A complex hyobranchial apparatus in a Cretaceous dinosaur and the antiquity of avian paraglossalia

Transcription

1 bs_bs_banner Zoological Journal of the Linnean Society, 2015, 175, With 10 figures A complex hyobranchial apparatus in a Cretaceous dinosaur and the antiquity of avian paraglossalia ROBERT V. HILL 1 *, MICHAEL D. D EMIC 2, G. S. BEVER 1,3 and MARK A. NORELL 3 1 Department of Anatomy, New York Institute of Technology College of Osteopathic Medicine, Northern Boulevard, Old Westbury, NY USA 2 Department of Anatomical Sciences, Stony Brook University, HSC-T8, Room 040, Stony Brook, NY , USA 3 Division of Paleontology, American Museum of Natural History, New York, NY 10024, USA Received 6 March 2015; revised 13 May 2015; accepted for publication 16 May 2014 The highly specialized feeding apparatus of modern birds is characterized in part by paraglossalia, triangular bones or cartilages in the tongue that constitute part of the rarely fossilized hyobranchial apparatus. Here, we report on a new, juvenile specimen of the ankylosaurid dinosaur Pinacosaurus grangeri Gilmore, 1933 that provides the first evidence of paraglossalia outside of crown group Aves. The specimen is remarkable in preserving a wellossified hyobranchial apparatus, including paired paraglossalia, first and second ceratobranchials, epibranchials, and evidence of a median cartilaginous basihyal. Reassessment of Edmontonia, another ankylosaur, also reveals evidence of bony paraglossalia. Ankylosaur paraglossalia closely resemble those of birds, but are relatively larger and bear prominent muscle scars, supporting the hypothesis that ankylosaurs had fleshy, muscular tongues. The other hyobranchial elements, surprisingly, resemble those of terrestrially feeding salamanders. Ankylosaurs had reduced, slowly replacing teeth, as evidenced from dental histology, suggesting that they relied greatly on their tongues and hyobranchia for feeding. Some curved, rod-like elements of other dinosaur hyobranchia are reinterpreted as second ceratobranchials, rather than first ceratobranchials as commonly construed. Ankylosaurs provide rare fossil evidence of deep homology in vertebrate branchial arches and expose severe biases against the preservation and collection of the hyobranchial apparatus. In light of these biases, we hypothesize that paraglossalia were present in the common ancestor of Dinosauria, indicating that some structures of the highly derived avian feeding apparatus were in place by the Triassic Period.. doi: /zoj ADDITIONAL KEYWORDS: ankylosauria birds dinosauria hyobranchial apparatus paraglossal bone. INTRODUCTION The hyobranchial apparatus (or hyobranchium) is emblematic of one of the most profound transformations in vertebrate evolution: the origin of tetrapods from exclusively gill-breathing ancestors. Among reptiles, the hyobranchium comprises derivatives of the second visceral (hyoid) arch, and one or two more caudal arches (Schwenk, 2000b). It supports the tongue within the *Corresponding author. rhill01@nyit.edu oral cavity, and its more caudal elements lie embedded within the pharynx. The morphology of the hyobranchial apparatus in reptiles is extremely diverse (Fürbringer, 1922; Gnanamuthu, 1937; Romer, 1956), and anatomical work over the last century has suffered from non-uniform terminology that impedes comparative study (Homberger, 1986; Reilly & Lauder, 1988; Homberger & Meyers, 1989). In general, the hyobranchium comprises a midline body (corpus hyoidei) and paired, laterally extending horns (cornua) derived from the visceral arches (Schwenk, 2000b). The reptilian hyobranchial 892

2 DINOSAUR HYOBRANCHIAL APPARATUS 893 apparatus is partly or mostly cartilaginous, and is incompletely preserved in the fossil record. The hyobranchial apparatus in fossil Dinosauria, therefore, remains poorly understood. Among non-avian dinosaurs, the elements most commonly fossilized are identified as first ceratobranchials (Morschhauser & Lamanna, 2013). These are usually rod-shaped and lack distinguishing anatomical landmarks. The most noteworthy exception comes from the ankylosaurid Saichania chulsanensis Maryańska, 1977, which preserves one pair of ceratobranchials, putative fragmentary ceratohyals, and a triangular midline lingual process thought to be composed of a combined basihyal and basibranchial that supported a long, flexible, and intrinsically muscular tongue (Maryańska, 1977). The specimen reported here is remarkable in that it preserves the most complex hyobranchial apparatus yet described for a non-avian dinosaur. It further provides new information about the anatomy and functional morphology of the feeding system in ankylosaurs, and the evolution of the hyobranchium in dinosaurs, including birds. INSTITUTIONAL ABBREVIATIONS AMNH FARB, American Museum of Natural History, Fish, Amphibians, Reptiles and Birds Collection, New York, USA; IGM, Institute of Geology, Ulaan Baatar, Mongolia; IVPP, Institute of Vertebrate Paleontology and Paleoanthropology, Beijing, China; ZPAL, Zoological Institute of Paleobiology, Warsaw, Poland. SYSTEMATIC PALAEONTOLOGY DINOSAURIA (OWEN, 1842) ORNITHISCHIA (SEELEY, 1887B) THYREOPHORA (NOPCSA, 1915) (SENSU NORMAN, 1984) ANKYLOSAURIA (OSBORN, 1923) ANKYLOSAURIDAE (BROWN, 1908) PINACOSAURUS GRANGERI (GILMORE, 1933) New referred specimen: IGM 100/3186, nearly complete skull and articulated cervical vertebral series, ossified and articulated hyobranchial apparatus, cervical ribs, first cervical osteodermal half ring, left radius, and associated osteoderms. A partial postcranial skeleton was also collected and is currently under preparation. Age and locality: Middle Campanian, Death Row sublocality, Ukhaa Tolgod, Ömnögovi Aimag, Mongolia. Facies S, structureless sandstone indicative of masswasting sandslides (Dingus et al., 2008). The burial of dinosaurs and other vertebrates at this locality was rapid and complete, resulting in well-preserved and articulated specimens. Diagnosis The specimen is referred to Pinacosaurus grangeri (Gilmore, 1933) on the basis of the following characters: primary airway bounded by dorsally embayed osteodermal mass, lacrimal incisure (pinching of snout in dorsal view), accessory openings into premaxillary sinus, squamosal horns small and pyramidal, and width across squamosal horns not greater than width across supraorbitals (Arbour & Currie, in press; Hill, Witmer & Norell, 2003; Burns et al., 2011). History of discovery and preparation IGM 100/3186 was discovered in the summer of 2005 at the Death Row locality of Ukhaa Tolgod, Ömnögovi Aimag, Mongolia (Dingus et al., 2008). The specimen is characteristic of dinosaur fossils from this locality, being relatively completely preserved and with individual bones in complete articulation or close association. A block containing the skull and cervical vertebrae was excavated and jacketed. The block was scanned by computed tomography (CT) using a GE Lightspeed VCT scanner at the Stony Brook University Medical Center. As a result of poor contrast between the bone and matrix, internal visualization of the block was limited, and so the specimen was scanned several times as mechanical preparation proceeded. Scans were taken at mm slices and intervals, 120 kv, and varying amperage ( ma) in order to produce the clearest images. Although bone was evident on the surface, the first scans did not reveal any internal structures. As the block was mechanically prepared it became clear that a nearly complete skull was preserved within. The dorsal surface was prepared partially, revealing the skull roof, osteoderms overlying the snout, and the first cervical osteoderm half ring. Mechanical preparation of the ventral surface of the specimen uncovered the mandible and part of the hyobranchial apparatus (Fig. 1). Additional CT scans allowed some superficial digital segmentation of the hyobranchium; however, similar density between bone and matrix made it impossible to fully characterize the anatomy of this structure. The predentary bone was removed and prepared separately. Two additional ossifications, found immediately caudal to the left hemimandible, were recognized as the interorbital ossifications, i.e. the orbitosphenoid and laterosphenoid bones. Further preparation isolated the hyobranchial apparatus, which was ultimately prepared separately. Scans of the isolated hyobranchium provided the contrast necessary to fully segment and digitally

3 894 R. V. HILL ET AL. Figure 1. Pinacosaurus grangeri (IGM 100/3186). Stereophotographs of articulated skull and cervical region in ventral view, showing hyobranchial elements at the centre. Abbreviations: bs, basisphenoid; cb2, second ceratobranchial; cvr, cervical rib; eb, epibranchial; lpg, left paraglossalium; ls, laterosphenoid; mo, mandibular osteoderm; os, orbitosphenoid; pmx, premaxilla; ppf, premaxillary palatal fenestra. reconstruct the structure using AVIZO 7.0 ( Three displaced teeth and the left articular bone remained in close association with the rostral elements of the hyobranchium. Moderate additional preparation of the palatal region exposed the basioccipital and exoccipital elements, which had become dislodged, and drifted into the palate. The relative independence and lack of fusion of these bones further attest to the juvenile age of the individual. Original CT data and movie files are publicly available as project P2101 at (O Leary & Kaufman, 2012). DESCRIPTION The skull exhibits morphology characteristic of derived ankylosaurids, being triangular in dorsal view, roughly as long as it is wide. The nasal region is heavily ornamented by osteoderms and other cranial rugosities (Vickaryous, Russell & Currie, 2001; Hill et al., 2003; Arbour & Currie, 2013). The skull roof caudal to the frontonasal suture is devoid of osteoderms or ornamentation, allowing the sutures between more caudal skull elements to be easily discerned. The parietals are fused into a single element. The lateralmost supraorbitals are elaborated into a sharp, pyramidal horn, and the secondary osteoderms overlying the squamosals are also pyramidal and rugose. IGM 100/3186 resembles other known specimens of Pinacosaurus in having multiple accessory apertures in the narial region (Fig. 2). Five openings are present on each side of the snout in IGM 100/3186, similar to the condition in IGM 100/1014 (Hill et al., 2003). The dorsalmost opening represents the true naris, which opens into the airway. Ventral to this are three more or less round openings (structures C1 C3; Hill et al., 2003), which open into a premaxillary sinus cavity. Caudal to these apertures is a slightly rostrocaudally elongate opening corresponding to structure B (Hill et al., 2003). As with other specimens of Pinacosaurus, it is unclear whether this aperture actually opened into a sinus or was an externally located fossa that housed soft tissue. The premaxillary palate is fenestrated by two large openings, each one located near the caudolateral margin of the premaxilla (Fig. 2). CT scans demonstrate that these openings communicate with the premaxillary sinus, and are therefore continuous with the external openings into this space (i.e. apertures C1 C3; Hill et al., 2003). The bilateral symmetry and finished edges of these openings suggest that they are actual fenestrae, rather than artefacts of breakage. Premaxillary palatal fenestrae have not been previously described in detail in Pinacosaurus or other ankylosaurs, although they appear to be present in certain illustrations depicting the palate (Burns et al., 2011), but are absent in others (Maryańska, 1977; Hill et al., 2003). Because the bone in this region is exceedingly thin, the margins of the fenestrae are susceptible to breakage. Their presence and natural size may be misinterpreted in all but the most completely preserved specimens. The first cervical half ring is preserved in association with the skull. It consists of six rectangular osteoderms that meet one another at deeply

4 DINOSAUR HYOBRANCHIAL APPARATUS 895 Figure 2. Pinacosaurus grangeri from the Upper Cretaceous of Mongolia, showing accessory narial apertures. Top, ZPAL MgD II/1, juvenile specimen of P. grangeri showing intact narial region; inset at lower left shows homologous portion on IGM 100/3186. Lower right, ventral (palatal) view of rostrum, showing the opening of the right premaxillary palatal fenestra. Abbreviations: B, structure B in the narial region; C, apertures C in the narial region, opening into the premaxillary sinus cavity; mx, maxilla; mxo, maxillary osteoderm, pmx, premaxilla; ppf, premaxillary palatal fenestra; v, vomer. interdigitating sutures. In articulation, the ring encircles more than two-thirds of the circumference of the neck. In ventral view the arch of the mandible is clearly visible, as it is preserved in articulation with minimal distortion. A long, triangular mandibular osteoderm is associated with each hemimandible. The predentary was dislodged into the oral cavity during or after deposition, but was preserved whole and then prepared separately. The hyobranchial apparatus (described below) dominates the space between the hemimandibles. Caudally, the ventral aspects of the cranial cervical vertebrae are visible. Each vertebra remains associated with its two cervical ribs; however, some shearing in this region has moved the ribs out of direct articulation. HYOBRANCHIAL APPARATUS The hyobranchial apparatus in IGM 100/3186 (Figs 3 4) consists of four bilaterally paired elements. The ventralmost and largest elements are interpreted here as paired paraglossalia (singular: paraglossalium; two fused paraglossalia = paraglossal). This is the first description of ossified paraglossalia outside of crown group Aves. Each paraglossalium has the shape of a right triangle, with a long, relatively straight rostrolateral margin. The ventral surface is flat overall, except at the rostrolateral margin where it is thickened and laterally rounded. The caudal margin is deeply incised, and separates a pointed caudolateral process from a more gently rounded caudomedial process. Each caudolateral process is itself excavated on its medial surface with two well-defined, equally sized fossae separated by a sharp ridge. The ventral surface of each paraglossalium also bears a prominent, linear muscle scar that lies parallel with the rostrolateral border of the bone. Near the caudal margin, each paraglossalium is pierced by a small foramen that communicates with the caudal margin via a slit-like perforation in the bone. The two paraglossalia meet along a loosely interdigitated midline suture, and are shifted slightly from life position. The dorsal surface of each paraglossalium is divided into a flat plate that meets its counterpart at the midline suture and a pronounced ridge that forms the rostrolateral edge of the bone. The ridge bears a prominent, lens-shaped fossa along its rostral half, and its caudal half forms the caudolateral process with its medially facing fossae. Caudodorsal to the paraglossalia are the first ceratobranchials. These are flattened bars elongated along a rostromedial caudolateral axis, with flared ends separated by a narrower waist. The rostromedial end of each element is gently concave ventrally, whereas the caudolateral end is slightly convex. On each bone there is a sharp rostrolateral process that is separated from the rest of the rostral end by a semicircular incisure. A tiny foramen enters each bone immediately caudal to its narrow waist, passing entirely through

5 896 R. V. HILL ET AL. Figure 3. Pinacosaurus grangeri (IGM 100/3186), detailed views of the hyobranchial apparatus. A, central part of prepared hyobranchial apparatus in dorsal view, showing deeply excavated fossae; B, oblique left ventrorostrolateral view of hyobranchial apparatus in situ; C, D, prepared right second ceratobranchial in (C) dorsal and (D) ventral views, showing complex curvature. Abbreviations: ar, articular; bs, basisphenoid; cb1, first ceratobranchial; eb, epibranchial; lpg, left paraglossalium; rcb1, right first ceratobranchial; rpg, right paraglossalium. to the dorsal surface. In dorsal view, the medial and lateral margins of each first ceratobranchial form sharp ridges that demarcate a trough-like fossa that extends for the entire length of the element. The lateral margins are roughened, bearing scars for muscle attachments. These muscle scars, along with the foramina and caudolateral processes of both the paraglossalia and first ceratobranchials, are interpreted as serially homologous features derived from consecutive visceral arches. Caudal to the first ceratobranchials is a pair of gracile, curved rods that represent the second ceratobranchial elements (Fig. 3C, D). These bones are gently flared at both their proximal and distal ends, and are complexly curved. First, they are curved about a dorsoventral axis, so that they present a concave rostral border and convex caudal border. Second, they are curved about two rostrocaudal axes, so that the lateral one-third is convex ventrally, whereas the medial onethird is slightly concave. The combined effect of these curvatures is a sigmoid bone that does not lie flat. We interpret a fourth pair of bones as epibranchials. These are by far the smallest hyobranchial elements and each has a flat, spatulate part that we interpret as the rostral end, tapering to a narrow, cylindrical caudal tip. These elements were found disarticulated, but in close association with other hyobranchial elements. These four paired elements represent derivatives of the laterally extending cornua ( horns ) of the hyobranchium. There is no ossified median element; however, there is evidence that a cartilaginous one was present, uniting these bones across the midline. The medial end of each second ceratobranchial bears an ovoid facet that faces rostromedially. The shape and orientation of these facets suggest articulation with a narrow, midline basihyal element. Its absence, despite the exquisite preservation of the specimen, suggests it remained cartilaginous, consistent with cartilaginous basihyal rods in many sauropsids (Fürbringer, 1922; Gnanamuthu, 1937). ONTOGENETIC STAGE OF THE SPECIMEN Pinacosaurus grangeri is unique among Ankylosauria, as it is known from multiple juvenile and subadult skulls (Arbour & Currie, in press; Vickaryous, Maryanska & Weishampel, 2004). Unlike most adult ankylosaur skulls, in which the cranial sutures are highly fused and obscured by fused osteoderms (Vickaryous & Russell 2003; Vickaryous et al., 2004), subadult specimens of P. grangeri lack the fusion of osteoderms in key regions, allowing the observation of sutures and permitting clearer anatomical interpretations. The holotype skull (AMNH FARB 6523) has generally been assumed to be an adult since its original description (Gilmore, 1933). It is badly crushed and shattered, leading to the erroneous interpretation that its skull roof was completely covered with osteoderms fused onto cranial bones. Re-examination of the holotype shows that sutures are visible in areas of the skull roof (Hill & Norell, 2008), and that the persistence or non-fusion of these sutures may be a generic and not an ontogenetic character. Nevertheless, several independent lines of evidence support prior interpretations that many Pinacosaurus skulls pertain to juvenile

6 DINOSAUR HYOBRANCHIAL APPARATUS 897 Figure 4. Pinacosaurus grangeri. Schematic reconstruction of life position of hyobranchial apparatus relative to skull bones. Composite reconstruction based on IGM 100/ 3186, IGM 100/1014 (Hill et al., 2003), IVPP V16853, IVPP V16283, and IVPP V16346 (Burns et al., 2011). Top: right lateral view, with right mandible removed to reveal hyobranchium. Bottom: ventral (palatal) view. or subadult individuals (Maryańska, 1971, 1977; Burns et al., 2011). The specimen described here, IGM 100/3186, is interpreted as a juvenile. In overall dimensions, it is 40% smaller than the holotype, and is nearly identical in all dimensions to IGM 100/1014 (Hill et al., 2003), which was also interpreted to be a juvenile. Like the latter specimen, it preserves the mandibular osteoderm in situ, yet unfused to the mandibular bones (Fig. 1). The neurocentral sutures of the cervical vertebrae remain unfused. Elements of the braincase, namely the basioccipital, orbitosphenoid/laterosphenoid, and otoccipital bones, are unfused to other skull elements, and after the death of the animal became disarticulated and transported to ectopic positions. The predentary was similarly dislodged, but this is not interpreted as evidence of a juvenile age, as the bone is thought to remain in loose articulation with the dentaries and allow jaw movements in multiple planes during feeding (Nabavizadeh, 2011). Foramina in the orbitosphenoid, laterosphenoid, and palate are large relative to the bones they penetrate, suggesting incomplete ossification. We took thin sections from an associated but isolated tooth (Fig. 5), as well as the left radius (Fig. 6), and an isolated osteoderm (Fig. 7), of IGM 100/3186. Specimens were embedded in epoxy resin, sectioned along the midline (in the labiolingual plane for the tooth, the transverse plane for the radius, and both a transverse and a sagittal plane for the osteoderm), mounted to glass slides and hand-ground and polished with increasingly fine sandpaper and alumina polish (320 grit to 0.05-μm powder) to a thickness of approximately 100 μm, following standard palaeohistology techniques (Lamm, 2013). The histological section of the large, disarticulated tooth shows very thin enamel (just a few dozen microns) around the tooth crown, surrounding well-preserved incremental lines of von Ebner through most of the dentine (Fig. 5). Incremental lines of von Ebner, which form daily in vertebrates (Erickson, 1996b), were on average 15.7 μm thick, with a standard deviation of 1.4 μm. The 16.5-mm-long disarticulated tooth that was sectioned took an estimated 75 days to form. We used this length age relationship to predict tooth age in functional and replacement teeth within the jaws of IGM 100/3186, following the method described by D Emic et al. (2013; Appendix S1). Tooth length was obtained via CT scans measured in IMAGEJ (Rasband, 2014). A maximum of one replacement tooth was present for each functional tooth, so only a minimum tooth replacement rate could be calculated, which was equivalent to the estimated age of each replacement tooth for a given alveolus. The oldest replacement tooth had an age of 53 days, and on average the replacement teeth were 48 days old. The tooth replacement rate could not have been higher than this number because in that case additional replacement teeth would be observed in each alveolus. This replacement rate is thus a minimum; furthermore, the replacement rate should be higher in this juvenile specimen than in adult specimens, based on the pattern in crocodilians (Erickson, 1996a). Tooth formation time, but not the replacement rate, was reported for the nodosaurid ankylosaur Edmontonia (Erickson, 1996b). The mean incremental line of von Ebner width of 13.9 μm inedmontonia is slightly less than the 15.7-μm width we found in Pinacosaurus. The much larger teeth of Edmontonia were estimated to have taken 279 days to form, nearly

7 898 R. V. HILL ET AL. Figure 5. Pinacosaurus grangeri (IGM 100/3186), dental histology of isolated tooth. A, close-up of labiolingual thin section showing well-preserved incremental lines of von Ebner (arrowheads); B, composite image of entire tooth sectioned in labiolingual plane. Incremental lines of von Ebner are traced in blue. four times the time estimated for Pinacosaurus. Our CT scan of Edmontonia reveals that, like Pinacosaurus, only a single replacement tooth was present in each alveolus, and there was a large size difference between functional and replacement teeth, suggesting relatively slow replacement. Though not quantifiable, it appears that ankylosaurs had substantially slower tooth replacement rates than did other ornithischians. The bone histology of the left radius of IGM 100/ 3186 reveals a well-vascularized cortex composed of woven to parallel-fibred tissue that grades internally into a spongy medullary cavity (Fig. 6). Vascular canals are mainly longitudinal in orientation, and are present and abundant to the periosteal margin, which is scalloped via partially formed primary osteons, indicating that bone deposition was continuing. Haversian remodelling is absent in the section, and at most a single, poorly defined LAG is present. No external fundamental system (EFS), an indicator of somatic maturity, was observed. These features indicate that the specimen was a juvenile at death, congruent with data from size and the relative lack of fusion among elements in the skeleton. In contrast to these juvenile histological features in IGM 100/3186, Stein, Hayashi & Sander (2013) reported moderate to extensive bone remodelling in the radii of three small ankylosaurs (two indeterminate specimens and one specimen of Edmontonia). This remodelling was present in specimens that were supposedly only 40 72% adult size, although these observations were based on bone cores instead of full sections, and it is unknown how representative each core is of the entire diaphyseal histology. The bone histology of the osteoderm of IGM 100/ 3186 (Fig. 7) is similar to that reported for a fragmentary osteoderm of P. grangeri (ZPAL MgD-II/27A) by Scheyer & Sander (2004), in that both share a spongy central region surrounded by a more dense cortex that is high in vascularity. Vascular canals are almost exclusively aligned with the long and intermediate axes, rather than the short (i.e. internal external) axis of the osteoderm. Growth lines are not evident in the section. There is no evidence of secondary remodelling, which is common in the other sectioned Pinacosaurus osteoderm (Scheyer & Sander, 2004), but that region of the thin section of IGM 100/3186 is poorly preserved in general. The combined skeletochronology data show that this individual was at least 1 year old at death. The dearth of growth lines in this specimen precludes the more precise estimates of longevity or growth rates available for other sampled ankylosaurs (Stein, et al., 2013). TAPHONOMIC CONSIDERATIONS The discovery of a well-preserved, ossified hyobranchial apparatus in P. grangeri suggests that this structure may be more widespread than previously suspected among fossil dinosaurs. Nevertheless, the holotype specimen of P. grangeri does not include elements of the hyobranchial apparatus, nor do the many referred specimens (Hill et al., 2003; Burns et al., 2011). There are several potential reasons for the scarcity of these elements. First, there is a significant preservation bias against cartilaginous portions of the hyobranchial apparatus. If the structure is not fully ossified, even in adults, it has a reduced chance of being preserved as a fossil. Even when ossified or calcified, there is a mechanical preservation bias against the hyobranchial apparatus, which consists of thin, delicate bones. In IGM 100/ 3186, the rostralmost projection of the paraglossal is less than 0.5 mm in thickness so thin as to be translucent, revealing the reddish sediment underneath it. To be completely preserved, as in this specimen, would require rapid burial in a very low-energy palaeoenvironment.

8 DINOSAUR HYOBRANCHIAL APPARATUS 899 Figure 6. Pinacosaurus grangeri (IGM 100/3186). A, associated left radius (ra) and fragment of ulna (ul); B, transverse thin section of the radius showing abundant longitudinally oriented vascular canals, crenulated periosteal margin, and the absence of growth lines. Tick marks indicate level of histological section. Dashed lines indicate missing bone. Second, even if bony hyobranchia are preserved, they may not be immediately recognized in the field or prep lab, because of the widespread unpredictability of, and unfamiliarity with, this complex structure. Bits of shattered bone that obscure the important data of the jaw Figure 7. Pinacosaurus grangeri (IGM 100/3186). A, B, associated osteoderm in superficial (A) and deep (B) views; C, transverse thin section of osteoderm showing abundant transverse and longitudinal vascular canals. Tick marks indicate level of histological section. Dashed lines indicate missing bone. Abbreviation: fm, foramen. and palate may be routinely swept away without recognition of their anatomical value as parts of the hyobranchial apparatus. Re-evaluation of known

9 900 R. V. HILL ET AL. specimens of ankylosaurs and other dinosaurs may reveal that they actually do preserve hyobranchial elements. One notable example worth reassessing is the unidentifiable bone described in the holotype skull of Gargoyleosaurus parkpinorum (Kilbourne & Carpenter, 2005). DISCUSSION COMPARATIVE MORPHOLOGY The specimen described here provides new information about the hyobranchium in ankylosaurs, and in dinosaurs in general. Unlike the specimen of Saichania (Maryańska, 1977), and indeed any extinct dinosaur, this specimen reveals a hyobranchium composed of multiple paired elements, indicative of the bilateral contributions to this complex structure. We reinterpret the curved rod-like elements of Saichania as second rather than first ceratobranchials, and the median, triangular element of that species as fused paraglossalia, rather than a combined basihyal/basibranchial, as previously suggested (Maryańska, 1977). The fused condition of the median, triangular element in Saichania, a presumably adult individual much larger than IGM 100/ 3186, may represent the end of an ontogenetic trajectory for ankylosaurs. Furthermore, the complex hyobranchium of Pinacosaurus suggests that the single pair of curved, rod-like hyobranchial bones found in many other dinosaurs often represent second rather than first ceratobranchials, or other elements, as previously interpreted. We surveyed the literature and the collections of the Yale Peabody Museum and American Museum of Natural History for evidence of other hyobranchial elements in ankylosaurs. One specimen of the ankylosaurid Euoplocephalus tutus Lambe, 1902 (AMNH FARB 5405) preserves a slender rod in the right cranioquadrate space. Its size and curvature are consistent with the novel interpretation that this is a second ceratobranchial that was dislodged after death and preserved alongside the braincase. A specimen of the nodosaurid ankylosaur Edmontonia rugosidens Gilmore, 1930 (AMNH FARB 5381) possesses two long, sinuous ceratobranchial rods, the rostral ends of which are concealed within a complex, folded sheet of gular skin, with hundreds of embedded ossicles (Vickaryous, 2006). CT scans (also available via the Morphobank project cited above) show a thin, triangular sheet of bone lying just deep to the gular armour, between the rostralmost tips of the ceratobranchials (Fig. 8). In location, shape, and relative dimensions, it resembles the paraglossalia of Pinacosaurus and Saichania, suggesting that the presence of these elements is probably primitive for Ankylosauria. Among extant taxa, the closest approximation of the paraglossal morphology observed here is found in birds. Both palaeognathous and neognathous birds possess paraglossal elements that originate as paired cartilaginous anlagen that may unite and/or ossify during ontogeny (Kallius, 1905; Fürbringer, 1922; Tomlinson, 2000). The composite paraglossal element is an arrowhead-shaped bone or cartilage with pointed caudolateral processes embedded in the base of the tongue. The presence of ossified paraglossalia in the new specimen is remarkable, because these elements have not been recognized previously in any non-avian dinosaur. In form and location, the bones together closely resemble the paraglossal bones of extant (Homberger & Meyers, 1989; Tomlinson, 2000) and fossil (Bertelli et al., 2010) neognath birds; however, there are several differences observed in the new specimen. First, instead of a single, midline ossification, the new specimen preserves paired, bilateral paraglossalia. Given the young age of this individual, the midline suture between the paraglossalia may not yet have fused. A second difference observed in the new specimen is the large size of each paraglossalium: over 50% of the lower jaw length. This is relatively much larger than in birds, in which the paraglossal bone is small relative to the oral cavity (Homberger, 1986; Tomlinson, 2000). Third, whereas the paraglossals of birds are generally flat and smooth both dorsally and ventrally, IGM 100/3186 preserves a suite of anatomical landmarks indicating extensive muscle attachment in P. grangeri. Other reptiles are not known to possess paraglossal bones; however, a long, pointed process in the rostral midline of the hyobranchium is found in many species of lizards and turtles (Fürbringer, 1922). Typically termed the entoglossal or lingual process, this structure derives from the hyoid (second visceral) arch, and is not likely to be strictly homologous with the paraglossal, owing to the presence of both the basihyal and the paraglossalia in birds. In crocodylians, the hyobranchial body is invariably composed of cartilage, with a square rostral margin. It is unclear whether this structure is homologous to the basihyal, paraglossal, or other combinations of elements. Caudal to the paraglossal elements of IGM 100/ 3186, the skeletal derivatives of the third and fourth visceral arches differ from those found in any other archosaurs. The flat, spatulate first ceratobranchials of P. grangeri project caudolaterally, and are slightly concave caudally. The second pair of ceratobranchials is concave rostrally, creating a lens-shaped interval between the elements. The first and second ceratobranchials meet at their distal tips, both articulating with a single epibranchial on each side (Fig. 4). This combination of features (broad first ceratobranchial, concave rostral curvature of second ceratobranchial, and single epibranchial articulating with both first and second ceratobranchial elements)

10 DINOSAUR HYOBRANCHIAL APPARATUS 901 Figure 8. Edmontonia rugosidens, AMNH FARB Top left: skull, mandible, and buccal osteoderm in oblique left lateral view. Top right: stereophotographs of head in ventral (palatal) view, showing gular ossicles and long ceratobranchials. Bottom: three orthogonal CT slices (not to scale) through region of hyobranchial apparatus. Lines show approximate level of slices. Location of paraglossalia outlined in red, dotted line. cb, second ceratobranchial; de, dentary; gost, gular osteoderms; ha, hyobranchial apparatus; oc, occipital condyle; t, teeth. characterizes certain salamandroid salamanders, such as Ambystoma, Dicamptodon, and Taricha. These salamanders feed terrestrially, using tongue prehension to pick up prey items. Some use a projectile means of propelling the tongue towards a food item, whereas others use the tongue as an adhesive pad to form to the contours of food items and draw them into the mouth (Wake & Deban, 2000). SOFT-TISSUE ASSOCIATIONS: MUSCLE RECONSTRUCTION Several anatomical studies provide a rich context for the interpretation of muscle attachment in Pinacosaurus (Fig. 9). These include studies of palaeognathous birds (Tomlinson, 2000) neognathous birds (Zweers, 1982; Homberger, 1986; Homberger & Meyers, 1989), and broader sauropsid clades (Gnanamuthu, 1937; Tanner & Avery, 1982; Bona & Desojo, 2011). The muscles that attach to the bones or cartilages of the hyobranchial apparatus are collectively called the hyolingual musculature. They can be organized into at least three categories, all of which were probably present in Pinacosaurus: intrinsic hyolingual muscles, extrinsic hyolingual protractors, and extrinsic hyolingual retractors. Intrinsic hyolingual muscles would have connected bones or cartilage of the hyobranchial apparatus to one another, or to the tongue. These would include the paired ceratoglossus, which in birds extends from the first ceratobranchial to the paraglossal. Any intrinsic tongue musculature is likely to have originated from the dorsal surface of the paraglossal. In salamanders that share the ceratobranchial morphology described here, the first and second ceratobranchials are adducted by a muscle that connects them: the subarcualis rectus. The serial homologues of this muscle adduct additional arches, when present. Some extant lizards have a connective tissue sheet between the first

11 902 R. V. HILL ET AL. Figure 9. Schematic reconstruction of hyolingual musculature of Pinacosaurus grangeri. Ventral view, showing hypothesized lines of action of hyolingual musculature. Blue arrows: intrinsic hyolingual musculature, e.g. hyoglossus, ceratohyoideus. Yellow arrows: extrinsic hyolingual protractors, e.g. mandibulohyoideus, branchiomandibularis. Red arrows: extrinsic hyolingual retractors, e.g. sternohyoideus. and second ceratobranchials (Tanner & Avery, 1982). Others lack this connection; however, most lizards possess a branchiohyoideus muscle between the first ceratobranchial and ceratohyal elements (Herrel et al., 2005; Meyers, Herrel & Nishikawa, 2002). It remains ambiguous as to which, if any, of these structures was represented in Pinacosaurus. A second set of hyolingual musculature comprises the extrinsic hyolingual protractors that draw the hyobranchium and tongue rostrally. In extant archosaurs, the branchiomandibularis muscle originates from the caudal end of the mandible and inserts onto the first ceratobranchial, drawing the entire hyobranchium rostrally when it contracts. It is present in both avians (Zweers, 1982; Homberger, 1986; Homberger & Meyers, 1989; Tomlinson, 2000) and crocodylians (Bona & Desojo, 2011), suggesting it is conserved even in highly derived archosaurs, and was probably present in Pinacosaurus. Other extrinsic hyolingual protractors include muscles that originate closer to the mandibular symphysis and insert on the hyobranchium. The homologies of these muscles among archosaurs are disputed; however, genioglossus is recognized as a medially placed protractor of the hyobranchium in both avians (Tomlinson, 2000) and crocodilians (Bona & Desojo, 2011). Palaeognaths also possess a more laterally placed protractor, the genioceratohyoideus, which originates from the lingual side of the mandible and inserts on the first ceratobranchial. This may be homologous with the lateral portion of genioglossus in crocodiles. In many lizards, the major hyolingual protractor is the mandibulohyoideus, with genioglossus representing a dedicated protractor of the tongue itself (Herrel, Cleuren & De Vree, 1995; Herrel, Cleuren & De Vree, 1997; Herrel, Timmermans & De Vree, 1998). The third set of hyolingual muscles comprises the extrinsic hyolingual retractors. These originate from non-hyobranchial bones, and run rostrally or rostromedially to insert on bones of the hyobranchium. Their action is to draw the protracted hyobranchium caudally, and the tongue along with it, pulling it back into the mouth. Birds and other sauropsids possess a constrictor colli intermandibularis (also found in other reptiles), and birds also possess specialized retractors (serpihyoideus in all avians and stylohyoideus in neognaths) that originate from the mandible and extend rostromedially to insert upon the basihyal or urohyal elements of the hyobranchium (Tomlinson, 2000). Other tetrapods generally possess extrinsic hyolingual retractors that originate from the sternum or pectoral elements. These include the rectus cervicis in salamanders (Deban & Wake, 2000), and the sternohyoideus, coracohyoideus, and episternobranchialis in crocodylians (Cleuren & De Vree, 2000; Bona & Desojo, 2011). Pinacosaurus probably retained a sternopectoral origin for its extrinsic hyolingual retractors. Like several other ankylosaurs, it possessed robust, ossified sternal elements with areas for muscle attachment consistent with the origin of a sternohyoideus-like muscle (Maleev, 1954; Coombs & Maryanska, 1990). Additionally, the large size of its hyobranchium relative to the oral cavity suggests that retractors with a mandibular origin would have had little mechanical advantage. A mandibular attachment for hyolingual retractors effectively allows the hyobranchium to become anatomically decoupled from the sternum and pectoral girdle. In the lineage leading to birds, this feature may have liberated the sternum to adapt to the demands of flight. The absence of a muscular attachment between the hyobranchium and sternum may also have facilitated the elongation of the neck common in saurischian dinosaurs (Taylor & Wedel, 2013), and may characterize Saurischia in general. FEEDING IN ANKYLOSAURS Ankylosaurs are traditionally interpreted as herbivores that fed predominantly on soft plants (Carpenter, 2012). Most analyses of ankylosaur feeding have focused on jaw mechanics and/or tooth morphology (Haas, 1969; Weishampel & Norman, 1989; Rybczynski & Vickaryous, 2001; Nabavizadeh, 2011). Recent analyses (Mallon & Anderson, 2013, 2014), however, have illuminated a great deal of variation in mandibular morphology, tooth size, and wear patterns among ankylosaurs,

12 DINOSAUR HYOBRANCHIAL APPARATUS 903 suggesting a more varied diet than previously proposed. Ankylosaurs in general were probably not well adapted to eating fibrous or woody plant matter, but may have eaten tough leaves as well as pulpy plant tissues, such as fruits (Mallon & Anderson, 2014). Fossilized gut contents of the basally branching ankylosaur Minmi preserve vascular segments, seeds, and fruits, supporting the hypothesis that at least some ankylosaurs were herbivores and ate multiple parts of plants (Molnar & Clifford, 2000). As a result of the scarcity of fossil hyobranchia, studies of feeding in dinosaurs have essentially ignored the role of the hyobranchium. The robust hyobranchium of Saichania has always been interpreted as providing attachment for a muscular tongue that may have aided in cropping vegetation (Maryańska, 1977; Carpenter, 2012; Mallon & Anderson, 2013). At least two lines of evidence from IGM 100/3186 suggest that P. grangeri also had a muscular tongue. First, dental histology reveals that Pinacosaurus had slowly replacing teeth with thin enamel (Fig. 5). This suggests less reliance on dental processing than in other ornithischians. Second, morphological features of the hyobranchium, including deeply excavated fossae and muscle scars on both dorsal and ventral surfaces, are consistent with a muscular tongue. Although specific behaviours (e.g. tongue protrusion, prehension) remain equivocal, it is clear that lingual processing was a key component of feeding in P. grangeri and other ankylosaurs. Muscular tongues, however, are not unique to herbivores. Many extant taxa, such as salamanders, chameleons, pangolins, armadillos, and anteaters, use their protrusible and muscular tongues to feed on insects (Reiss, 2000; Schwenk, 2000a). An insectivorous diet was proposed for ankylosaurs by Nopcsa (1928), who suggested that the North American ankylosaurid Scolosaurus may have subsisted on as little as 1 kg of insects per day, as an adaptation to a presumably desert environment. Consistent with this interpretation, the distinctive facies at Ukhaa Tolgod collectively indicate a relatively arid environment (Dingus et al., 2008). Dunes accumulated during particularly dry periods, catastrophically collapsing during rains and burying vertebrates alive. All facies at Ukhaa Tolgod indicate a sparsely vegetated environment, and there is evidence of pupal chambers of beetles (Johnston, Eberth & Anderson, 1996) and other arthropod activity (Kirkland & Bader, 2010). Millipede fossils are frequently found still in their burrows (M.A. Norell, pers. observ.) at Ukhaa Tolgod. Insects are numerous, speciose, and constitute a large percentage of the terrestrial biomass on Earth (May, 1988). It is therefore possible that an ankylosaur of relatively small body size, such as Pinacosaurus, could have subsisted on a diet partially or even primarily composed of insects. HOMOLOGY OF HYOBRANCHIAL ELEMENTS The completeness of IGM 100/3186 allows for a reevaluation of the homology of other fossil dinosaur hyobranchial elements, which almost always consist of a single pair of rods that vary slightly in their curvature and roundness in cross section. These elements are most often identified as first ceratobranchials, with exceptions including a putative ceratohyal or basihyal in Triceratops (Morschhauser & Lamanna, 2013), ceratohyals or ceratobranchial II in Protohadros (Head, 1998), ceratohyals in Psittacosaurus (Sereno, 2010), the presence of a midline corpus element in Carnotaurus (Bonaparte, Novas & Coria, 1990), and thyrohyals in Camarasaurus (Gilmore, 1925). These exceptions aside, the common identification of the single pair of rods in most taxa as first ceratobranchials is often based on the fact that this element tends to ossify with the greatest frequency in extant reptiles (Fürbringer, 1922), rather than on strict comparative anatomical grounds. Some curved, rodlike elements of other Mesozoic dinosaurs identified as first ceratobranchials [e.g. Corythosaurus (Ostrom, 1961), Anchisaurus (Galton, 1976), Ouranosaurus (Taquet, 1976), Iguanodon (Norman, 1980), Omeisaurus (Dong, Zhou & Zhang, 1983), Brachylophosaurus (Prieto-Marquez, 2005), Europasaurus, (Marpmann et al., 2015), Tethyshadros (Della Vecchia, 2009), Hypsilophodon, (Galton, 2009), Jinzhousaurus (Barrett et al., 2009), Parasaurolophus (Farke et al., 2013)], ceratohyals [e.g. Khaan (Balanoff & Norell, 2012)], thyrohyals [e.g. Camarasaurus (Gilmore, 1925)], or simply as hyoid bones [e.g. Citipati (Clark, Norell & Rowe, 2002), Zhuchengceratops (Xu et al., 2010), Hungarosaurus (Ősi, 2005)] more closely resemble the second than the first ceratobranchials or other elements of Pinacosaurus (or resemble both ceratobranchials in different ways), suggesting that these elements may be more or equally parsimoniously interpreted as second ceratobranchials in those taxa. Indeed, the bone identified as a second ceratobranchial in the ornithopod Protohadros (Head, 1998) most closely resembles the second ceratobranchial of the new specimen of Pinacosaurus in its sinuosity, the presence of an oval facet on one end, and the relative flattening on the other end. Under the reinterpretation of these elements as second rather than first ceratobranchials, the dorsoventrally expanded rostral end of ceratobranchial II in advanced ornithopods would be derived relative to the more unexpanded shape present in more primitive ornithopods (e.g. Iguanodon and Ouranosaurus), a hypothesis that awaits testing via more descriptive work in the clade. A hyoid element

13 904 R. V. HILL ET AL. attributed to Triceratops (Lull, 1933) is curved in two planes, similar to the second ceratobranchial found in the new specimen. Its flared caudal end and pointed rostral tip are reminiscent, however, of the other elements seen in Pinacosaurus. THE ORIGIN OF THE AVIAN PARAGLOSSAL The paraglossal has long been considered a neomorphic bone and an autapomorphy of neognathous birds (Kallius, 1905; Crompton, 1953; Tomlinson, 2000). In all extant birds, two cartilaginous anlagen (the paraglossalia) give rise to the adult paraglossal elements. In most birds the paraglossalia migrate towards one another during development, uniting to form a midline element that may later ossify. In the ostrich, the paraglossalia never meet, and instead remain as separate cartilage into adulthood. In other palaeognaths the paraglossalia unite but remain cartilaginous, and may leave an open fenestra in the midline (Crole & Soley, 2012). In most palaeognaths the paraglossal has a similar arrowhead-shaped morphology to that described above (Tomlinson, 2000; Crole & Soley, 2009); however, in certain tinamous the paraglossal is triangular with scalloped margins (Tomlinson, 2000). Neognath paraglossals also fuse from paraglossalia that unite and ossify late in the development of the hyobranchium (Kallius, 1905; Hogg, 1980; Tomlinson, 2000; Maxwell & Harrison, 2008; Genbrugge et al., 2011). In certain parrots the element is bipartite, with a median syndesmosis separating the two paraglossal bones (Homberger, 1986). Most other neognaths typically possess a single bone that bears no outward traces of its bipartite origin. We propose that the paraglossal bones described here for Pinacosaurus and Edmontonia are likely to have arisen from a similar developmental process. The presence of bilaterally symmetrical bones, sutured at the midline, supports a bipartite origin for the single paraglossal found in Saichania. The paraglossalia of P. grangeri are unusual in that, as preserved, they lie ventral to the first ceratobranchial elements (and presumably, the midline, cartilaginous basihyal). In most birds, the paraglossal articulates with the tip of the basihyal dorsally. There are a few possible explanations for this dorsoventral inversion in the expected relationship. First, although the hyobranchium is preserved largely in situ, there has been some taphonomic disturbance of the specimen, as evidenced by the disarticulated braincase elements and the predentary, as well as several scattered teeth. Second, the large size of the paraglossalia, relatively much larger than in any bird, may constrain the development to the more ventral location, also compensating for increased muscle attachment to their dorsal surfaces. Third, this condition may simply reflect the already recognized variability in the system, as in Struthio, the unfused paraglossalia also lie ventrolateral to the basihyal. This configuration leads to some slight peculiarities of muscle attachment in Struthio, including the absence of hyoglossus muscles (Tomlinson, 2000). The most parsimonious explanation for the presence of ossified paraglossalia in both Pinacosaurus and neogath birds is that their cartilaginous anlagen were present in the common ancestor of all dinosaurs (Fig. 10). These structures may have remained cartilaginous throughout life in other ornithischians and saurischians, or may have been lost secondarily in certain clades. Alternatively, ossified paraglossalia may be more common, and simply not recognized in the fossil record, because of their fragility and unfamiliar shape. Paraglossalia may even further characterize birdline archosaurs; however, thus far the fossil record of proximate dinosaur out-groups (e.g. pterosaurs) has yielded only rod-like ceratobranchials (Romer, 1956; Padian, 2008). Referring to the paraglossal as simply a neomorph avoids the question of its deeper evolutionary and developmental identity. One hypothesis proposed for the origin of the avian paraglossal is that it represents the paired ceratohyal elements; i.e. the lateral elements derived from the hyoid arch (de Beer & Barrington, 1934). These structures are large and wing-like in a variety of vertebrates (Tanner & Avery, 1982; Grande & Bemis, 1988; Deban & Wake, 2000), and in salamanders (as described above) they closely mirror the location, morphology and relative size of the paraglossal bones of P. grangeri, except in that they are typically positioned dorsal to the basihyal (Deban & Wake, 2000). Nevertheless, Crompton (1953) rejected a ceratohyal origin for the paraglossal, based on his work with early penguin embryos, because the paraglossal anlagen lie rostral to the basihyal and presumably the other second arch structures. The question of the developmental identity of the paraglossal will only be fully addressed with comprehensive evolutionary/developmental studies that take into account growth series of birds, crocodiles, and other tetrapods. Although they generally have low preservation potential, fossil hyobranchia stand to greatly enhance the understanding of the evolution of the vertebrate head and neck. Our description refines the search image for future palaeontological fieldworkers and preparators. Future research will also illuminate the homologies of other elements of the hyobranchial apparatus, which remain controversial (Reilly & Lauder, 1988; Wake & Deban, 2000). CONCLUSIONS 1. Pinacosaurus grangeri had a well-ossified hyobranchial apparatus, with at least four paired