Almost all known sauropod necks are incomplete and distorted

Catalogue of complete necks

Unambiguously complete necks are known from published accounts of only a few sauropod specimens. In chronological order of description, the following specimens were found with their necks complete and articulated, and have been adequately described—each of them known to include the posteriormost cervicals because the vertebral column is articulated through into the dorsal region:

• CM 11338, a referred specimen of Camarasaurus lentus described by Gilmore (1925). This is a juvenile specimen, and thus does not fully represent the adult morphology. (McIntosh et al., 1996: 76 claim that this specimen is the holotype, but this is not correct: YPM 1910 is the holotype—see below.)

• CM 3018, the holotype of Apatosaurus louisae, described by Gilmore (1936). The neck was separated from the torso but articulated from C1–C15, though the last three cervicals were badly crushed: see below for details.

• CCG V 20401, the Mamenchisaurus hochuanensis holotype, described by Young & Zhao (1972). Each vertebra is broken in half at mid-length, with the posterior part of each adhering to the anterior part of its successor; and all the vertebrae are badly crushed in an oblique plane.

• ZDM T5402, a Shunosaurus lii referred specimen, described in Chinese by Zhang (1988), with English figure captions. Their figure 22 depicts the atlas. Unlike the holotype T5401, this specimen is mature.

• BYU 9047, the Cathetosaurus lewisi holotype, described by Jensen (1988). (Jensen incorrectly gives the specimen number as BYU 974.) This specimen was redescribed, and the species referred to Camarasaurus, by McIntosh et al. (1996). Although all 12 cervicals are present, “10–12, particularly 12, have suffered such severe damage that it is impossible to restore them” (McIntosh et al., 1996: 76).

• MACN-N 15, the holotype of Amargasaurus cazaui, described by Salgado & Bonaparte (1991) who described “22 presacral vertebrae articulated with each other and attached to the skull and sacrum, relatively complete” (Salgado & Bonaparte, 1991: 335, translated).

• ZDM 0083, the holotype of Mamenchisaurus youngi, described in Chinese by Ouyang & Ye (2002) with English figure captions. Their figure 14 depicts the atlas and axis.

• MUCPv-323, the holotype of Futalognkosaurus dukei, initially described by Calvo et al. (2007a) and redescribed by Calvo et al. (2007b). The neck was found in two articulated sections which fit together without needing additional vertebrae in between (Jorge O. Calvo, 2021, personal communication).

• SSV12001, the holotype of Xinjiangtitan shanshanesis, described by Zhang et al. (2018). The original description of this specimen by Wu et al. (2013) included only the last two cervicals, which were the only ones that had been excavated at that time.

A few additional specimens are known to have complete and articulated necks, but have not yet been described:

• USNM 13786, a referred subadult specimen of Camarasaurus lentus recently mounted at the Smithsonian. The specimen “was almost completely buried before the sinews had allowed the bones to separate” (letter from Earl Douglass to William J. Holland, 22 August 1918), and photographs kindly supplied by Andrew Moore show that the atlas was preserved.

• MNBH TIG3, the holotype of Jobaria tiguidensis. Sereno et al. (1999: 1343) assert that this species has 12 cervicals in all and say “One articulated neck was preserved in a fully dorsiflexed, C-shaped posture”. Paul C. Sereno (2021) personal communication, confirms that the articulated neck is MNBH TIG3.

• SMA 002, referred to Camarasaurus sp. Tschopp et al. (2016), in a description of its feet, say that this specimen “lacks only the vomers, the splenial bones, the distal end of the tail, and one terminal phalanx of the right pes. The bones are preserved in three dimensions and in almost perfect articulation”.

• MAU-Pv-LI-595, the “La Invernada” Titanosaur. Filippi et al. (2016) give a very brief account in an abstract. Filippi (2021) personal communication, says that the entire preserved specimen was articulated.

• MAU-Pv-AC-01, an unnamed titanosaur mentioned in abstracts by Calvo, Coria & Salgado (1997) and Coria & Salgado (1999). The specimen was found in perfect articulation from skull down to the last caudal vertebrae (Rodolfo A. Coria, 2021, personal communication).

The first cervical (the atlas) in sauropods is very different in form from the other vertebrae, and small and fragile. Consequently it is easily lost. Some further specimens have necks that are complete and articulated from C2 (the axis) backwards:

• MB.R.4886, the holotype of Dicraeosaurus hansemanni, described by Janensch (1929), has a neck that is complete and well preserved from C2 to C12 (the last cervical). Janensch referred to this as “specimen m” and writes “It was found articulated from the 19th caudal vertebra to the 9th cervical vertebra inclusive. The proximal part of the neck from the 8th cervical vertebra up to the axis was bent ventrally and lay at right angles to the distal part of the neck.” (Janensch, 1929: 41).

• PMU 233, the holotype of Euhelopus zdanskyi, described by Wiman (1929) as “exemplar a” and redescribed by Wilson & Upchurch (2009).

• ZDM T5401, the subadult holotype of Shunosaurus lii, described in Chinese by Zhang, Yang & Peng (1984). The quarry map (Zhang, Yang & Peng, 1984: figure 1) suggests that the atlas is missing.

• MCT 1487-R, informally known as “DGM Series A”, described by Powell (2003). Gomani (2005: 9) summarises as “12 cervical vertebrae, except the atlas, preserved in articulation with three proximal dorsal vertebrae”.

• GCP-CV-4229, the holotype of Spinophorosaurus nigerensis, described by Remes et al. (2009). This species is known from two specimens, of which the holotype was found in very good condition and well articulated from C2 to C13, the last cervical (Fig. 1). The atlas seems to be missing (Kristian Remes, 2021, personal communication; Ralf Kosma, 2021, personal communication).

One other sauropod is complete from the first cervical, but probably not to the last:

• MOZ-Pv1232, the holotype of Lavocatisaurus agrioensis, described by Canudo et al. (2018). This is complete from C1–C11. Canudo’s guess is that this is the complete neck (Jose I. Canudo, 2021, personal communication), but the specimen doesn’t demand that conclusion and no known eusauropod is definitively known to have fewer than 12 cervicals. However, Upchurch (2021) personal communication, notes that the dicraeosaurid neosauropod Amargasaurus has 22 presacral vertebrae in total, and difficulties in locating the cervicodorsal junction (see below) make it impossible to state confidently whether it had 12 or only 11 cervicals, so caution is warranted here.

Other sauropod specimens have necks that are complete and articulated from further back in the cervical sequence:

• YPM 1910, the holotype of Camarasaurus lentus, a mounted specimen described by Lull (1930). The neck is complete from C2 or C3, Lull was uncertain which. (It should be possible to clear that up by re-examining the specimen in light of what has subsequently become known about Camarasaurus lentus, but this work has not yet been done.)

• SMA 0004, the holotype of Kaatedocus siberi, described by Tschopp & Mateus (2012). Cervicals 3–14 are preserved.

• AODF 888 (informally “Judy”), probably referrable to Diamantinasaurus, briefly described by Poropat et al. (2019). Preserved from C3 or maybe C4. “One posterior cervical (XIII or XIV) found several metres from articulated series, but appears to slot nicely into the gap between the articulated cervical series and the unprepared thoracic section, which might include at least one additional cervical (XIV or XV)” (S. F. Poropat, 2021, personal communication).

Several necks are probably nearly complete, but it is not possible to know due to their not being found in articulation:

• CM 84, the holotype of Diplodocus carnegii, described by Hatcher (1901). C2–C15 are preserved, though not all in articulation; C11 may be an intrusion: see below for details.

• ZDM T5701, the holotype of Omeisaurus tianfuensis, described by He, Li & Cai (1988). The neck was not articulated (He, Li & Cai, 1988: figure 1), and was missing “two elements or so” (He, Li & Cai, 1988: 120).

• QJGPM 1001, the holotype of Qijianglong guokr, described by Xing et al. (2015). On page 8, the authors say “The axis to the 11th cervical vertebra were fully articulated in the quarry. The atlas intercentrum and the 12th–17th cervical vertebrae were closely associated with the series.”

• MNBH TIG9, a referred specimen of Jobaria tiguidensis. Wilson (2012: 103) writes that this specimen “includes a partially articulated series of 19 vertebrae starting from the axis and extending through the mid-dorsal vertebrae.”

• MNBH TIG6, another referred specimen of Jobaria tiguidensis, which has not been mentioned in the literature. Paul C. Sereno (2021) personal communication, says that it is “a subadult partial skeleton with excellent neck” and that “the sequence was articulated from C2–C11. Most of the ribs were attached as well.”

Note. The Jobaria tiguidensis individuals previously had specimen numbers beginning MNN, but the Musée National du Niger changed its name to Musée National Boubou Hama and the specimen numbers have changed with it.

The breakdown of these complete and near-complete necks is interesting (Fig. 2). Non-neosauropods are relatively well represented, both inside and outside of Mamenchisauridae—although it is unfortunate many of these specimens are not well described in English: two of the ten are of Jobaria, for which the cursory summary of Sereno et al. (1999) remains the only published description, and some of the Chinese sauropods are described only in Chinese.

Diplodocoids are surprisingly poorly represented, with only a single specimen in each of Dicraeosauridae and Diplodocidae that is complete. Brachiosaurids have absolutely no representation—see below on how unconvincing the neck of Giraffatitan is. More advanced titanosauriforms are better represented, but there is still only one with a complete neck, Futalognkosaurus dukei. By contrast, the single genus Camarasaurus is very well represented, with five specimens of which four are fully complete (though only two of those have been described). Probably this does not indicate a taxon-specific taphonomic signal, but follows from the sheer abundance of Camarasaurus specimens—an abundance likely influenced by over-lumping of multiple rather different species into a single genus.

It is surprising, though, that the second and third best represented sauropods in museums, Diplodocus and Apatosaurus, are both barely represented in terms of complete necks. And while the number of complete and nearly-complete necks among somphospondyls, including titanosaurs, is encouraging, it is disappointing that so many of them are not yet described.

At the time of writing (21 December 2021), the Paleobiology Database (https://paleobiodb.org/) lists 342 sauropod species. The nine unambiguously complete and articulated necks therefore represent only one in 38 known sauropod species—and recall that even these are mostly “complete” only in the weak sense of preserving some part of each cervical vertebra.

As best we can tell, only one sauropod species, Camarasaurus lentus, is known from more than a single complete neck. Of the two individuals, CM 11338 is a juvenile and USNM 13786 is a subadult, so the mature morphology is unknown. If we allow necks missing the atlas, then there are also two individuals of Shunosaurus lii: ZDM T5401, the subadult holotype, which is missing its atlas; and ZDM T5402, an adult referred specimen whose neck is complete. (These specimens have not been described in English.) With sample sizes this small, it is not possible even in principle to determine whether there is a bimodal distribution in the length of any sauropod’s neck.

Several well-known sauropod specimens are often thought of as having complete, undamaged necks, but in each case the truth is less clear. I now discuss three important specimens.

Diplodocus carnegii CM 84

The Carnegie Diplodocus is one of the most recognised dinosaurs in the world: not only is the original specimen, CM 84, on display as a mounted skeleton in the Carnegie Museum in Pittsburgh, but casts are displayed in many other major museums (e.g. the Natural History Museum in London, the Museum für Naturkunde Berlin and the Muséum National d’Histoire Naturelle in Paris.) The neck appears complete in these mounted skeletons, with fifteen cervical vertebrae, and is illustrated as such by Hatcher (1901: plate 8); Fig. 3. However, the situation is not as clear as it appears in these exhibits.

Holland (1900: 816), in the first published account of the Carnegie Diplodocus, assigned to this specimen only eleven cervicals, noting (on p. 817) that:

• The cervicals were for the most part interarticulated, all lying in such position as to show the serial order […] Eleven are found in the specimen at the Carnegie Museum, atlas and axis being as yet undiscovered.

Allowing for the missing atlas and axis, Holland concluded only that the cervical count was “at least 13”.

However, Hatcher (1900: 828–829) corrected this count later the same year:

• About 45 feet (14 m) of the vertebral column is preserved in our specimen. When discovered the vertebrae did not lie in a connected and unbroken series, yet there can be little doubt that they all pertain to the same individual […] Unfortunately no diagram was made, at the time of exhuming the remains, showing the relative position of each of the vertebrae in the quarry […] Early last spring, at the request of the writer, Mr. W. H. Reed (who assisted in unearthing the skeleton), while again on the ground, made a diagram of the quarry, showing the relative positions, as he remembered them, of the various bones of the skeleton.

Despite this uncertainty, Hatcher asserted (p. 828–829):

• “In all 41 vertebrae are represented, including 14 cervicals (all but the atlas) […] Assuming that no vertebrae are missing from our series the vertebral formula of Diplodocus should now be written as follows: Cervicals, 15 […] The number of cervical vertebrae in Diplodocus is definitely fixed at at least 15.”

Hatcher’s (1900) paper is unsatisfactory in that it gives no reason for his revision of the cervical count. Hatcher also hedged by leaving open the possibility of there being more than 15 cervicals. The lack of a reliable quarry map is unfortunate.

In his subsequent monograph, Hatcher (1901: 4) expanded on the completeness and condition of the material as follows (emphasis added):

• [Diplodocus carnegii holotype CM 84] has been entirely freed from the matrix and is found to consist of [appendicular material and] forty-one vertebrae divided as follows: fourteen cervicals including the axis, eleven dorsals, four sacrals, and twelve caudals. These vertebrae are for the most part fairly complete, though unfortunately the sacrals and anterior cervicals are more or less injured. This series of forty-one vertebrae are believed to pertain to one individual and to form an unbroken series from the axis to the twelfth caudal, although as was shown in a previous paper, there is some evidence that there are perhaps one or more interruptions in the series and that one or more vertebrae are missing. On the other hand, as will appear later, it is not entirely impossible that at least one vertebra of this supposed series pertains to a second individual belonging perhaps to a distinct genus.

Hatcher (1901: 11) went on to quote a statement from A. S. Coggeshall, who had assisted in the excavation, explaining in more detail how the elements of the neck were discovered:

• [The] last (fifteenth) cervical was considerably removed from the succeeding dorsals and less so from the preceding cervicals. Commencing with the next vertebra (cervical fourteen), the direction of the entire cervical series was altered so that it lay with its axis almost at right angles to that of the dorsal series. The cervicals extended in an almost straight line from the fourteenth to the fifth, but there was a considerable gap between cervicals 11 and 10, while the axis and cervicals three, four and five were doubled back under the succeeding vertebrae.

This account almost explains why Holland underestimated the number of cervicals: the anteriormost four, lying under more posterior cervicals, had not yet been found. However, if ten cervicals (C6–C15) had been found and the atlas and axis were both missing, Holland (1900: 816) would surely have stated “Ten are found in the specimen at the Carnegie Museum, atlas and axis being as yet undiscovered” rather than eleven. Some mystery remains: perhaps Holland was aware of one of the anteriormost four preserved cervicals.

Coggeshall’s description is somewhat corroborated by Reed’s quarry map, which is included as Plate 1 of Hatcher’s (1901) monograph (Fig. 4). However, the map is in some respects at odds with the description: for example, it shows all 13 vertebrae C2–C14 in a single straight line rather than indicating that C2–C5 were doubled back; and it shows gaps both between C10 and C11 (as stated), and also between C11 and C12 (not mentioned in the text).

Regarding the vertebra that might pertain to “a second individual belonging perhaps to a distinct genus”, Hatcher (1901: 22) explained: “Eleventh Cervical.—This vertebra is so unlike either the immediately preceding or succeeding vertebrae that if it had been found isolated it would have been unhesitatingly referred to a different genus. Mr. Coggeshall, however, assures me that it was interlocked with the succeeding, or twelfth cervical.” Yet, as noted, the quarry map suggests that there was some distance between C11 and C12, perhaps invalidating Coggeshall’s assertion. It is to be lamented that both the map and the description were created some time after the excavation actually took place, by which time memories had evidently become unreliable.

In conclusion, Diplodocus carnegii most likely had fifteen cervicals, but may have had more (if some vertebrae were not recovered), or maybe fewer (if C11 was misassigned). Furthermore, the anterior cervicals are damaged in a way that is not at all apparent from Hatcher’s drawings (plate III) or photographs (plate IV) because they were restored before these illustrations were prepared. As Hatcher (1901: 23) noted, “The work of freeing these vertebrae from the matrix and restoring them was for the most part done during my absence in the field. Unfortunately no drawings or photographs were taken prior to the process of restoring with colored plaster.” (In the early 20th Century, it was routine to restore damaged fossils in ways that completely obscured the degree of damage: see Fig. 5)

Apatosaurus louisae CM 3018

Apatosaurus louisae is the best known species of Apatosaurus, since its holotype CM 3018 is much more complete and better preserved than that of the type species A. ajax (YPM 1680), or that of the closely related Brontosaurus excelsus (YPM 1980).

The specimen was collected by Earl Douglass in 1909 and 1910, from what was then known as the Carnegie Museum Dinosaur Quarry near Jensen, Utah, and is now Dinosaur National Monument. It was mounted for exhibition in 1913, and somewhat belatedly named the type of a new species in a brief initial description by Holland (1915). He noted that “the specimen consists of a series of vertebrae, complete from the atlas to nearly the end of the tail” and appendicular material; but also that “the cervical vertebrae had been separated from the dorsals and shifted, but the entire series was found articulated in regular order” (p. 143). (Holland’s description also mentioned that “a skull, which judging by its location, belongs to the specimen, was found within eleven feet of the atlas. It does not differ greatly in form from the skull which belongs to Diplodocus”. Had Holland stuck to his guns, Apatosaurus could have been restored with its correct skull 63 years before Berman & McIntosh (1978) corrected Marsh’s long-standing and influential misapprehension that it had a Camarasaurus-like skull.)

Holland stated (p. 144) that he had “in preparation a large monographic paper relating to the genus, based in part upon [CM 3018]”. However, completion was long delayed, and Holland died in 1932 before in was ready to be published. It was eventually brought to completion by Gilmore (1936) and it is from this monograph that the species is primarily known.

Gilmore’s monograph explains that all is not as it seems in the neck of his specimen. He notes (p. 191) that “there was some distortion due to the compression to which [the cervicals] had been subject, but this has been largely corrected during preparation”—a questionable decision, as it means that the shapes of the vertebrae as originally found are now lost, and cannot be subjected to more modern retrodeformation techniques (e.g. Tschopp, Russo & Dzemski, 2013). He continues “Cervicals thirteen, fourteen, and fifteen, however, were so badly crushed that it was thought best to replace them in the mounted skeleton by plaster restorations of these vertebrae”, although he does claim that “they are, however, sufficiently well preserved so that most of their important characteristics can be determined”. The caption to Gilmore’s plate XXIV reads “Cervical vertebrae of Apatosaurus louisae. Type, No. 3018 […] Cervicals 13, 14 and 15 have been much restored from badly crushed originals, and should be used with caution.” It is also evident from this plate that most of C5 is also missing, although this is not acknowledged in the text. As noted by Upchurch (2000), the poor condition of the posterior cervical vertebrae, and their replacement by plaster models in the mounted skeleton, compromise the validity of biomechanical modelling based on this specimen, such as that of Stevens & Parrish (1999).

In conclusion, while the articulation of the cervical sequence of CM 3018 leaves little doubt that all cervicals are present and in the correct order, the crucial posterior cervicals are largely uninformative.

Giraffatitan brancai MB.R.2181

This specimen is the paralectotype of Giraffatitan brancai (= “Brachiosaurus” brancai). Much of the material is incorporated in the mounted skeleton in the atrium of the Museum für Naturkunde Berlin, which remains the largest substantially real mounted skeleton of a terrestrial animal anywhere in the world. (There are larger mounts of sauropods, such as the skeletons of Patagotitan at the AMNH and FMNH, but these are casts and sculptures, not real material.) While most of the material of the Berlin brachiosaur mount is real fossil bones, the presacral vertebrae are too heavy and fragile to mount: instead, high quality sculptures are used, and the vertebrae themselves are held in collections.

The presacrals in the mount are based on real bones that are from two specimens—the lectotype MB.R.2180 (formerly known as SI) and the paralectotype MB.R.2181 (formerly SII). The former includes cervicals 2–7, an assignment that can be accepted with some confidence if the vertebrae indeed form a sequence because the axis, C2, in sauropods is very distinctive, having a completely different anterior articular surface from all the subsequent cervicals (see e.g. Janensch, 1950: figs. 9–16, cf. figs. 17–48.) MB.R.2181 includes cervicals assigned by Janensch to positions 3–13 (although almost all of them are damaged, some very severely).

However, the two individuals MB.R.2180 and MB.R.2181 were found together in a single quarry (designated Quarry S). Bones of the two individuals were jumbled up together, with little articulation, as shown in the quarry map, redrawn by Heinrich (1999: figure 16; Fig. 6) from an original drawn in the field by Werner Janensch. Any reconstruction—or even assignment of individual vertebrae to one specimen or the other—must be considered provisional.

I have previously suggested (Taylor, 2009: 800–801) that the distinctively high-spined dorsal vertebra usually considered the fourth of MB.R.2181 may not actually belong to that specimen, or even that taxon. Instead, this unusually tall vertebra may belong to an animal more closely resembling the Tendaguru titanosauriform briefly described by Migeod (1931) and which I am in the process of redescribing (Taylor, 2005, Taylor in prep at https://github.com/MikeTaylor/palaeo-archbishop/). If this vertebra is indeed not part of MB.R.2181 then the most likely inference is that it is part of MB.R.2180. This would be unfortunate if these two specimens were indeed representatives of different taxa. The smaller and less complete MB.R.2180, rather than the larger, more complete and better known MB.R.2181, is the lectotype (Janensch, 1935). Therefore, the ICZN rules dictate that the name Giraffatitan brancai would adhere to MB.R.2180, and that a new name would be required for the better-known MB.R.2181. Since this species was thought until relatively recently to be a species of the North American genus Brachiosaurus (see Taylor, 2009), a further reassignment would mean that this charismatic and iconic specimen would become known by a third different name in little more than a decade. To avoid this outcome, an ICZN petition may be warranted.

Janensch (1950: 33) indicates that the confusion of the cervical vertebrae is not as bad as that of the dorsals, but the situation is still far from clear, as Janensch’s description explains:

• The vertebrae from the 3rd to 15th presacrals [of MB.R.2181] lay in articulation in a consolidated lime sandstone lens; of them, the 3rd to 5th vertebrae are tolerably complete, the remaining 10 vertebrae were articulated with one another, with one interruption that arose when the 8th presacral vertebra rotated out of the series and was displaced. [Translation by Gerhard Maier.]

Janensch’s words are somewhat at odds with the quarry map, which shows cervical material scattered around the northeastern part of the quarry (Fig. 6). Other quarry sketches drawn by Janensch, including one reproduced by Heinrich (1999: figure 18), certainly show at least some cervicals having been found in articulation, but it is not clear what the correspondence is between the field numbers on these sketches and subsequently assigned element identifications.

So it is possible, though unlikely, that there might have been other displaced cervicals, before or after the one designated “8th”, that were not recovered. Neither can we be wholly confident that the anteriormost preserved cervical in the MB.R.2181 series is really C3. Its identification is based on the overlap with vertebrae of MB.R.2180, but we cannot be certain that MB.R.2180 is a member of the same species as MB.R.2181. Perhaps the anteriormost preserved cervical is really C4? Or perhaps some of the “MB.R.2181” cervicals really belong to MB.R.2180. This is not particularly likely, as Janensch pointed out, due to the difference in size between the two specimens—the MB.R.2181 cervical centra are about 30–40% longer than those assigned the same serial position in MB.R.2180. But the size difference cannot be considered a conclusive argument, especially given the uncertainty in serial-position assignments in both specimens.

In conclusion: Giraffatitan brancai probably had thirteen cervicals, but may have had more, or possibly less; and the neural arches are only known for cervicals 3, 4, 5 and 8 in MB.R.2181 (if these are the correct serial positions for those vertebrae). If MB.R.2180 is indeed a member of the same species then cervicals 2–7 are known from well-preserved elements, but no more. All of this uncertainty is exacerbated by the problem that no complete or even nearly complete neck of any other brachiosaur has been described.

To summarise this section, not only are complete sauropod necks in very short supply, even those that are considered complete cannot generally be confidently considered so, and complexities of interpretation bedevil the best-known specimens.

Taphonomic factors

All of the problems with sauropod neck preservation arise from the nature of the animals and the general procedures of taphonomy.

First, sauropods are big. This is a recipe for incompleteness of preservation: small skeletons are more easily destroyed by taphonomic processes, but if they survive are more easily preserved whole, while large skeletons less rarely survive intact (Brocklehurst et al., 2012). It is no accident that the most completely preserved sauropods are small individuals such as CM 11338, the cow-sized juvenile Camarasaurus lentus described by Gilmore (1925). For an organism to be fossilised, it is necessary for the carcass to be swiftly buried in mud, ash or some other substrate. This can happen relatively easily to small animals, such as the many finely preserved small theropods from the Yixian Formation in China, but is much less possible with a large animal (Mannion, 2010: 284).

Cleary et al. (2015: 528 and figure 6) showed that medium-sized ichthyosaurs preserve more completely than either small or large individuals, but since these are aquatic animals their preservational context is not applicable to the case of sauropods. Brown et al. (2012) found that in the Dinosaur Park Formation, “large-bodied” dinosaurs preserved more completely than smaller ones, but their sample contained no sauropods, their threshold for “large” was only 60 kg, and the largest animals included were 4.5-tonne hadrosaurs. It may be that if the methods of Brown et al. (2012) were used to analyse the sauropod-bearing Morrison or Tendaguru formations, the result would be similar to those of Cleary et al. (2015), with medium-sized animals having the most complete preservation. At a larger scale, Cashmore et al. (2020: 963) found only a weak, and weakly significant, correlation between sauropod body mass and specimen completeness (R² = 0.03; p = 0.04), using the SCM2 skeletal completeness metric of Mannion & Upchurch (2010). However, analysis of their figure 5A, plotting log body mass against log completeness shows a rapid falling away in completeness above body masses with log 10—presumably natural log, representing about 22 tonnes.

It is also possible that the light construction of highly pneumatic cervical vertebrae would have rendered them particularly prone to water transport, disarticulating and scrambling the necks of even some otherwise adequately preserved specimens.

Except in truly exceptional circumstances, sediments simply are not deposited quickly enough in terrestrial environments to cover a 25 m, 30 tonne animal with a light neck skeleton before it is broken apart by scavenging, decay and water transport. Fossilisation of the very largest sauropods tends to produce even more fragmentary remains. In light of this, it is not surprising that the very longest sauropod necks are usually known from particularly inadequate specimens. The longest neck for which we have direct evidence is that of the diplodocid Supersaurus, possibly 15 m long, but the only cervical material of the largest specimen is a single 1.4 m cervical (BYU 9024, formerly BYU 5003; Jensen, 1985, 1987). Similarly, the giant basal titanosauriform Sauroposeidon probably had a neck about 11 m long, but the only definite material belonging to it is a sequence of three and a half cervicals from the middle of the neck (OMNH 53062; Wedel, Cifelli & Sanders, 2000). The longest known titanosaur necks are probably those of Patagotitan, Puertasaurus and Dreadnoughtus, all at around 9–10 m, but the cervical material from which they are known is meagre: only three vertebrae in the Patagotitan holotype MPEF-PV 3400, of which the longest is 120 cm long (supplementary information to Carballido et al., 2007); a single 118 cm Puertasaurus vertebra, MPM-PV 10002 (Novas et al., 2005); and a single 113 cm vertebra of Dreadnoughtus MPM-PV 1156 (Lacovara et al., 2014).

Secondly, even when complete sauropod skeletons are preserved, or at least complete necks, distortion of the preserved cervical vertebrae is almost inevitable because of their uniquely fragile construction. As in modern birds, the cervical vertebrae were lightened by extensive pneumatisation, so that they were more air than bone (Taylor & Wedel, 2013a: figure 4), with the air-space proportion typically in the region of 60–70% and sometimes reaching as high as 89% (Taylor & Wedel, 2013a: table 2; Wedel, 2005: figure 7.4C). While this construction enabled the vertebrae to withstand great stresses for a given mass of bone, it nevertheless left them prone to crushing, shearing and torsion when removed from their protective layer of soft tissue. For highly pneumatized cervicals in particular, the chance of the shape surviving through taphonomy, fossilisation and subsequent deformation would be tiny.

Possible mitigations

Some information about sauropods necks that is not directly observable can be inferred from phylogenetic bracketing: the observation that if two close outgroups to an organism of interest both have a feature, then the null hypothesis is that the organism of interest inherited that feature from the common ancestor (see Witmer, 1995). For example, the neck of Sauroposeidon proteles is known only from a sequence of four vertebrae (Wedel, Cifelli & Sanders, 2000). But as a brachiosaurid it is bracketed by Giraffatitan brancai, which is thought on reasonable evidence (see above) to have had 13 cervicals, and Camarasaurus lentus, which is known from complete specimens to have had 12 cervicals (Gilmore, 1925: 367). It is therefore parsimonious on this basis to conclude that the most recent common ancestor of Giraffatitan and Camarasaurus probably had 12–13 cervical vertebrae, and that Sauroposeidon proteles likewise also had 12–13 cervicals.

While this level of inference is better than no information, the present example illustrates three problems with the use of phylogenetic bracketing for sauropod morphology. First, due to incomplete knowledge of most sauropods, it is often necessary to move some way down the tree from the organism of interest before reaching specimens for which morphology is sufficiently known. In this example, while Giraffatitan is probably close to Sauroposeidon, Camarasaurus is some way basal to it—but more closely related brachiosaurids such as Abydosaurus and Europasaurus, while possessing good cervical material, do not have anything close to a complete neck, and so cannot be used in bracketing. As a result, the Giraffatitan–Camarasaurus bracket used in this example is broader than we would wish. Second, the phylogenetic context in which we interpret brackets is not always well established. In the present example, I have been following Wedel, Cifelli & Sanders (2000) referral of Sauroposeidon to the clade Brachiosauridae; but if D’Emic & Foreman (2012) are correct in their assessment that Sauroposeidon is in fact a somphospondylian, then it would be more appropriate to bracket it between Giraffatitan brancai and Euhelopus zdanskyi—which has 17 cervical vertebrae (Wiman, 1929: 7), suggesting a significantly higher cervical count of 13–17 for Sauroposeidon. Third, in the case of sauropods, the morphology of the bracketing taxa is often not as solidly established as we might wish. In this case, while the cervical counts of Camarasarus lentus and Euhelopus zdanskyi are well established, that of Giraffatitan brancai is less so, as documented above, rendering the cervical-count ranges for Sauroposeidon even more uncertain. In conclusion, while phylogenetic bracketing remains a useful tool, it must be used with care.

The problem of distortion in individual vertebrae can be addressed to some extent through retrodeformation (e.g. Schlager et al., 2018). This term encompasses a suite of techniques in which mathematical processes are applied to virtual models of fossils in an attempt to restore them to the shape they had before their deformation in matrix. Various methods have been proposed in the literature, largely based on an assumption of symmetry. Most of the early methods were designed for 2D transformations on morphologically simple invertebrate fossils (e.g. Cooper, 1990), but more recent developments have yielded methods more capable of handling complex 3D shapes (e.g. the Single Axis Method of Kazhdan et al., 2009). Application of these techniques to sauropod vertebrae has so far been limited, but Tschopp, Russo & Dzemski, 2013 conducted experiments on cervical vertebrae of the diplodocid Kaatedocus siberi using an artificially deformed model of a dodo cervical as a control. They found that while the retrodeformation techniques they tested were capable of restoring symmetry to a deformed vertebra, they often did so at the expense of artificially making the vertebra shorter, broader or thinner than it should be. Surprisingly, they also found that applying the same process twice to a vertebra first made it more robust than the original, then more slender. At this stage it seems that more sophisticated techniques, preserving more aspects of the 3D geometry, will be necessary before retrodeformed sauropod cervicals can be taken at face value.

But this does not mean that other strategies cannot help in dealing with distorted material. For example, Hedrick & Dodson (2013) used 3D geometric morphometric techniques to better understand individual and taphonomic variation in a set of 30 skulls representing three nominal species of Psittacosaurus. They showed that the apparent morphological variation between these “species” actually represent the results of different taphonomic processes, and that all 30 skulls in fact represent members of the same species. It may be possible to use similar techniques on sauropod vertebrae, although modifications would be required to deal with the additional complexity introduced by serial variation in vertebral morphology.

One thing that is apparent in all this is the necessity of careful documentation on how specimens were found, in what states of articulation, in what kinds of matrix, and under what strains: for example, Angielczyk & Sheets (2007) use strain measurements to quantify deformation. These descriptions should where possible be accompanied by quarry maps, and ideally by photos and even 3D models of the excavation—something that is increasingly possible given the unlimited capacity for online supplementary information in many journals. Such additional documentation may not materially increase our understanding of the specimens but does at least make it more explicit what we do and do not know, yielding rigidly defined areas of doubt and uncertainty. Similarly, when fossils are restored, it should be done in such a way that it is apparent what parts of the restored specimen are real bone and which are reconstructed. Happily, modern palaeontologists are much better in all these respects than our predecessors were. Some “golden age” palaeontologists were concerned primarily with what would constitute a spectacular museum mount: it is for this reason that O. C. Marsh had YPM 1980, the holotype of Brontosaurus excelsus, so enthusiastically “restored” that it is now impossible to determine which parts of the cervical vertebrae are real (personal observation; Barbour, 1890). But these days are mostly behind us, and nearly all modern specimens are prepared in a form that accurately conveys what was actually preserved.

Implications

Both the incompleteness and distortion of sauropod necks have grave consequences for our ability to reason about sauropods. As noted above, the very small sample of complete necks makes it quite impossible to perform meaningful statistical analyses. This incomplete information also impedes our ability to understand the evolution of sauropod necks, making it difficult to determine the plesiomorphic cervical morphology and vertebral counts at the bases of clades. Similarly, the frequent, unpredictable and sometimes dramatic distortion of what vertebrae we do have renders mechanical analysis of neutral poses and ranges of motion extremely problematic. For vertebrae small and robust enough to be manipulated by hand, this can be readily observed in physical space (Fig. 13). There is no reason to think that computer modelling of vertebrae and their articulations should yield models any more informative than the distorted fossils that they are based on.

On a more positive note, the lack of complete necks does not mean that we are without information. For many sauropods that lack complete necks, enough vertebrae are preserved with enough fidelity that we can have a good idea how morphology varies between anterior, middle and posterior cervicals, even if precise identification of the vertebrae is not possible. Crucially, this degree of completeness suffices for the majority of characters to be scored in phylogenetic analyses: apart from a few characters specific to the atlas or axis, most such characters pertain only to anterior, middle or posterior cervicals.