997) on the basis of a single specimen. Kuchling (2014) identified 2 additional specimens in the WAM collection and provided evidence for validity of the taxon and for its likely provenance from the Ord River region of Western Australia. I located 2 further specimens from the same region in the MAGNT collection (R34788, Cockburn Creek, W.A., and R34791, Kununurra, W.A.) that are morphologically referrable to C. kuchlingi; however, it is clear that further investigation of this poorly known taxon is required. The Cockburn Creek specimen was examined using CT scanning and had no neurals. The shell morphology of 2 larger specimens from Parry Creek in the same region (MAGNT R34786 and R34789) is somewhat different but they may be the same taxon. Of these, R34786 has 1 small neural close to the suture between costal pairs 3 and 4, while R34789 has no neurals.

Results of the review presented here can be summarized as follows:

1. Early chelids, mainly from South America, where they date back as far as the Cretaceous, but spanning the period forward to the Eocene in both South America and Australasia, are consistent in their possession of a series of neurals. While not all fossils of these early taxa are entire, all those that are have a series of at least 5 surface neurals, and in a number of taxa there is a complete series of 8 surface neurals. All Eocene and pre-Eocene fossils that are fragmentary have contiguous anterior surface neurals and are thus consistent with their having had a longer contiguous series of neurals. More recent fossils and extant forms are variable for the condition. The review here thus lends weight to the concept, discussed in some detail by Lapparent de Broin and Molnar (2001) that a complete series of neurals is plesiomorphic in chelids and the reduction of neurals from a complete series to partial or complete loss has occurred in various lineages of South American and Australasian chelids. Among long-necked forms, consistent reduction and complete loss has occurred only in parts of the Australasian lineage.

2. Some level of reduction in the series of neurals is almost ubiquitous across extant chelids, such that a continuous series extending from the nuchal to the suprapygal is exceptional.

3. A contiguous series of neurals is consistently present in extant South American Chelus. Hydromedusa. Phrynops, and Rhinemys and Australian Chelodina (Macrodiremys). In some of these taxa neurals appear to be close to the plesiomorphic condition both in their number and their shapes on the dorsal carapace surface.

4. Neurals are consistently absent from extant South American Platemys and Australasian Emydura. Elusor. Rheodytes, and Pseudemydura.

5. Presence of a series of neurals differs consistently between species within Australasian Myuchelys and between subgenera of Australasian Chelodina.

6. Among South American Acanthochelys and Mesoclemmys, the number and distribution of neurals is variable between species but neurals are routinely absent from A. pallidipectoris and A. macrocephala.

7. Phrynops. Rhinemys. Myuchelys purvisi. Chelodina (Macrodiremys), and Chelodina (Macrochelodina. burrungandjii from its type locality, all routinely possess a series of neurals, but in all of these, the continuity of the neural series and the number of neurals are individually variable.

8. Small, isolated neurals occur in occasional individuals of some species of Myuchelys. Elseya, and Chelodina in which a contiguous series of neurals is routinely absent.

A continuous series of neurals is the ancestral condition of the Chelidae (Zangerl 1948 [in Rhodin and Mittermeier 1977]); Gaffney 1977; Pritchard 1988; Thomson and Georges 1996; Lapparent de Broin and Molnar 2001). Reduction in the number of neurals, and of their contribution to the carapace surface is now widespread across the chelid radiation. This reduction shows a general progressive tendancy in which 1) neurals no longer reach the suprapygal, 2) neurals no longer contact the nuchal, 3) the number of neurals reduces, and ultimately, 4) neurals are completely absent. In this progression an increasing number of the costal pairs are no longer separated by neurals and meet each other on the vertebral midline.

DISCUSSION

Anatomical Investigation of Neurals. — Neurals are dorsal or dorso-lateral processes of the carapacial neural arches but the anatomical structure of these bones in chelids that possess, or do not possess, neurals has rarely been clearly described. Neurals have frequently been discussed, or explicity described as distinct bones, even in the literature of turtle osteology (e.g., Zangerl 1969; Rhodin and Mittermeier 1977; Pritchard 1988; Thomson and Georges 1996).

Scheyer et al. (2008) investigated the embryonic development of the vertebral carapace in pleurodires by comparing that of a chelid (Emydura subglobosa) in which neurals are not known, with that of 2 pelomedusoides in which neurals occur. Their histological series demonstrated how neurals form from the embryonic periosteal collars of neural arches in taxa that have neurals, but that there was no such development and consequently no neurals in Emydura subglobosa. My results from examination of ossified shells for a sample of 13 Australian chelid species, encompassing 4 genera, agree with theirs in demonstrating that neurals are outgrowth processes of vertebral neural arches and are not separate bones. This is in agreement with the anatomy of nonchelid turtles (see, e.g., the cryptodire Pseudemys concinna; Fig. 4C), as also noted by Cherepanov (2016).

For the taxa I examined, my results show that where neurals are absent, the neural arch is simply sutured across the ventral midline of the visceral surface of costal bones. Scheyer (2007) noted the same for a specimen of Platemys platycephala that he sectioned and which is shown here by kind permission (Fig. 3D).

Thomson and Georges (1996) described their discovery of small and “isolated neurals” as distinct bones lying dorsal to the neural arches of the vertebral column and embedded under the surface of the medial costals' suture and thus not exposed on the skeletal carapace surface. They reported these from Elseya dentata. Emydura subglobosa worrelli, and Chelodina longicollis. Their investigation included sectioning and examination of 5 species from 3 Australian genera. Subsequently they, and Georges and Thomson (2006) and Thomson (2015), have considered that the presence of subsurface neurals is a derived condition common to all chelids in which neurals are not exposed on the skeletal surface of the carapace. In combination, the anatomical work of Scheyer (2007), Scheyer et al. (2008), and my investigation include 1 South American species and 13 Australian species encompassing 5 chelid genera. Our combined results are at odds with those of Thomson and Georges in that we found that “neurals” are processes of neural arch bones and are not separate bones and we found no subsurface neurals. At the least, these results confirm that subsurface neurals are not ubiquitous in chelids that lack neurals on the carapace surface.

Neurals and Evolution of the Chelidae. — Phylogenetic reconstructions should be created on the basis of the maximum available evidence and the condition of neurals is just one character. However, various previous authors have given substantial weight to the condition of neurals in consideration of chelid relationships (Burbidge et al. 1974; Gaffney 1977, 1979; McDowell 1983; Manning and Kofron 1996; Thomson and Georges 1996, 2009) and a number of these have considered the evolutionary significance of the presence or absence of neurals in particular chelid taxa and/or in phylogenetic reconstructions for the family. However, a concensus has yet to be reached about a fundamental aspect of chelid phylogeny and a brief background to this is as follows.

Extremely long necks, approaching or equalling the carapace length, are a feature of 3 extant chelid genera (Chelus and Hydromedusa of South America and Chelodina of Australasia) and of Cretaceous Yaminuechelys of South America. Primarily on the basis of cranial morphology and prior to descriptions of Yaminuechelys, Gaffney (1977) developed a hypothesis in which the 3 extant long-necked genera and Phrynops form a clade. However, Pritchard (1984) considered that Chelodina was not closely related to the 2 extant South American long-necked genera and that both its long neck and similarity of its skull morphology, particularly to Hydromedusa, are the results of convergent evolution. In essence, the question is whether long-necked chelids evolved once and all known long-necked forms belong to a single clade, or whether long-necked forms evolved twice with 2 resultant clades represented by known South American forms and Australasian Chelodina. In the latter case, long-necked chelids inhabiting the 2 continents may each share more recent common ancestry with the short-necked forms of their respective continents than with long-necked taxa of the other landmass.

Subsequent considerations of this question based on morphology of extant taxa (Gaffney 1977; Gaffney and Meylan 1988) and fossil forms (Bona and de la Fuente 2005; de la Fuente and Sterli 2005) suggest monophyly of long-neck chelids and that their common ancestor lived before the rupture of the southern Gondwanan landmass. By contrast, investigation of chelid relationships based on molecular analyses suggest that the long-necked chelid genera are a polyphyletic assemblage (Seddon et al. 1997; Georges et al. 1998; Fujita et al. 2004) and that evolution of long necks occurred separately in South American and Australasian forms, possibly subsequent to the fragmentation of Gondwana that sequentially separated South America, Antarctica, and Australia (de la Fuente and Sterli 2005). Respectively, these 2 hypotheses have come to be termed the “morphological” and the “molecular” hypotheses.

A number of studies have specifically addressed this question from different angles (Shaffer et al. 1997; de la Fuente and Sterli 2005; Scheyer 2009; Wilson and Sánchez-Villagra 2011; Holley et al. 2015; Maniel et al. 2018) but the evidence to date has not resulted in concensus for either hypothesis. There are nuances to the morphological hypothesis. Investigations since Gaffney (1977) have excluded Phrynops from the proposed long-necked monophyly and a cladistic analysis of total evidence by de la Fuente and Sterli (2005) indicated that Chelus is not part of this assemblage (in which case the long-necked condition must have arisen twice if the remaining long-necked chelids are monophyletic, or 3 times if they are not). Holley et al. (2015) examined the question using molecular clock analyses and fossils as calibration points in different combinations for each hypothesis. Their analyses indicate an early diversification of the chelids and the evolution of long-necked chelids before the final breakup of southern Gondwana. Although they offer evidence suggesting that Chelodina is younger than South American long-necks, they were not able to determine whether long-necked forms evolved more than once. Thus, at present, phylogenies for both the morphological and the molecular hypotheses are tenable.

The foundation of molecular phylogeny for the family was provided by Georges and Adams (1992), Seddon et al. (1997), and Georges et al. (1998). A number of studies before and since have provided molecular evidence for relationships at the genus level or lower (Derr et al. 1987; Georges and Adams 1996; Georges et al. 2002, 2014; Reid et al. 2011; Fielder et al. 2012; Huebinger et al. 2013; Le et al. 2013; Friol et al. 2015; Spinks et al. 2015; Thomson et al. 2015). Phylogenies derived from both hypotheses for chelid genera are provided by Scheyer (2009) and de la Fuente et al. (2014). It is worth noting that the 2 hypotheses relate to early divergences in the evolution of the family. Relationships within major clades (respectively, South American and Australasian short-necked genera, South American long-necked genera, and Australasian long-necked subgenera) and terminal taxa within these, are not controversial.

In order to explore the retention and loss of neurals in the evolution of chelids, I have redrawn the basic phylogenies for each of the 2 hypotheses from de la Fuente et al. (2014) and added all extant genera and relevant species (Fig. 6). Extinct forms are included where there is information about their neurals and a putative relationship to an extant form has been suggested by a previous author (or authors). Their inclusion provides some indication of the timescale in which variation in neurals may have occurred. The timescales of possible divergences are based, where available, on published sources but should be treated as indicative only. On both phylogenies I have mapped retention, partial loss, and complete absence of a contiguous series of neurals on the premise that possession of a contiguous series of 8 neurals is the plesiomorphic condition for the family and that parsimony thus dictates that possession of a series of neurals is a retained primitive condition.

View Figure

• Download as Powerpoint
• Download Figure

While some degree of reduction of the neural series is almost ubiquitous in extant chelids, I have mapped the character according to 3 condition states as follows, which fundamentally accord with those recognized by de la Fuente et al. (2016):

1. Neurals are “retained”: where the majority of the plesiomorphic 8 neurals are consistently present with most of them in a contiguous series.

2. Neurals are “partially lost”: where some neurals are consistently present but they usually number fewer than 4 and/or continuity of the series is irregular or inconsistent.

3. Neurals are “absent”: where neurals are routinely absent from the majority of specimens.

Mapping the 3 neural condition states on each of a molecular and a morphological phylogeny indicates that complete loss of neurals has occurred at least 8 times and partial loss has occurred 3 times in both phylogenetic reconstructions. The 2 trees do not differ in this respect because the presence, partial presence, and absence of a series of neurals differs between genera, subgenera, and species within clades that are clearly closely related rather than more deeply within the evolution of the family. In this respect neurals do not appear to provide evidence in support of either phylogenetic hypothesis.

Previous attempts to consider the condition of neurals in phylogenies of the Chelidae (Gaffney 1977; Thomson and Georges 1996) have not had the benefit of more recently available information and, at least in part, have been confounded by polarizing the character state as one in which a series of neurals is either “present” or “absent”. The review presented here demonstrates that the condition is considerably more subtle and complex, with variable degrees of retention and reduction in the presence of neurals across the chelid radiation.

The information reviewed here suggests support for Gaffney's (1977) view that loss of neurals has occurred independently in disparate chelid lineages. We can add that the process of their reduction appears to be in progress as a widespread phenomenon in the family. The retention of a long series of neurals indicates plesiomorphy in that character but its retention or loss is otherwise of limited value to our understanding of chelid phylogeny or taxonomy.

While the presence of a series of type 6A neurals is accepted as the plesiomorphic condition of chelids (Pritchard 1988), Thomson and Georges (1996, 2009) and Thomson (2003) consider that the series of neurals in C. colliei and C. burrungandjii is a secondarily derived condition. Their rationale for this is that a contiguous series of neurals in Chelodina is associated with expansion of the rib heads to accommodate enlarged musculature necessary for rapid extension of the long necks of these 2 species for capture of motile prey, albeit that they achieve this function by the use of different muscles (Thomson 2003). Georges and Adams (1992), Seddon et al. (1997), and Georges et al. (1998) have offered evidence that C. colliei is the sister of C. (Chelodina) and Thomson (2003) determined that, “based on morphology, the species Chelodina colliei is in fact a highly derived member of the Chelodina group which converged on the Macrochelodina condition”. Because C. (Chelodina) does not routinely have neurals, the existence of a long series of neurals in C. colliei does require explanation. Two possibilities are 1) that neurals have reemerged as a derived condition or 2) that it has retained neurals as an ancestral condition.

For a neural series to be derived as an element of the skeletal carapace surface after this condition has been lost in an ancestor, it may not be necessary for them to have evolved as an entirely new condition. Because neurals are extensions of the underlying neural arches, it is plausible to conceive a mechanism in which the overlying costal bones become thinner and ultimately the neurals could emerge through and between them. By comparison with all other Chelodina, and virtually all other chelids, the costals of C. colliei are quite thin.

Burbidge et al. (1974) and Pritchard (1988) proposed that the neural series in C. colliei is ancestral and likely to have been retained in long isolation from congeners that subsequently lost a neural series. The current distribution of C. colliei is substantially within a southern portion of the Yilgarn Craton of Western Australia. It is notable that the Craton has an ancient history of high stability without major perturbations such as vulcanism, marine transgression, or significant orogenesis since long before the earliest known chelids. Whether the ancestor of C. colliei inhabited the region prior to it rifting from Antractica in the early–mid Cretaceous (Veevers 2000) is moot, although that antiquity for it was suggested by Burbidge et al. (1974). Regardless, this portion of Western Australia has since been substantially isolated from the rest of Australia by periods of north–south marine transgressions dividing the continent and subsequent aridity.

In the case of C. burrungandjii, in which at least a portion of its population consistently possesses a number of neurals, there does not seem to be any compelling case for the view that this is other than partial retention of the ancestral condition. Its costals are at least as thick as those of other species of C. (Macrochelodina) that entirely lack neurals. The topography of the Arnhem Land Plateau of Australia's Northern Territory, to which C. burrungandjii is substantially confined, is considered to have remained essentially unchanged since the Late Jurrasic (Nott 1995), and the taxon may also have been geographically isolated from congeners for much of its history. However, while I found variable series of neurals were present in 2 specimens from South Alligator River drainage at or near the type locality, 3 from the Mann River and Maningrida area had, respectively, no neurals and 2 small isolated neurals. Further investigation of neurals in this species is required, particularly because the presence or absence of a neural series has been thought to be a character distinguishing C. burrungandjii from C. walloyarrina (McCord and Joseph-Ouni 2007b).

On the basis of parsimony and the following evidence, I agree with earlier authors (Burbidge et al. 1974; Manning and Kofron 1996) and consider that the series of neurals in C. colliei and in C. burrungandjii is more likely to be a retained ancestral condition than a derived state.

• A series of 8 contiguous neurals was present in some early Chelodina, as evidenced by the Early Cenozoic Chelodina alanrixi, and this suggests the condition is plesiomorphic for the genus.

• In C. colliei and some C. burrungandjii the configuration of the neural series, in which it is contiguous but is isolated from both the nuchal and suprapygal and a variable number of anterior and posterior costals, is consistent with the pattern in other chelids that have retained a series of neurals as an ancestral condition.

• Although somewhat modified, the basic shape of neurals 2–5 in C. colliei is consistent with the ancestral condition in which the bones are shaped 6A (Pritchard 1988). Neurals of this shape are also present in Early Cenozoic Chelodina alanrixi.

• Other than Chelus. Hydromedusa, and Rhinemys, in which the neurals generally retain contact with the nuchal and may do so with the suprapygal, the arrangement of neurals in C. colliei conforms more closely to the ancestral condition (Pritchard 1988) than it does in other extant taxa across the entire chelid radiation.

Pritchard (1988) offered hypotheses for the loss or reduction of neurals in chelids related to shell function and movement mechanics and Scheyer et al. (2008) have provided working hypotheses for the preclusion of embryonic development of neurals. These hypotheses remain areas of significant interest for further investigation.