The first oviraptorosaur (Dinosauria: Theropoda) bonebed: evidence of gregarious behaviour in a maniraptoran theropod

A monodominant bonebed of Avimimus from the Nemegt Formation of Mongolia is the first oviraptorosaur bonebed described and the only recorded maniraptoran bonebed from the Late Cretaceous. Cranial elements recovered from the bonebed provide insights on the anatomy of the facial region, which was formerly unknown in Avimimus. Both adult and subadult material was recovered from the bonebed, but small juveniles are underrepresented. The taphonomic and sedimentological evidence suggests that the Avimimus bonebed represents a perimortem gregarious assemblage. The near absence of juveniles in the bonebed may be evidence of a transient age-segregated herd or ‘flock’, but the behaviour responsible for this assemblage is unclear. Regardless, the Avimimus bonebed is the first evidence of gregarious behaviour in oviraptorosaurs, and highlights a potential trend of increasing gregariousness in dinosaurs towards the end of the Mesozoic.

Institute of Paleontology and Geology, Mongolian Academy of Sciences (MPC-D) in Ulaanbaatar, Mongolia. The assemblage is the first reported record of a monodominant bonebed of oviraptorosaurs-as well as the first Late Cretaceous maniraptoran bonebed-and provides insights on the anatomy and behaviour of Avimimus.

Results
Geological and Sedimentological Context. The Avimimus bonebed occurs in the lower portion of the alluvial Nemegt Formation, at the Nemegt locality within the Nemegt Basin 32,33 (Fig. 1). Here, Upper Cretaceous strata of the Nemegt Formation interfinger with and overlie the Baruungoyot Formation 33 . In general the Nemegt Formation is characterized by abundant deposits of ephemerally active meandering channels, splays and sheetfloods, and ponds and wetlands 32,33 . The Nemegt Formation has yielded few bonebed deposits; the only two recorded instances include an assemblage of Saurolophus 34 and this Avimimus bonebed. The precise age of the Nemegt Formation is difficult to determine because of the discontinuity of beds and exposures, absence of microfossil biostratigraphy, and lack of datable volcanics 33 . Nonetheless, based on the better-documented late Campanian to early Maastrichtian age of the underlying Djadokhta Formation 35 and the presence of Maastrichtian dinosaurs such as Saurolophus in the Nemegt Formation, an early Maastrichtian age for the Nemegt Formation is currently accepted.
The exposed Nemegt Formation around the bonebed is only 35 m thick due to truncation by a regionally-expressed Quaternary unconformity. The Avimimus bonebed occurs 10.5 m above an interfingered Nemegt-Baruungoyot contact and is associated with the lower portions of sigmoidal and offlapping inclined beds of silty, pebbly, fine-to medium-grained sandstone 33 (Fig. 2). Two types of matrix surround the bonebed. The base of the bonebed is a fine-grained sandstone, which is overlain by a coarse-grained sandstone with some clay rip-up clasts. Inclined beds are typically less than 5 cm thick, exhibit a total vertical relief of 40 cm, and dip toward the south and southwest, suggesting that they are offlapping deposits of a migrating point bar in a meandering river channel 32,33 . Large-scale trough cross beds drape the toes of the point bar and immediately overlie the bonebed. Paleocurrent data collected from them indicate that flow at the base of the point-bar was toward the west-southwest, ranging from 240-280°. Bones are preserved in both matrix types, and some bones span the contact between the layers. This suggests that the beds were deposited in a single coarsening-upwards event, probably tied to the migration of the point bar. The presence of localized mudstone pebbles and Avimimus remains at the base of the beds, as well as non-predictable grain size changes between beds, poorly organized mixtures of trough and ripple cross-strata within beds, soft-sediment deformation structures, and localized millimeter-thick clay drapes all suggest highly variable conditions through time at the bonebed site. These conditions include erosive hydraulic flow, standing water, subaerial exposure, and trampling by dinosaurs. Accordingly, at times this channel-hosted site may also have acted as a waterhole, attracting vertebrates in search of food and/or water.
Assemblage. The bonebed assemblage is dominated by Avimimus, which comprises 160 (90.4%) of the 177 accessioned specimens (Table S1). The other 17 accessioned specimens include indeterminate dinosaur elements (6), an oviraptorid dorsal centrum, an oviraptorid ilium, embryonic hadrosaur bones, a bird tarsometatarsus, a lizard vertebra, a mammalian limb bone, eggshell, gastropod and bivalve casts, and wood. The relative abundance and dominance of Avimimus in the bonebed is underestimated because, in some cases, multiple elements are preserved in articulation or as fused single functional elements that were accessioned together. The non-poached, newly excavated part of the bonebed was collected exhaustively, so it is unlikely that there was a collection bias towards Avimimus at the expense of other taxa. Avimimus material collected includes a variety of cranial elements, vertebrae, forelimb material, some parts of the pelvis, and many hindlimb elements. Ontogenetic stage of elements from the bonebed was assessed using size and fusion of the elements. Although these qualities are not strictly tied to developmental age, similar criteria have been used previously to assess relative age in other assemblages where histological sections were unavailable 8,36 . Adults were identified by fusion of the tibiotarsus or tarsometatarsus. The lengths of fused tibiatarsi (n = 17) from the bonebed vary by less than 10% (246 mm -280 mm), suggesting that adults were either all of a similar ontogenetic stage, or that growth was determinate in adult individuals. All tibiotarsi longer than 246 mm (n = 17) are fused, and none shorter than 246 mm are fused, indicating that fusion of the tibiotarsus is strongly tied to body size. Subadult individuals were identified by lack of hindlimb fusion but proximity in size to the largest fused individuals (tibiae > 80% of 280 mm). Of the 33 tibiae recovered from the bonebed (Table 1), only one tibia (MPC-D 102/38), an estimated 202 mm in length, fell below the 80% cutoff of the length of the largest tibia (280 mm), and can be considered a juvenile individual. The state of fusion for this individual cannot be assessed, because the distal end is missing. The presence in the assemblage of small elements from Avimimus, such as phalanges, and small material from other taxa, including embryonic hadrosaur bones, a lizard vertebra, and mammal limb bones, suggests that the dearth of juvenile Avimimus is real, rather than the result of winnowing. Evidence from other bonebeds suggests that non-avian theropod dinosaurs had a tendency to form juvenile-dominated herds 17,[36][37][38] . Adult-dominated bonebeds of non-theropod taxa also typically contain a small percentage of juvenile animals 8,34 . The Avimimus bonebed is therefore unusual in its near absence of juvenile individuals.
The minimum number of individuals (MNI) represented by the assemblage was estimated using tibiae. The distal ends of 13 right tibiae are present and indicate that at least this many individuals contributed to the assemblage. However, combining the measurements of these 13 specimens with data from left tibiae ( Table 1) shows that there are at least 18 tibiae of different sizes in the assemblage. Thus, an MNI of 18 appears to be the best estimate for the number of individuals that contributed to the assemblage, which represents the largest monodominant assemblage of maniraptorans yet reported.
Taphonomy. Given an MNI of 18, overall skeletal element representation of Avimimus is low in the bonebed (~4-5%, Table S2). Hindlimb elements are strongly overrepresented (~30-94%) in the assemblage compared to all other elements (Fig. S1). Long bones excavated from the quarry (Fig. 3) show a significantly preferred NE-SW orientation (Rao t = 201, p < 0.001; Raleigh Z: t = 0.7211, p < 0.001), subparallel to sedimentological indications of a SW-oriented paleochannel. All Avimimus bones recovered from the site share the same bone modification signature, lacking signs of prolonged subaerial exposure, insect feeding traces, tooth marks, or weathering. Small fragments of hadrosaur bones in the bonebed are typically abraded, weathered, and powdery, suggesting they have been subject to different taphonomic conditions. Most of the bones were preserved horizontally, but in the coarse-grained sandstone layer several elements (ilium, humerus, phalanges) were vertical or inclined, suggesting that they were moved from resting position and buried quickly. Few of the unfused elements are preserved in articulation, with the notable exceptions of two nearly articulated premaxillae (Fig. 4J,K) and eight distal caudals ( Fig. S2) found articulated. Most of the associated material comes from compound or fused elements, but the presence of some associated material (ilium and unfused sacral ribs) indicates that the bones were not transported far. The bones have been hydraulically sorted, so that small elements are rare-but still present-and there is a bias towards the preservation of thick-walled elements like femora and tibiae. Despite this, the bones show little to no abrasion, and delicate elements, like cranial material and fibulae (Fig. S3), are unbroken, which suggests that hydraulic flows that sorted or reoriented elements were not intense or prolonged. Most of the broken bones from the assemblage were surface collected, suggesting that they were damaged by the poachers. Only two theropod teeth (cf. Velociraptor) were recovered from the site, suggesting that scavenging, if present at all, was limited. It is unlikely for a number of reasons that the sediments surrounding the bonebed represent the first burial of the material. First, such an assemblage would be dominated by articulated or associated material, rather than isolated bones. Second, the matrix is representative of normal deposition in a channel, rather than a catastrophic flood capable of killing multiple Avimimus. The pristine condition of the bones suggests they were originally buried rapidly, which protected them from subaerial weathering, trampling and scavenging, but allowed most of the soft tissues to decompose over the course of months to years. The bonebed was then uncovered by a medium-to high-energy flow, represented by the two sandstones, which disarticulated most of the material and transported it a short distance. The second flow event had enough energy to reorient most of the long bones to a N-NE to S-SW orientation (Fig. 3), but was not powerful enough to remove large elements. Numerous examples of similar multistage depositional events in monodominant assemblages are known from North America 10,39,40 .
Craniomandibular Skeleton. The bonebed produced cranial elements that were formerly unknown for Avimimus and provide important anatomical information. A jumble of associated bones (MPC-D 102/34) includes the premaxillae and nasals (Fig. 4), and an additional two semiarticulated premaxillae (MPC-D 102/108) were recovered (Fig. 4J,K). The unfused premaxillae are hollow; each has a long dorsal process with a lateral facet for the nasal and a flat medial surface for the adjoining premaxilla. The tomial margin has five denticulations (Fig. 4J). Uniquely amongst oviraptorosaurs, the laterally flaring posterior process (Fig. 4F,G) that separates the maxilla from the external naris has a deep depression, probably confluent with the antorbital fossa. Although Kurzanov 20 reconstructed Avimimus with a conjoined naris and antorbital fenestra, the presence of the posterior process, which is missing in the holotype, indicates that it would have separated the naris and antorbital fenestra, as in all other theropods. The fused nasals (MPC-D 102/46; Fig. 4) form an unusual anchor-shaped bone. Posteriorly the nasals have ventrolaterally extending, hatchet-shaped lateral descending processes. Anteriorly there is a longitudinal groove on the dorsal surface, which opens into a slot for the premaxillae (Fig. 4). The posterior margin of the fused nasals is concave in dorsal view, and the nasals would have been largely separated posteriorly by the frontals. There is a longitudinal ridge on the ventral side of the midline process of the nasals. The united nasals have a smooth exterior surface and lack the pneumatic pitting that is present in oviraptorids [41][42][43] .
Rearticulating the premaxillae and nasals shows that snout was short, with a vertical anterior margin and anteriorly facing nares. A partial skull (MPC-D 102/81) preserves the posterior part of the cranium, which has coossified into a single unit-here called the neurocranial unit-as in birds. The coossified unit of Avimimus incorporates more bones than that of any bird, including the frontals, parietals, postorbitals, pterygoids, quadrates, squamosals, and bones of the braincase. Sutures between bones are obliterated, except for faint lines between the opisthotic-exoccipital unit and the basioccipital. The body of the apneumatic quadrate (Fig. 5) fuses along its whole medial margin to the prootic and pterygoid, so that the only communication of the post-temporal fenestra with the region anterior to the quadrate is the foramen for the middle cerebral vein (Fig. 5C,D). Dorsally, the pterygoid wing of the quadrate and the fused squamosal are separated from the exoccipital by an anteriorly-facing recess with a large ventral foramen and a smaller dorsal fossa, both for the middle cerebral vein. An aperture in the dorsal part of  44 . The right quadratojugal is an anteriorly directed prong that is indistinguishably fused to the lateral margin of the quadrate. There is no evidence of a quadratic foramen or fenestra between the quadrate and quadratojugal, a feature that is usually present in oviraptorosaurs 41,42 . The occipital process of the squamosal (Fig. 5) is conjoined and fused to the paroccipital process of the exoccipital, which is unusual for oviraptorosaurs. The posterior part of the pterygoid (Fig. 5) is relatively large and is horizontal, contrasting with the typical dorsomedial-ventrolateral orientation of the oviraptorid pterygoid. The pterygoid contacts the basisphenoid along most of its length, rather than just at the basipterygoid process. The pterygoid contact with the quadrate is anteroposteriorly extensive and lies far dorsal to the mandibular condyles of the quadrate. In oviraptorids, the pterygoid typically contacts the quadrate just medial to the mandibular condyle 41 . The broken pterygoid ramus is hollow, unique for oviraptorosaurs (Fig. 5). The occipital condyle (Fig. 5) is kidney-shaped and smaller than the foramen magnum, which is nearly circular as in oviraptorids 43 . The basal tubera are large and separated by a shallow median depression, with a possibly pneumatic foramen at its center. There are no basisphenoid recesses. The basipterygoid processes face laterally and are continuous with the greatly expanded posterior wing of the pterygoid, unlike in oviraptorids 41 . The supraoccipital has a longitudinal sagittal crest, but lacks a transverse nuchal crest. The opisthotic-exoccipitals form small laterally directed paroccipital processes that do not extend ventrally to the level of the basal tubera (Fig. 5). There is only one jugular opening on the posterior surface of each opisthotic-exoccipital-apparently   unique for a dinosaur-that served as the exit for cranial nerves IX-XII (Fig. 5). On the medial wall of the exoccipital portion of the braincase (Fig. 5J,K), there are five foramina. The largest of these, dorsal to the others, is a dorsoventrally oriented slit for cranial nerves IX-XI and communicates with the jugular opening on the posterior side of the exoccipital. Of the four smaller, ventral foramina, the anterior one is probably for a blood vessel, and the other three were for branches of cranial nerve XII. The last three foramina merge posteriorly to exit through the large jugular opening on the posterior face of the exoccipital. The medial surface of the braincase is pierced by a large floccular recess, ventral to which is a shallow depression pierced by cranial nerves VII and VIII (Figs 5b  and 7J,K). Anterolaterally, the prootic is pierced by a small anteriorly-facing foramen for cranial nerve VII.
The edentulous apneumatic dentaries (MPC-D 102/16) of Avimimus are partly coossified, although a suture is visible ventrally (Fig. 6). The lingual surface of the dentary has a complex series of ridges and grooves, although the relief is not as great as in caenagnathids 44 . There is a distinct lingual groove on the occlusal surface of the dentary, which is bounded medially by a weakly pronounced lingual ridge. There is an incipient symphysial shelf, similar in development to most oviraptorids 41 .The occlusal margin projects above the rest of the lingual surface, but is not concave in lateral view (Fig. 6). The Meckelian grooves are separated at the midline by a distinctive ventrally tapering buttress of bone (Fig. 6), which demarcates two shallow lateral fossae in posterior view. The lateral surface of the mandible is marked by several minute foramina, which suggests that there was a keratinous beak as in birds 44 . The posterodorsal ramus of the dentary is not bifurcated transversely, which indicates that its contact with the surangular or coronoid was simple, as in oviraptorids 41 .

Discussion
The predominance of thick-walled, hydrodynamically dense elements (Figs 7 and 8, Table S1) and the taphonomic signatures of the bones, combined with sedimentological and paleocurrent data, suggest that this assemblage represents a secondary deposit of previously buried skeletal material. The original death assemblage was probably formed by a catastrophic mass death event and the remains were then accumulated in the paleochannel during a second depositional event. The cause of the mass death cannot be determined with certainty, although the assemblage is somewhat similar to the ornithischian bonebeds from the Late Cretaceous of the Western Interior of North America. These include well-studied bonebeds such as the Centrosaurus bonebeds of Dinosaur Provincial Park 8 and the Pachyrhinosaurus bonebeds of the Wapiti Formation 16 . These assemblages consist primarily of single species, although there are often isolated elements from other taxa. Larger individuals dominate, and assemblages are preserved as disarticulated, hydraulically sorted elements in channel lag deposits. These bonebeds are thought to result from the catastrophic deaths of many individuals in groups or herds, drowned during flooding events 8,11,16 . The high proportion of distal hindlimb elements is unusual, especially considering that other dense elements, like sacra and the fused pelvic elements 27 , are underrepresented (Fig. S1). This may point to a miring event as the cause of the mass death. Unfortunately, any sedimentological indications of miring have been erased by the second flow event, and therefore the cause of death is ambiguous. The composition of the bonebed and the high number of individuals have implications for the behaviour of Avimimus. The death assemblage strongly suggests that Avimimus engaged in gregarious behaviour, although the particular nature of that behaviour is not clear. The morphology of the mandible (Fig. 6) in Avimimus is similar to oviraptorids and caenagnathids, which were probably herbivorous 44,45 , so it is unlikely that the bonebed is evidence of pack hunting or a scavenging-driven assemblage. The presence of more than two adults suggests that the bonebed is not an isolated family group, and the mix of subadults and adults in such a large aggregation argues against an assemblage of parents and their offspring. Other assemblages of multiple omnivorous or herbivorous theropods have been discovered, most notably therizinosaurs 46 and ornithomimids 17 . Kirkland et al. 46 studied a paucispecific bonebed of Falcarius, with more than 300 individuals and a range of developmental stages 47 . The abundance of material (> 2000 specimens) 47 , and the 99% dominance of Falcarius at the site 46 strongly suggest that the site is the result of a catastrophic mass death. The presence of multiple developmental stages indicates that the Falcarius assemblage reflects typical population structure, which suggests that it may represent a non-transient herd. This appears to be untrue of the Avimimus bonebed, which, despite a smaller sample, shows a strong bias in the presence of certain developmental stages. It is possible that the near absence of juveniles in the bonebed is the result of reproductive seasonality and rapid growth. In this case, young Avimimus are born at the same time of year and grow to near-adult size within a year. The result is a mixed-age assemblage of similar size. In the absence of histological data, however, this assertion cannot be tested, but the presence of one possibly juvenile individual (MPC-D 102/38) argues against it.
Although speculative, the paucity of juveniles in the bonebed may instead indicate that Avimimus formed age-segregated assemblages. These groups may have enjoyed reproductive, antipredator, or foraging benefits, but the contribution of these factors to the formation of the assemblages is unclear. Lekking behaviour, where individuals group to display to potential mates, is known in multiple groups 48,49 , especially birds. Aggregations may include as many as 100 individuals 50 of varying age, size, and sexual fitness. The near absence of juveniles is consistent with, but not indicative of, a lekking assemblage. The sex of any individual in the assemblage cannot be assessed, and therefore the ratio of genders cannot be used to evaluate the hypothesis of lekking. Alternatively, the bonebed assemblage may be evidence of flocking or communal roosting behaviour for any number of reasons. The anti-predator effectiveness of flocking and communal roosting 51 is documented especially well in birds [52][53][54] and mammals [55][56][57] , among other animals. Multiple studies show that vigilance 58 is reduced in larger groups, increasing foraging efficiency 59,60 . Kobayashi and Lu 17 described an assemblage of at least 14 articulated Sinornithomimus from China, with a high proportion of juveniles 37 , which they suggested is the result of a predator avoidance strategy. However, in the absence of a larger sample size and evidence of the cause of the mass death event, the specific behaviour that the Avimimus death assemblage represents cannot be assessed.
Monodominant and monotaxic associations of bones and skeletons are known for many taxa of dinosaurs (Table S3). The distribution of these bonebeds is demonstrably non-random with respect to phylogeny and stratigraphy, although the effects of outcrop availability and sampling are unclear. Certain animals, especially during the Cretaceous period, show a strong tendency to form monodominant or monotaxic assemblages 11 . These include the hadrosaurine hadrosaurs, especially Edmontosaurus 9 , and centrosaurine ceratopsids 8,15 . Oviraptorosaurs can be added to this list, as numerous associations of multiple oviraptorosaurs are known, including Anzu wyliei 61 , Conchoraptor gracilis (MPC-D 102/3; M. A. Norell pers. comm.), Heyuannia huangi 62 , and Khaan mckennai 63 . Although it could be argued that this is simply an artifact resulting from the association between certain lineages and environments that are subject to flooding events, such as coastal plains, the fact remains that many dinosaurs and other taxa that inhabit these environments do not occur in such assemblages. Furthermore, many of the associations, especially those of oviraptorosaurs (Conchoraptor, Khaan), are not from fluvial sediments. In any case, the Avimimus bonebed provides the first strong evidence of gregarious behaviour in oviraptorosaurs, and improves our knowledge of social behaviour in maniraptorans.

Methods
Collection. The bonebed was excavated systematically and mapped (Fig. 3) using a 1 m 2 grid system. Once mapped, specimens were collected and given coordinate numbers referring to their location within the map grid (see supplementary information). Plan view, bi-directional orientations (azimuthal trends) of elements were obtained from the map using ImageJ 1.48 v. These data were then plotted on a rose diagram divided into 45-degree quadrants (Fig. 3). Rao's Spacing and Raleigh Z tests from the R software package "circular" were used to assess whether the in situ assemblage exhibited statistically significant preferred orientations. Specimens were mechanically prepared at the MPC by "Dinosaurs of the Gobi" participants, using a combination of manual and air-pressured tools and a variety of consolidants (Butvar, cyanoacrylate).
Taphonomic analysis. Taphonomic bone-modification data 64 were assessed through simple visual inspection of prepared elements. The adult skeleton of Avimimus is characterized by the fusion of many bones into compound elements, reducing the number of discrete skeletal elements through ontogeny. As such, the mixture of adults and subadults in this bonebed makes it difficult to accurately assess skeletal element representation. Skeletal representation was therefore calculated separately for both an adult skeleton and a subadult skeleton with no fused elements (Table S2). This provides minimum and maximum boundaries, although the true pattern is likely towards the maximum boundary, because compound elements were more commonly fused than unfused in the bonebed. Minimum number of individuals was estimated using the maximum number of unique elements (distal right tibiae). Combining this MNI (n = 13) with length estimates of partial tibiae based on the proportions of complete tibiae gave a better estimate of the number of tibiae of substantially different lengths (n = 18).

Measurement.
Calipers were used to measure small and medium size elements (< 150 mm in maximum dimension) to an accuracy of 0.5 mm. For large elements measuring more than 150 mm in maximum dimension, a fabric measuring tape was employed. For MNI information, complete tibiae were measured and ratios of anteroposterior proximal width, transverse distal width, and transverse shaft diameter were correlated with tibia length. This allowed the lengths of partial tibiae to be estimated from other available measurements based on the proportions of complete tibiae. None of the length estimates from opposite sides were close in value, so it is unlikely that two tibiae from which lengths were estimated belonged to the same individual.