Reassessment of the enigmatic Late Cretaceous theropod dinosaur, Bagaraatan ostromi

Abstract

The Late Cretaceous Bagaraatan ostromi, described by Osmólska in 1996, is one of the most enigmatic theropod dinosaurs. The holotype possesses a peculiar combination of features, which Osmólska suggested were indicative of a primitive position among theropods that could not be resolved further. Other researchers have pointed to affinities with either derived bird-like coelurosaurs or tyrannosaurids. Here, we reanalyse all material collected by Osmólska, which reveals it to be a chimaera of multiple theropod taxa. The femur, tibiotarsus, one pedal phalanx, and most of the bones undescribed in Osmólska’s paper are identified as Caenagnathidae indet. The mandible, cervical vertebrae, pelvis, tail, and one pedal phalanx, which we consider the holotype of B. ostromi, show tyrannosaurid affinities, which are here supported by phylogenetic analyses. We find only two potentially unique diagnostic features of the Bagaraatan holotype: double surangular foramina and a horizontal ridge on the lateral surface of the postacetabular process of the ilium. Both, however, may be ontogenetically or intraspecifically variable, and thus we conclude that the holotype of B. ostromi represents an indeterminate tyrannosaurid. The small size of the holotype and its possession of many features known to characterize juvenile Tyrannosaurus rex indicate that the skeleton belongs to a juvenile, which is one of the smallest juvenile tyrannosaurids currently known.

Introduction

In 1996, Halszka Osmólska described and named a new species of a theropod dinosaur, Bagaraatan ostromi, based on a bone association collected in 1970 in the Nemegt Formation of Mongolia by the Polish–Mongolian paleontological expeditions (ZPAL MgD-I/108). The incomplete skeleton, estimated to belong to a moderately sized carnivore that was 3.0–3.5 m long, included a partial mandible and the hind part of the body (pelvis, hindlimb, and proximal tail), and was described by Osmólska (1996) as showing a mosaic of features that made it impossible to determine its relationship to other theropods. She pointed out similarities to Dromaeosauridae (i.e. general structure of the mandible and slenderness of the hindlimb), but noticed that the presence of the propubic pelvis, short caudal prezygapophyses, and lack of ossified caudal rods distinguishes it from dromaeosaurids. She also discussed affinity with tyrannosaurids, owing to the features of the mandible (i.e. shape and robustness of the anterior end, and two glenoid processes), but noted that the femoral trochanters are different in B. ostromi, in contrast to tyrannosaurids. Because of this unusual combination of features, Osmólska (1996) tentatively considered B. ostromi to be a medium-sized tetanuran with ‘primitive’ characteristics and a probable sister taxon to the Allosauridae and Coelurosauria. This placement, however, was not the result of a phylogenetic analysis.

Later authors continued to be confused by the skeleton of B. ostromi. For example, Csiki and Grigorescu (1998) noticed similarities in the hindlimb structure between B. ostromi and theropod hindlimb material from the Haţeg Basin of Romania (Bradycneme draculaeHarrison & Walker, 1975 and Elopteryx nopcsaiAndrews, 1913), i.e. fusion of the tibia and proximal tarsals, presence of a horizontal groove on the anterior surface of the astragalus, presence of a lateral tubercle on the femoral lateral articular condyle, and the femoral head positioned above the greater and lesser trochanter. Bradycneme draculae and Elopteryx nopcsai were identified by Csiki and Grigorescu (1998) as a non-coelurosaurian tetanuran and a maniraptoran, respectively. The authors proposed that those species, together with B. ostromi, might represent a new clade of peculiar small theropods with a fairly close relationship to birds, also including the ‘Iren Nor avimimid’, PIN 2549/100 (Kurzanov 1987, Csiki and Grigorescu 1998).

The idea that B. ostromi belonged to a clade of small, derived theropods was raised again with the description of Xinjiangovenator parvusRauhut & Xu, 2005, known from a partial articulated hindlimb. This species was proposed to be closely related to B. ostromi, forming a poorly known Asian lineage of coelurosaurs (Rauhut and Xu 2005). Subsequently, Br. draculae was identified as an indeterminate representative of Maniraptora, Elopteryx nopcsai was recognized as a troodontid or a non-ornithuromorphan pygostylian bird (Naish and Dyke 2004), the ‘Iren Nor avimimid’ was revealed to be a troodontid (Currie and Peng 1993), and X. parvus was placed in a polytomy with other basal Coelurosauria (Choiniere et al. 2010). Indeed, the earliest phylogenetic analyses in which B. ostromi was included placed it as a sister taxon to the derived coelurosaurian clade Maniraptoriformes (Holtz 1998) or even within the Maniraptora (Rauhut 2000, Rauhut et al. 2010), closely related to birds.

Some authors, however, noted striking similarities between B. ostromi and tyrannosaurids. Holtz (2004) claimed that the prominent and laterally expanded surangular shelf, reduced and broadened retroarticular process, and the presence of a ventral flange on the ischial peduncle of the pubis are shared between B. ostromi and tyrannosaurids. However, he also pointed out that this comparison should be treated with caution owing to the fragmentary nature of the B. ostromi holotype. Later, B. ostromi was recovered as a basal tyrannosauroid in two phylogenetic analyses (Carr and Williamson 2010, Loewen et al. 2013). In the first analysis, it was found in a polytomy, together with Iliosuchus incognitusHuene, 1932, Stokesosaurus clevelandiMadsen, 1974, Eotyrannus lengiHutt et al., 2001, Aviatyrannis jurassicaRauhut, 2003, and cf. Alectrosaurus sp. (see Carr and Williamson 2010). The second analysis (Loewen et al. 2013) positioned B. ostromi together with non-proceratosaurid tyrannosauroids closer to tyrannosaurids, i.e. Dilong paradoxusXu et al., 2004, Eotyrannus lengi, Raptorex kriegsteiniSereno et al., 2009, and Dryptosaurus aquilunguis (Cope, 1866). Thus, for the last two decades, B. ostromi has often been presumed to be a tyrannosauroid (Currie 2003a, Choiniere et al. 2010, Alifanov 2012, Bell et al. 2012, Dalman 2013).

Recently, however, B. ostromi has been considered a chimera of tyrannosaurid and non-tyrannosaurid bones (Brusatte 2013, Brusatte and Carr 2016, Napoli et al. 2021), an intriguing hypothesis that might explain the ‘mosaic’ of primitive and derived features noted by Osmólska (1996). This idea has not yet been supported by careful description, comparisons, and documentation of each bone of B. ostromi.

Here, we reassess the material of B. ostromi presented by Osmólska (1996). Although Osmólska (1996) explicitly noted that there is no doubt that the bones of B. ostromi belonged to a single individual, unpublished material catalogued under the same specimen number (ZPAL MgD-I/108) indicates that the association was a multitaxic assemblage, thus putting the taxonomic identity of individual bones into question. Thus, herein we redescribe, revise, and provide documentation of all bones of the association, including the previously unpublished bones, and reassess the status of B. ostromi. We find that what Osmólska (1996) described as a single dinosaur skeleton is, in fact, a chimera of two taxa: an oviraptorosaur and bones belonging to a juvenile tyrannosaurid. With this realization, we clarify the systematics of B. ostromi by designating an associated skeleton in the assemblage as the holotype, discuss features that might support B. ostromi as a valid taxon, but ultimately conclude that it is most likely to be an indeterminate tyrannosaurid owing to its juvenile status, thus removing a genus and species from the roster of Nemegt theropods. The holotype specimen becomes one of the smallest juvenile tyrannosaurids known, providing valuable information on tyrannosaurid growth and ontogeny.

Materials and methods

Material

The holotype of Bagaraatan ostromi, ZPAL MgD-I/108, was found in 1970 in Northern Sayr (Nemegt Formation, Late Cretaceous) during the seventh Polish–Mongolian palaeontological expedition. The specimen was described as a partial skeleton including the mandible, incomplete pelvis, incomplete left hindlimb, and proximal caudal vertebrae. According to Osmólska (1996), the anterior part of the body of ZPAL MgD-I/108 was strongly weathered, thus not collected, but the position of the mandible in the field indicated that it belonged to the same individual as the back part of the body. In the field, in the area where the holotype of B. ostromi was found, several other bones were collected and catalogued under the same specimen number: two cervical and one caudal vertebra, manus phalanx II-1 and ungual II-3, proximal end of a rib, and a few indeterminate bones. Despite that, Osmólska excluded those bones from the description of B. ostromi (Osmólska’s personal notes: ‘Found together with B. ostromi but doubtful – not described’ [translation from Polish by the authors]). She made a preliminary identification of those bones as belonging to indeterminate Oviraptorosauria. Unfortunately, no sketch or map of the arrangement of the bones of B. ostromi was drafted, and no photographs were taken before the excavation of the bones. Thus, it is impossible to make an independent assessment of the position of the bones assigned to ZPAL MgD-I/108. Herein, we assess all the bones found in the ZPAL MgD-I/108 assemblage, including the undescribed fossils, some of which have been used already in phylogenetic analyses (Loewen et al. 2013).

To account for the multitaxic character of that composite specimen, we redefine ZPAL MgD-I/108 as a catalogue number. The mandible, being the first specimen described in detail, figured, and referred to B. ostromi by Osmólska (1996), in addition to the most autapomorphic bone in the association, is here assigned the number ZPAL MgD-I/108. We also include the cervical and caudal vertebrae and pelvis under this specimen number, because they were apparently found in continuity with the mandible in the field, linked by the eroded remains of the vertebral column (Osmólska 1996: p. 3). The size, preservation, and general phylogenetic affinities of these bones are also consistent with each other, and we consider them to belong to an associated skeleton, which we refer to as the holotype of the species B. ostromi. Two pedal phalanges were also found in the assemblage; the size and taxonomic identity of the pedal phalanx IV-1 corresponds to ZPAL MgD-I/108, and it is assigned to it. We exclude all the non-tyrannosaurid material from the holotype of B. ostromi ZPAL MgD-I/108 and assign it the number ZPAL MgD-I/108/1.

Methods

The material was photographed using a Canon EOS M6 mark II camera. The fossils were also digitalized using a Shining 3D EinScan Pro 2X 3D scanner fixed on a tripod with EinScan Pro 2X Color Pack (texture scans), Ein-Turntable (alignment based on features), and EXScan Pro v.3.2.0.2 software.

The anteroposterior diameter of the surangular foramen and the skull length in 38 tyrannosauroid specimens of various body sizes (listed in the listed in the excel spreadsheet in the Supporting Information) was measured based on photographs and three-dimensional models or published data in ImageJ v.1.53e. The analysis includes 10 individuals of Tarbosaurus bataar (Maleev, 1955), one of R. kriegsteini, 10 of Gorgosaurus libratusLambe, 1914, five of Tyrannosaurus rexOsborn, 1905, six of Daspletosaurus spp., and one each of Alioramus altaiBrusatte et al., 2009, Qianzhousaurus sinensisLu et al., 2014, Bistahieversor sealeyiCarr & Williamson, 2010, Teratophoneus currieiCarr et al., 2011, and Lythronax argestesLoewen et al., 2013. In the case of ‘Shanshanosaurus huoyanshanensis’ Dong, 1977 and Go. libratus TMP 1994.12.155, the mandible length was taken instead of the skull length owing to lack of complete skulls. The mandible and skull length in Ta. bataar is reported to be almost equal, with ~1% of difference between in length between the lower jaw and the length of the skull from premaxilla to occipital condyle (Maleev 1974), and the skull and mandible in tyrannosaurids grow isometrically (Currie 2003b). Therefore, we included those individuals owing to their importance, because they represent small juveniles. Raptorex kriegsteini is not included here within Ta. bataar; although it was proposed that Raptorex is a juvenile of the latter species (Fowler et al. 2011), recently it was once again suggested to be a valid taxon (Carr 2023). Raptorex kriegsteini still requires a proper description in order to understand its affinities within the Tyrannosauroidea fully. Bagaraatan ostromi ZPAL MgD-I/108 is included, however; the length of the mandible is estimated to 30 cm based on the similar-sized specimen of Ta. bataar MPC-D 107/7 (Tsuihiji et al. 2011). Owing to the presence of two surangular foramina in ZPAL MgD-I/108, that specimen is represented on the plots by two points: one for the measurement of the posterior surangular foramen alone, and the second for the measurement of the distance from the anterior margin of the anterior surangular foramen to the posterior margin of the posterior surangular foramen. The Pearson correlation coefficient was calculated in order to determine the linear dependence of the skull length and the surangular foramen diameter. Then, a regression analysis was performed for all tyrannosaurids and separately for the specimen groups of Tarbosaurus, Gorgosaurus, Tyrannosaurus, and Daspletosaurus to find the relationship between the size of the surangular foramen and the skull length, in order to determine possible allometries and quantify the classification of the surangular foramen into imprecise size categories (‘small’, ‘medium’, and ‘enlarged’). Although it is accepted that the small or absent surangular foramen is present in basal Tyrannosauroidea, and the enlarged fenestra (~30% depth of the posterior end of the surangular; Sereno et al. 2009: 52; Carr and Williamson 2010: 204) is present in the Tyrannosauridae, the imprecise and inconsistent ‘small’, ‘medium’, and ‘enlarged’ categories are still applied in the literature concerning tyrannosaurids (e.g. Tsuihiji et al. 2011, Carr 2020, Voris et al. 2021). All statistical analyses were conducted in PAST v.4.03.

The distal end of the femur was cross-sectioned above the condyles historically, but the resulting histological thin sections were never published. The section was taken using standard procedures (Padian and Lamm 2013) in the Institute of Paleobiology, Polish Academy of Sciences. The 100 and 50 µm ground sections were analysed under normal and polarized light, including gypsum wedge using a Nikon Eclipse LV100 POL polarizing microscope with a DS-Fil camera. The pictures were combined in NIS-Elements microscope imaging software. For the description of bone histology, we used standard terminology and definitions following Chinsamy-Turan (2005) and Padian and Lamm (2013).

The amended holotype of B. ostromi (ZPAL MgD-I/108) was scored by us in two phylogenetic datasets to assess its position among coelurosaurian theropods broadly and tyrannosauroids in particular. First, it was added to the Theropod Working Group (TWiG) phylogenetic analysis including a broad sample of 152 coelurosaurian theropods scored for 853 morphological characters (Brusatte et al. 2014). We analysed the dataset in TNT v.1.5 (Goloboff et al. 2008), with Allosaurus fragilis Marsh, 1877 as an outgroup. We began with a new technology search (with default parameters for sectorial search, ratchet, tree drift, and tree fusion), which recovered a minimum length tree in 10 replicates, in order to sample tree space broadly and identify major tree islands. The returned most parsimonious trees were then subjected to a round of additional traditional search (TBR) branch swapping, to explore more fully each tree island identified by the new technology search. Second, to assess the position of the B. ostromi holotype among the tyrannosauroids in particular, we scored it in a phylogenetic analysis including 31 tyrannosauroid species (and four outgroups) and 366 characters (the Nesbitt et al. 2019 version of the Brusatte and Carr 2016 dataset). The dataset was analysed in TNT using the same parameters as above. For both analyses, clade support was determined via Bremer supports and jackknife percentages.

Systematic palaeontology

Dinosauria Owen, 1842; Saurischia Seeley, 1887; Theropoda Marsh, 1881; Coelurosauria Huene, 1914; Tyrannosauroidea Osborn, 1905; Tyrannosauridae Osborn, 1905; Tyrannosauridae indet.
Bagaraatan ostromi Osmólska, 1996
Holotype

ZPAL MgD-I/108: incomplete right mandible (dentary, angular, surangular, prearticular, and articular), left and right incomplete ilia, nearly complete left pubis, partial right pubis, proximal end of left ischium, left pedal phalanx IV-1, two cervical vertebrae, 25 caudal vertebrae, and two haemal arches.

Note on diagnostic characters

We provide a full diagnosis below, because we must first describe all the bones of the Bagaraatan series before untangling which different taxa they belong to. However, we note here that this holotype individual can be referred to the Tyrannosauridae because of eight features: (i) presence of the dentary ‘chin’; (ii) transition between the anterior and ventral edges of the dentary placed below the fourth alveolus; (iii) dorsoventrally narrow Meckelian groove deeply inset into the medial side of the dentary; (iv) extremely reduced retroarticular process of the articular; (v) prominent surangular shelf; (vi) convex anterior margin of the pubis; (vii) cervical vertebrae with a hypapophysis; and (viii) thick posterior centrodiapophyseal laminae.

Locality and age

Northern Sayr, Nemegt, Ömnögov, Mongolia; Nemegt Formation.

Description

Mandible

Only two fragments of the left mandible are preserved: the anterior part of the dentary with poorly preserved supradentary, and a piece that includes articulated posterior parts of the surangular, angular, and prearticular, and the incomplete articular.

Dentary

The dentary is slender in general outline and shows an anterior expansion in comparison to the midregion (28 mm deep at the third vs. 25.5 mm deep at the ninth alveolus; Figs 1, 2), which is D-shaped in cross-section. Also, the dentary is labiolingually expanded anteriorly: the anterior end is wide labiolingually (measuring 16.3 mm) in comparison to the posterior part of the preserved dentary (12.1 mm width; Fig. 1C). The anterior tip of the dentary is missing; however, clearly it was positioned higher than the level of the tooth row (its preserved base is already dorsal relative to the rest of the bone; Fig. 1A, B). The anteroventral margin is relatively straight and strongly inclined posteroventrally, creating an angle of 135° with the ventral margin of the dentary. This creates a distinct ‘chin’ (i.e. slightly protruding region at the place where the anteroventral and ventral margin meet) between the anterior and ventral surfaces, which is positioned underneath the third and fourth alveoli. The ‘chin’ underneath the fourth alveoli is commonly seen in juvenile tyrannosaurines (Carr 2020: character 117) and Alioramus altai (Brusatte et al. 2012), but in adults the ‘chin’ is placed ahead of the fourth alveolus, as Ta. bataar (Fig. 3), and Ty. rex (Brusatte and Carr 2016: character 171). A low angle of the symphyseal region relative to the ventral margin is found in juvenile tyrannosaurids with narrow jaws, contrasting with the steeper rostroventral margin of deep-jawed adult individuals, where the ‘chin’ migrates further anteriorly (Figs 2, 3; Carr and Williamson 2004, Carr 2020).

The dorsal margin of the dentary is strongly concave in lateral view, even in the anterior part, which is a feature of derived tyrannosaurids (Brusatte and Carr 2016: character 177) that is also seen in juveniles and subadults (Currie and Dong 2001, Tsuihiji et al. 2011, Brusatte et al. 2012, Funston et al. 2020b). The ventral margin of the dentary is only very slightly convex (Figs 1A–C, 2E, F). The lateral surface is smooth; the neurovascular foramina pierce the bone along an anteroposterior sulcus (i.e. dentary groove; Figs 1A, 2E) 7.8 mm below the tooth row. The foramina are more numerous in the anterior part of the dentary, close to the symphysis (Figs 1F, 2B). The dentary groove is an ontogenetically variable feature in tyrannosaurids, being sharp and deep in juveniles and shallow in mature individuals (Fig. 3; Brusatte et al. 2016, Carr 2020). On the ventral side of the dentary, a second row of foramina, parallel to the ventral margin, is present. Anteriorly, those foramina are larger and closely spaced; posteriorly, the foramina are smaller and widely spaced (Figs 1A, F, 2B, E).

The medial side of the dentary is smooth, with a deep and narrow groove that extends anteroposteriorly between the interdental plates and the rest of the dentary (Figs 1B, 2F). The interdental plates are poorly preserved, but their triangular shape is visible in medial view. The symphysis is elongated, aligned anterodorsally, and has a nearly smooth surface (bearing only minute, very subtle striations). A ‘chin’ is present, as in other tyrannosaurids (Brusatte and Carr 2016: character 172), including small juveniles (Funston et al. 2020b), with the exception of Q. sinensis (Lu et al. 2014, Foster et al. 2021). The ventral margin of the symphysis ends below the fourth tooth alveolus, where a single anterior Meckelian foramen is present (Figs 1B, 2F). The position is similar to that in Ta. bataar (e.g. ZPAL MgD-I/4 and ZPAL MgD-I/175; Fig. 3) and other tyrannosaurids (Brusatte et al. 2010, Funston et al. 2020b), but in Alioramus altai the foramen is positioned further posteriorly, below the fifth tooth alveolus (Brusatte et al. 2012). The anterior Meckelian foramen is located anterior to the anterior end of the Meckelian groove, which is shallower anteriorly and cuts more deeply into the dentary posteriorly. The deep and sharp inset of the Meckelian groove is a characteristic of tyrannosaurids and close relatives (Brusatte and Carr 2016: character 178) and is seen in small juveniles of Ty. rex (Carr 2020) and other juvenile tyrannosaurids (Funston et al. 2020b). Anteriorly, the groove is positioned in the middle of the medial surface of the dentary, but posteriorly it is positioned in the upper third of the dorsoventral height of the dentary. The distance between the Meckelian groove and the tooth row also shortens posteriorly (from 13.25 mm anteriorly to 8 mm posteriorly). In dorsal view, the preserved part of the dentary is straight (Figs 1C, 2A), similar to Alioramus altai (Brusatte et al. 2012) and juvenile Ta. bataar (Tsuihiji et al. 2011).

The dentary shows 11 tooth alveoli. Nine dentary teeth are broken off, but nine complete tooth alveoli are preserved, along with most of a tiny mesial-most alveolus at the front of the jaw, and the anterior end of the 11th alveolus at the back. The preserved part of the first alveolus is exceptionally small in comparison to the other alveoli, whereas the second is larger than the first, but still smaller than the remaining teeth and with a circular outline (Figs 1F, 2B; Table 1). This indicates that the first two teeth in the jaw were smaller and more circular in cross-section than the remaining teeth, as is common in tyrannosauroids (Brusatte and Carr 2016: character 175), including small juveniles (Funston et al. 2020b). The alveoli posterior to the first two are elongated mesiodistally and have an eight-shaped outline in dorsal view (Figs 1C, 2A). The labiolingual width is the largest at the third alveolar position, and the anterior and posterior alveoli are narrower. The anteroposterior length of the alveoli decreases anteriorly, such that the 10th alveolus is the longest. These alveoli indicate that the associated teeth are ziphodont, with labiolingual widths < 60% mesiodisal lengths, as is the case in most theropods and juvenile tyrannosaurids, but differing from the labiolingually widened incrassate teeth of large adult tyrannosaurids (Brusatte and Carr 2016: character 201).

Supradentary

As correctly noted by Osmólska (1996), only a small, poorly preserved splinter of the supradentary is present in articulation, dorsal to the dentary and lingual to the interdental plates, at the level of the fifth to the seventh teeth (Figs 1B, 2F). Some uninformative, miniscule scraps of bone are also present posteriorly. As preserved, the supradentary appears to be dorsoventrally narrow, covering less than one-fifth of the mandible height.

Splenial

We could not confirm the presence of a triangular, slightly hooked anterodorsally anterior part of the splenial suggested by Osmólska (1996). The triangular element is most likely to be a cracked and inset ventral bar of the dentary.

Surangular

Only the posterior part of the left surangular is preserved (Figs 1A–E, 4). The surangular is a generally thin, plate-like bone, which expands labiolingually at the dorsal margin. Lateroventrally, the surangular is covered by the flat and mediolaterally thin angular (Figs 1A, 4A). The angular ends very close to (only 4 mm below) the surangular foramen. In ventral view, the connection between the surangular, articular, and prearticular is visible. The ventromedial edge of the surangular contacts the prearticular. This contact is visible externally in the posterior part, but more anteriorly the surangular is partly obscured by the angular; it continues only as a narrow splinter along the posterior half of the preserved part of the angular. As preserved, the contacts between the bones in that area appear split as a result of compaction, hence their precise layout might be displaced, and thus it is possible that in vivo the surangular was either not exposed from under the angular or that the exposure was slightly larger but now is obscured and/or partly eroded. In any case, the deformation most probably was not substantial. The angular tightly covers the surangular, such that the margin between those bones is barely visible laterally but well marked ventrally.

The most conspicuous aspect of the surangular is the presence of two surangular foramina: one smaller (2.3 mm × 1.5 mm) and positioned anterodorsally, and the second larger (diameter: 5.63 mm × 3.68 mm) and placed posteroventrally to the first one (Figs 1A, 4A). Both are elongate, ovoid in shape rather than circular, with the long axes directed posterodorsally. The bone is extremely thin between those foramina (Figs 1E, 4F). This condition is different from most tyrannosaurids, where a single surangular foramen is enlarged, such that its dorsoventral depth is > 30% of the depth of the surangular (Brusatte and Carr 2016: character 179). This is the case in Nemegt tyrannosaurids, like alioramins (Brusatte et al. 2009, 2012, Lu et al. 2014), Ta. bataar (e.g. ZPAL MgD-I/4 and ZPAL MgD-I/31; Fig. 5), and the young juvenile Raptorex (Sereno et al. 2009). However, in some other juvenile tyrannosaurids there is a single surangular foramen, but it is small (Currie and Dong 2001, Tsuihiji et al. 2011), and it has been determined that the size of the foramen changes during the ontogeny of Ty. rex (Carr 2020: character 126). Both surangular foramina in B. ostromi are located in a fossa below the lateral surangular shelf. There is no pneumatic pocket posterodorsal to the surangular foramen, whereas nearly all other tyrannosaurids have one (Brusatte and Carr 2016: character 183), although it is absent in some specimens of Ta. bataar (e.g. ZPAL MgD-I/4 and ZPAL MgD-I/31; Fig. 5; and MPC-D 107/7, Tsuihiji et al. 2011). In B. ostromi, the two foramina are separated by a laterally convex, lateroposterodorsally inclined, dorsally thickening (up to ~5 mm), and gently posteroventrally bowed bar (Figs 1A, 2A). The bar buttresses the posterior part of the lateral surangular shelf.

There is a lateral surangular shelf above the foramina, close to the dorsal margin of the bone (Figs 1A, E, 2A, B, F). Its lateral protrusion is not as prominent as in Alioramus altai (Brusatte et al. 2012) or Ta. bataar (e.g. ZPAL MgD-I/4, ZPAL MgD-I/5, and ZPAL MgD-I/31; Fig. 5), but the lateral protrusion of the surangular shelf is subtle, as in some juvenile tyrannosaurids (Currie and Dong 2001, Tsuihiji et al. 2011, Foster et al. 2021). In lateral view, the shelf extends straight anteroposteriorly, paralleling the long axis of the mandible, as in tyrannosaurids, but differing from the anteroventral or anterodorsal orientation in most other theropods (Brusatte and Carr 2016: character 182) The smooth surface of the adductor fossa dorsal to the shelf faces almost equally dorsally and laterally. This is similar to both species of Alioramus (Kurzanov 1976, Brusatte et al. 2012) and juvenile tyrannosaurids (Currie and Dong 2001, Tsuihiji et al. 2011), but differs from the strongly laterally facing state in large adult tyrannosaurids (Brusatte and Carr 2016: character 184). In older Ta. bataar, the fossa is immediately medial to the shelf, extends medioventrally, and forms a depression (more pronounced in smaller specimens), but more medially the adductor fossa curls up and faces strongly laterally (e.g. ZPAL MgD-I/31, ZPAL MgD-I/4, and ZPAL MgD-I/5; Fig. 5). The dorsally pointing posterior edge of the adductor fossa is more pronounced than in Alioramus altai (Brusatte et al. 2012). There is a triangular fossa on the lateral surface of the surangular shelf anteroventral to the glenoid, a distinguishing feature of derived tyrannosauroids (Brusatte and Carr 2016: character 185). The glenoid on the surangular (lateral glenoid socket of Osmólska 1996) is a deep and anteroposteriorly narrow transverse concavity bound anteriorly and posteriorly by dorsally extended processes (the preglenoid process and conelike process, respectively; Figs 1, 4). This is similar to Alioramus altai (Brusatte et al. 2012) and juvenile Ta. bataar (Tsuihiji et al. 2011); in larger Ta. bataar the glenoid is anteroposteriorly wider (e.g. ZPAL MgD-I/4, ZPAL MgD-I/5, and ZPAL MgD-I/31; Fig. 5). Posteromedially to the glenoid fossa (lateral glenoid socket sensuOsmólska 1996), in dorsal view, a deep and narrow fossa is present (medial glenoid socket sensuOsmólska 1996). In Ta. bataar, the two glenoid fossae are not marked by the upraised lateral margin of the surangular. Two glenoid depressions are present in that species, but similar in depth and separated by a gradual elevation (e.g. ZPAL MgD-I/4). In B. ostromi, the medial glenoid is much deeper than the lateral glenoid. There is a fossa on the lateral surface of the surangular, ventral to the glenoid, as is seen in derived tyrannosauroids (Brusatte and Carr 2016: character 186). This fossa is smooth, as in Alioramus altai (Brusatte et al. 2012), not rugose, as in Ta. bataar (ZPAL MgD-I/4, ZPAL MgD-I/5, and ZPAL MgD-I/31). Distal to the glenoid, behind the posterior dorsal (conelike) process, a second, groove-like concavity (cleft of Osmólska 1996) is present, bound posteriorly by a small but well-defined dorsal projection, which continues medially as a sharp, distinct ridge (Figs 1C, 4A). This feature occurs in both Alioramus altai and Ta. bataar.

The retroarticular process of the surangular is tiny, corresponding to the small corresponding process on the articular (Fig. 1A, B). This is a feature of tyrannosauroids (Brusatte et al. 2014: character 76). This process is straight and slopes posteroventrally, similar to Alioramus altai (Brusatte et al. 2012) and Q. sinensis (Lu et al. 2014). In Ta. bataar (ZPAL MgD-I/4 and ZPAL MgD-I/5), it is oriented vertically. The medial hook process is nearly perpendicular to the prearticular axis of the surangular and constitutes almost 50% of the width of the surangular.

Angular

Only the left posterior part of the angular is preserved (Figs 1D, 4A, C). It is plate-like, laterally convex, securely sutured, and tightly covers the surangular. Its margins are marked in the lateral view by a shallow groove, historically marked with a pen, making exact observation difficult (Figs 1A, 4A). The dorsal margin of the posterior plate of the angular is convex below the anterior of the two surangular foramina and concave below the distal margin of the posterior foramen, where the dorsoventral height of the angular decreases posteriorly. The distance between the dorsal margin of the angular and the ventral margin of the posterior surangular foramen is short (4.2 mm). The posterior margin is convex, pointing slightly upwards, and the ventral margin is straight and contacts the surangular posteriorly and the prearticular anteriorly. The posterior tip of the angular is not complete. The preserved posterior end of the angular extends past the level of the posterior margin of the posterior surangular foramen.

Prearticular

The posterior process of the left prearticular is preserved and is tightly articulated with the articular posteriorly, the angular ventrally, and the surangular dorsally, laterally, and posteroventrally (Figs 1B–E, 4C, D). The posteromedial tip of the prearticular is broken off. The preserved part of the prearticular is medially concave in ventral view. The ventral margin between the prearticular, angular (anteriorly), and surangular (posteriorly) runs sigmoidally in ventral view, and only posteriorly does the margin between the bones curve medially (note that the bones are slightly split along the ventral surface of the mandible, but that does not seem to distort their general layout). The articular, surangular, and angular are tightly articulated with the prearticular. The prearticular is not fused to the surangular and articular, similar to juvenile Tarbosaurus (Currie and Dong 2001, Tsuihiji et al. 2011) and Alioramus altai (Brusatte et al. 2012) and in contrast to large Ta. bataar (ZPAL MgD-I/4 and ZPAL MgD-I/5). The posteroventral margin of the prearticular is pointed downwards (similar to Alioramus altai), whereas in Ta. bataar it is oriented posteriorly. The distal concave margin contacting the articular is shallower than in Ta. bataar.

Articular

The articular is almost complete, lacking only the ventromedial part. It is tightly articulated with the prearticular anteromedialy and contacts the surangular laterally. The posterior surface is smooth, gently concave, and elliptic, more than twice as tall as it is wide. The retroarticular process is extremely reduced (Figs 1, 4D, E), as in all Tyrannosauroidea, but differing from the much larger processes in dromaeosaurids and other theropods (Brusatte et al. 2014: character 76). The attachment site for the jaw muscles on the articular is mediolaterally narrower than the glenoid articular surface, and there is a very narrow non-articular region between the glenoid and the muscle attachment. Both features are characteristic of most tyrannosauroids, but not other theropods (Rauhut et al. 2010, Brusatte and Carr 2016: characters 189 and 190).

Antarticular

We could not confirm the presence of a separate antarticular suggested by Osmólska (1996). As preserved, the structure in question is a cracked medial edge of the surangular.

Postcranial skeleton: Cervical vertebrae

Two incomplete amphiplatyan cervical vertebrae are preserved (Fig. 6). They are similar in structure and size: the anteroposterior length of the anterior cervical centrum (Fig. 6A–F) measures 35.8 mm, and the posterior cervical centrum (Fig. 6G–L) measures 36.5 mm. The articular surfaces of the centrum are oval, slightly concave, and shallow dorsoventrally. The height to width ratio of the centra is 0.7 and 0.6 for the anterior and posterior cervical vertebra, respectively. The centra are concave laterally, and they thicken close to the parapophyses, which are oval in lateral view and directed laterally (Fig. 6B, H). On the lateral sides of the centra, pleurocoels (lateral pneumatic fossae) are present. Above the pleurocoels, the posterior centrodiapophyseal laminae are thick and laterally offset, and they demarcate a deep infradiapophyseal fossa anteriorly, as in all tyrannosaurids, but differing from the thinner laminae of more basal tyrannosauroids (Brusatte and Carr 2016: character 213). Sutures between the centra and neural arches are open. Small, eroded hypapophyses on the anterior region of the ventral surface of the cervical vertebrae are present, as in tyrannosaurids and close relatives (Brusatte and Carr 2016: character 214), including juveniles, such as the Alioramus altai holotype (Brusatte et al. 2012).

The cervical vertebrae are similar to the mid- or posterior cervical vertebrae of the juvenile tyrannosaurid ‘Shanshanosaurus huoyanshanensis’, because both exhibit flat ventral surfaces of the centrum, which are also narrow-waisted, biconcave, and with a large and single pleurocoel on the lateral side (Currie and Dong 2001).

Caudal vertebrae

Twenty-one caudal vertebrae were found in articulation (Figs 7–14). Four distal caudal vertebrae and two haemal arches were also found (Figs 14, 15) but cannot be fitted to the articulated tail. The first preserved caudal is taller dorsoventrally than long (Fig. 7A–F; Table 2), whereas the second is roughly equal in height and length (Fig. 7G–L), and all successive centra are longer than tall (Figs 8–14; Table 2). The transverse processes disappear starting from the 15th preserved caudal (Fig. 12M–R). In Ta. bataar (ZPAL MgD-I/4, ZPAL MgD-I/175, and ZPAL MgD-I/177), the height of the centrum is similar to its length in the fifth caudal vertebra, and the transverse processes disappear starting from the 18th caudal vertebra. Thus, we estimate that the preserved articulated part of the tail ZPAL MgD-I/108 represents the fourth to 24th caudal vertebrae. Moreover, in the first preserved caudal, the transverse processes are oriented posteriorly (Fig. 7A–F), which is a typical condition of the proximal caudal vertebrae of tyrannosaurids.

The neural arches of the caudal vertebrae in ZPAL MgD-I/108 are co-ossified with the centra in all bones, but the remnant of the suture is visible in the proximal centra, up to the 18th caudal (Fig. 12M–R). This suture is also present in the proximal caudal vertebrae of other tyrannosaurids, including Ta. bataar, and also in some other theropods, such as ornithomimids (e.g. Gallimimus bullatusOsmólska et al., 1972, ZPAL MgD-I/94).

The caudal centra are all amphicoelous; only the first preserved caudal of ZPAL MgD-I/108 is somewhat concave anteriorly and flat posteriorly (Fig. 7A–F). In both Ta. bataar (ZPAL MgD-I/4, ZPAL MgD-I/175, and ZPAL MgD-I/177) and Ty. rex (Brochu 2003), the caudal vertebrae are amphicoelous, and the first four centra are somewhat concave anteriorly. This supports the identification of the first preserved caudal of ZPAL MgD-I/108 as the fourth caudal vertebra (Fig. 7A–F). The lateral surfaces of the centra do not have any pleurocoels or other pneumatic features, and on the ventral surfaces there are no ridges (Figs 7–14). The articular surfaces for the haemal arches are present at the posteroventral end of the centra; these are well visible and similar in shape to those in Ta. bataar (e.g. ZPAL MgD-I/175).

The neural arches are generally incomplete. The robust and rectangular neural spines of the proximal caudal vertebrae lack their dorsal ends, but even as preserved they project beyond the level of the posterior limit of the respective centra, as in most other tyrannosaurids (Brusatte and Carr 2016: character 229). The ontogenetic component to this character was noticed by Carr (2020). In Ty. rex, the spinous processes of the caudal vertebrae do not extend behind the level of the posterior edge of the centrum, as in juvenile Ta. bataar (Tsuihiji et al. 2011) and, apparently, the Raptorex holotype (Sereno et al. 2009). In adult Ty. rex, the spinous process of the caudal vertebrae extends posterior to the centrum (Carr 2020), as in Ta. bataar (ZPAL MgD-I/3 and ZPAL MgD-I/175). The neural spines of ZPAL MgD-I/108 are inclined posteriorly along the tail, similar to Alioramus altai (Brusatte et al. 2012) and Q. sinensis (Lu et al. 2014) and in contrast to Ta. bataar, in which the neural spines project more vertically (ZPAL MgD-I/4, ZPAL MgD-I/175, and ZPAL MgD-I/177). Further distally, the neural spines become more strongly inclined posteriorly, and from the 16th caudal vertebra they become short dorsoventrally and elongated anteroposteriorly (Figs 12–14). The dorsal expansion present on the posterodorsal end of the neural spine in other tyrannosaurids (Brusatte et al. 2012) is not preserved in B. ostromi, and thus its presence cannot be confirmed.

The transverse processes are mostly incomplete in the caudal series of B. ostromi. Proximal caudal vertebrae have anteroposteriorly long and dorsoventrally thin, distally narrowing transverse processes. From the ninth caudal vertebra onwards, the transverse processes are still thin and flat, and directed laterally. Then, the 15th caudal vertebra shows reduced transverse processes, much shorter and narrow anteroposteriorly. The 16th and 17th caudal vertebrae have minute transverse processes, and the 18th and further caudal vertebrae lack the transverse processes (Figs 12–14). On the anteroventral surface of each transverse process, where the process meets the prezygapophysis, there are two laminae that define a deep, triangular concavity. This is present in most other tyrannosaurids, including juvenile specimens such as the Alioramus altai holotype (Brusatte et al. 2012), but absent in more basal tyrannosauroids and other theropods (Brusatte and Carr 2016: character 231). A triangular depression was noticed in Alioramus altai (Brusatte et al. 2012) at the region where the transverse process meets the neural spine, but in B. ostromi it is proportionally wider and shallower. In Ta. bataar, the depression is rather broad and shallow, regardless of the size of the animals (ZPAL MgD-I/3, ZPAL MgD-I/4, and ZPAL MgD-I/175); however, the depth and width of the depression depend on the preservation: in the caudal vertebrae of Ta. bataar ZPAL MgD-I/3, the depression is narrow and deep on the left side but shallow and wide on the right side. The depth and breadth of the fossa is best explained by taphonomic deformation, and thus its taxonomical value is limited.

The prezygapophyses of the proximal caudal vertebrae are positioned more vertically than in Ta. bataar (ZPAL MgD-I/3, ZPAL MgD-I/4, and ZPAL MgD-I/175) and Alioramus altai (Brusatte et al. 2012). Further distally, the prezygapophyses point more anteriorly, and from the 17th caudal onwards they are longer and project even more anteriorly (Figs 12–14). The surface and shape of the articular surfaces of the prezygapophyses is not visible owing to their tight articulation with the postzygapophyses or damage. The postzygapophyses are positioned behind the centrum, and their articular surfaces face lateroventrally, more laterally than in in Ta. bataar (ZPAL MgD-I/3, ZPAL MgD-I/4, and ZPAL MgD-I/175) and Alioramus altai (Brusatte et al. 2012).

Owing to the close articulation between the caudal vertebrae, the hypantrum between the prezygapophyses is not visible. The hyposphene between the postzygapophyses is large and rectangular in B. ostromi, similar to Ta. bataar (ZPAL MgD-I/3, ZPAL MgD-I/4, and ZPAL MgD-I/175) and in contrast to the delicate hyposphene found in Alioramus altai (Brusatte et al. 2012).

Ilia

The ilia are incomplete; the left and right ventral postacetabular processes, part of the left proximal preacetabular process, and apparently, two fragments of the dorsal edge of the left ilium blade are preserved (Fig. 16). Osmólska (1996) mentioned (but did not illustrate) a thin bone fragment found some distance from the remainder of the pelvis, with an even natural dorsal edge and dense, perpendicular striations on one of the surfaces, which she interpreted as the dorsal edge of the ilium. The material catalogued under ZPAL MgD-I/108 includes two fragments fitting that description (Fig. 16K–N). Given the presence of other dinosaur species in the association and the lack of articulation with the remainder of the skeleton, their affinity to B. ostromi is uncertain, although possible.

The base of the preacetabular process was positioned above the pubic peduncle, as marked by the attachment site of the muscle iliofemoralis internus, the cuppedicus fossa (Fig. 16A, B), characteristic for tyrannosaurids and other tetanurans (Hutchinson 2001, Carrano and Hutchinson 2002). Dorsally, the cuppedicus fossa is a wide and slightly concave area, which curls down laterally and forms the ventral margin of the preacetabular process. The dorsal margin of the preserved element of the preacetabular blade is crushed diagenetically. Above the ventral margin of the preacetabular blade, a depression is present.

Above the acetabulum, on the lateral surface of the right iliac blade (Fig. 16G–J), an eroded linear ridge is present (Fig. 16I). This structure is present in all tyrannosauroids, including the juvenile MPC-D 107/7 (Tsuihiji et al. 2011), but excluding R. kriegsteini and Q. sinensis (Lu et al. 2014, Brusatte and Carr 2016: character 258). Possibly, the absence of this feature in the latter two might reflect an individual or growth variation.

The right postacetabular process is taphonomically compressed mediolaterally, and its pubic peduncle and the supraacetabular crest are eroded (Fig. 16C–F). The ischial peduncle is robust, and the acetabular surface is flat. Distally, the ischial peduncle is laterally, ventrally, and medially surrounded by a shallow depression. Further posteriorly from the ischial peduncle, ventrally, a large and deep brevis fossa is present. It is concave, wide mediolaterally, and gradually widens posteriorly, from 11 mm anteriorly to 28 mm distally. Such widening occurs also in Alioramus altai (Brusatte et al. 2012). There is no foramen at the base of this fossa, as in Ta. bataar (ZPAL MgD-I/4), but the foramen is present in Alioramus altai (Brusatte et al. 2012). The medial and lateral walls of the brevis fossa are formed by the medial and lateral flanges of the postacetabular process. The lateral flange is thicker than the medial flange, as in Ta. bataar (ZPAL MgD-I/3) and Alioramus altai (Brusatte et al. 2012). The brevis fossa is visible in lateral view only anteriorly; further posteriorly it is concealed by the lateral flange of the postacetabular process. Above the beginning of the brevis fossa, the lateral flange of the postacetabular process continues dorsally as a dorsal, ~24-mm-long crest described by Osmólska (1996), surrounded by anterior and posterior depressions.

On the medial surface of the right ilium of ZPAL MgD-I/108, parts of three sacral ribs are present: one above the acetabulum, the second above the pubic peduncle, and the last positioned on the medial flange (Fig. 16C–J). Owing to the position of the sacral ribs, we agree with Osmólska (1996) that they belong to the third to fifth sacral vertebrae. If so, the laterally exposed brevis fossa terminates posteriorly at the level of the anterior part of the fifth sacral vertebra.

Pubes

The left pubis (proximal part and shaft preserved) is more complete than the right (where only the proximal part is preserved; Fig. 17). The articulation facet for the ilium is preserved in the left pubis (Fig. 17A–E). The contact with the pubic peduncle of the ilium is clear: the lateral margin is laterally extended with a rugose surface. In dorsal view, the pubic portion of the acetabulum is wider transversely, but shorter anteroposteriorly, than the ischial part. Below the acetabulum, the pubis narrows medialolaterally and forms a thin plate. The pubic tuberosity is incomplete, but it is present as a distinct convex structure, as in many tyrannosauroids, including juveniles such as the Raptorex holotype, but it does not have the highly rugose from of large subadult and adult tyrannosaurids, such as Ta. bataar (ZPAL MgD-I/3 and ZPAL MgD-I/5) (Brusatte and Carr 2016: character 270). In B. ostromi, the tubercle is essentially level with the obturator notch, as in tyrannosaurids (Brusatte and Carr 2016: character 271). Ventral to the pubic tuberosity and the articulation surface with the ischium, the pubis narrows anteroposteriorly and slightly widens transversely. Here, the main shaft of the pubis is anteriorly concave when seen in lateral view (Fig. 17A), as in tyrannosaurids generally, but differing from the straighter condition in the juvenile Raptorex holotype (Brusatte and Carr 2016: character 269). On the posteromedial surface of the bone, the beginning of the pubic apron is preserved as a sigmoidal crest running along the medial surface of the pubic shaft (Fig. 17B, D, E). Its shape is similar to Ta. bataar (ZPAL MgD-I/175). The medial surface of the pubic apron is missing. The pubic shaft is circular in cross-section, starting from the region where the pubic apron appears, and remains circular until the end of the preserved part of the pubis (although the lateral surface of the pubic shaft is missing). In distal view, the proximal part of the pubis (above the shaft) is less bowed laterally than in Ta. bataar (ZPAL MgD-I/3 and ZPAL MgD-I/175). This, however, can be accounted for by fact that the these two subadult individuals of Ta. bataar are twice the size (~7 m in length) of B. ostromi.

Ischium

Only the proximalmost left ischial plate, including the peduncles, is preserved (Fig. 17A–D). The articular surface of the pubic peduncle is tightly articulated with the ischial peduncle of the pubis. The pubic peduncle is separated from the ischial peduncle by an elliptic concavity. In dorsal view, the concavity is walled laterally by a wide and low margin (5 mm wide mediolaterally in the narrowest place), which expands anteriorly and posteriorly until reaching the peduncle margins, forming an hourglass-shaped margin (Fig. 17C). Medially, the concavity is walled by a straight, mediolaterally thin, and dorsally extended lamina, which is also present in other tyrannosaurids. The articular surface of the pubic peduncle is 26 mm tall proximodistally and 20 mm wide mediolaterally. The lateral surface of the preserved part of the proximal ischium is concave, whereas the medial surface is only slightly concave. The articular surface of the iliac peduncle is 31 mm wide mediolaterally and 23.5 mm long anteroposteriorly. The lateral margin of the iliac peduncle is strongly extended laterally. In dorsal view, it is elliptical and has a concave articular surface with the ischial peduncle of the ilium, similar to other tyrannosaurids (Brusatte et al. 2012).

Pedal phalanx

The left phalanx IV-1 is 33 mm long (Fig. 18), its length to width ratio is 1.5. The proximal articular surface is wider (22 mm) than tall (19 mm; unlike Ta. bataar, where the proportions are the opposite: ZPAL MgD-I/29, ZPAL MgD-I/175, and ZPAL MgD-I/206), however, the dorsal and plantolateral margins of the phalanx are incomplete. The proximal articular surface is concave, in a similar manner to Ta. bataar individuals. The medial margin of the articular surface is slightly concave, and the opposite lateral margin is convex. In the dorsal and planar view, the phalanx IV-1 of ZPAL MgD-I/108 is rectangular, only slightly narrowed in the middle. In the lateral and medial view, the phalanx is triangular in overall shape, clearly narrowing (stronger on the lateral than medial side) immediately before the distal condyles. In dorsal view, a supracondylar basin is present, immediately behind the slightly elevated margin of the distal articular surface. The supracondylar basin is only slightly wider mediolaterally than long proximodistally (the length to width ratio is 0.8; in Ta. bataar specimens, the basin is much wider than long, and the ratio is ~0.4), and in comparison to Ta. bataar individuals, the basin is shallower. The lateral condyle is smaller than the medial condyle, and the lateral ligament pit is shallower in comparison to the medial one, as in all Ta. bataar individuals studied (ZPAL MgD-I/3, ZPAL MgD-I/4, ZPAL MgD-I/5, ZPAL MgD-I/29, ZPAL MgD-I/175, ZPAL MgD-I/206, and ZPAL MgD-I/331). The distal margin of the medial condyle is circular, its dorsal end does not form a pointed posteriorly tip, as in young Ta. bataar (ZPAL MgD-I/29), but in contrast to larger individuals (ZPAL MgD-I/3, ZPAL MgD-I/4, ZPAL MgD-I/5, ZPAL MgD-I/175, ZPAL MgD-I/206, and ZPAL MgD-I/331), where the tip is present. In dorsal view, the distal margin of the medial condyle is pointing anteromedially. The medial condyle is higher plantodorsally and wider medialolaterally than the lateral condyle. The distal condyles are separated by a cleft (which is acute and narrower in comparison to Ta. bataar individuals) along the entire articular surface. The rounded margin of the lateral condyle in lateral view is not complete on the plantar side. On the dorsal side, the margin of the articulation surface is smooth, only slightly lifted up. In larger individuals of Ta. Bataar, the dorsal end of the articular surface in lateral view is clearly demarcated.

The pedal phalanx IV-1 of young Ta. bataar ZPAL MgD-I/29 shows the same length to width ratio as ZPAL MgD-I/108. In subadults of Ta. Bataar, the ratio is 1.3 (e.g. ZPAL MgD-I/175), and in adults it is 1.2 (e.g. ZPAL MgD-I/206). Despite the fact that the phalanx IV-1 of B. ostromi is more slender than in subadult and adult Ta. bataar, it is short and wide, as is typical for tyrannosaurids, in contrast to the elongated and slender pedal phalanges of ornithomimids (length to width ratio is 1.7 for Ga. bullatus ZPAL MgD-I/94), caenagnathids (length to width ratio is 2.1 for Elmisaurus rarus Osmólska, 1981 ZPAL MgD-I/98), or troodontids (length to width ratio is 1.7 for Borogovia gracilicrusOsmólska, 1987, ZPAL MgD-I/174).

Oviraptorosauria Barsbold 1976; Caenagnathoidea Stenberg, 1940; Caenagnathidae Stenberg, 1940

Referred material

ZPAL MgD-I/108/1: Left manus phalanx II-1, manus ungual I-2, proximal and distal ends of the left femur, tibiotarsus, and rib.

Note on diagnostic characters

We provide full details below, because we must first describe all the bones of the Bagaraatan original series before untangling which different taxa they belong to. However, we note here that this set of bones can be referred to Caenagnathidae because of: (i) the presence of lateral pleurocoels in the proximal caudal centra; (ii) lesser and greater trochanters in contact; (iii) clearly demarcated accessory trochanter; and (iv) gracile and straight shape of the manual phalanx.

Locality and age

Northern Sayr, Nemegt, Ömnögov, Mongolia; Nemegt Formation.

Description

Caudal vertebrae

The centrum of one caudal vertebra is preserved (Fig. 19A–F). It is 28 mm long, 19.5 mm tall, and 23 mm wide (the height to width ratio of the centrum is 0.8). The centrum is oval, only slightly compressed dorsoventrally. Laterally, the centrum bears one pleurocoel (pneumatic foramen) on each side. The presence of lateral pleurocoels in the caudal vertebrae is a synapomorphy of Caenagnathoidea (Lamanna et al. 2014). The centrum is only slightly concave laterally. Ventrally, two parallel ridges extend along the centrum, as in Elmisaurus rarus (specimen MPC-D 100/119 ‘Nomingia gobiensis’ Barsbold et al. 2000).

Ribs

Only a proximal part of a dorsal rib is preserved; the rib is broken at the tuberculum (Fig. 19G, H). The capitulum is bulbous. Behind the slightly convex articular surface, no depression is present, in contrast to tyrannosaurids (Ta. bataar, e.g. ZPAL MgD-I/3, ZPAL MgD-I/4, and ZPAL MgD-I/175, and Alioramus altai; see Brusatte et al. 2012). Also, in contrast to the latter, the tuberculum is enlarged. Because the capitulum and tuberculum are at a similar level, the rib is likely to come from the posterior part of the ribcage. The overall shape of the preserved part of the rib corresponds to the morphology in caenagnathids (e.g. Caenagnathidae indet. ZPAL MgD-I/99).

Manus phalanx II-1

The left phalanx is straight and elongated, measuring 76.6 mm (Fig. 19O–T). The proximal articular surface is taller (18 mm) than wide (16 mm) and divided by a low ridge, which is narrow dorsally and wide ventrally. On both sides of the ridge, the articular surfaces are teardrop-shaped and strongly concave. The distal medial condyle (13.5 mm high) is smaller than the lateral one (15.4 mm high) and separated by a deep and narrow furrow. The medial ligament pit is shallower than the lateral ligament pit. The width of the distal end is 15 mm; the length to width ratio of the phalanx is 4.7.

The gracile and straight shape of the manus phalanx II-1 of ZPAL MgD-I/108/1 is the same as in Elmisaurus rarus ZPAL MgD-I/98, although the phalanx of ZPAL MgD-I/108/1 is larger. The length of manus phalanx II-1 of ZPAL MgD-I/98 is 66 mm, the proximal width 14 mm, and the distal width 12 mm (the length to width ratio is 4.7, same as for ZPAL MgD-I/108/1). The manus phalanx II-1 of ZPAL MgD-I/108/1 shares also with Elmisaurus rarus slightly downturned distal condyles and expanded articular surfaces of the distal condyles. Other theropods known from the Nemegt Formation, i.e. tyrannosaurids (Ta. bataar, e.g. ZPAL MgD-I/3 and ZPAL MgD-I/4), ornithomimids (Ga. bullatus, cast of MPC-D 100/11; Deinocheirus mirificusOsmólska & Roniewicz, 1970, cast of MPC-D 100/18), avimimids [alvarezsaurids (Mononykus olecranusPerle et al., 1993 (Perle et al. 1994)], and oviraptorids [e.g. Oksoko avarsan (Funston et al. 2020a), Nemegtomaia barsboldiLu et al., 2005 (Fanti et al. 2012)] do not have manual phalanges that are so straight, slender, and elongated.

Manus ungual II-3

The ungual is elongated (54 mm in length), curved, and very narrow, and the proximal articular surface is 13 mm wide (Fig. 19I–N). The ungual lacks only the distal tip. The proximal articular surface is oval (longer dorsoventrally than mediolaterally). A vertical ridge, which is dorsally and ventrally expanded but constricted in the middle section, extends across the middle of the articular surface. The articular surfaces on both sides of the ridge are strongly concave. The dorsal edge of the articular surface forms a robust dorsal lip, surrounded by a depression. A ventral process is present on the ventral edge of the articular surface. The articular surface is separated by a notch from the ventrodistally located enlarged flexor tubercle. Laterally and medially, the collateral groove extends along the entire ungual, starting from the area above the flexor tubercle.

The manus ungual II-3 is not known in Elmisaurus rarus; however, the presence of the distinctive dorsal lip indicates that the ungual corresponds to the manual unguals of Caenagnathidae. In comparison to the manual unguals of an North American caenagnathid, Chirostenotes pergracillisGilmore, 1924, CMN 2367 (Funston 2020), the ungual of ZPAL MgD-I/108/1 is less curved than the phalanges I-2 and III-4, but more straight, similar to II-3. Moreover, the proximal articulation is offset, and the flexor tubercle is distally positioned and smaller in contrast to unguals I-2 and III-4, which further supports its identification as II-3 of a caenagnathid. Other theropods known from the Nemegt Formation, i.e. tyrannosaurids (Ta. bataar, e.g. ZPAL MgD-I/3 and ZPAL MgD-I/4), ornithomimids (Ga. bullatus, cast of MPC-D 100/11; De. mirificus, cast of MPC-D 100/18), alvarezsaurids [M. olecranus (Perle et al. 1994)], and oviraptorids [O. avarsan (Funston et al. 2020a) and N. barsboldi (Fanti et al. 2012)] do not have such enlarged, curved, and transversely narrow manual unguals with an enlarged flexor tubercle distinctly separated from the ventral process and distinctive dorsal lip.

Femur

Two parts of the left femur are preserved: the proximal and distal end; most of the shaft is missing, hence the length of the femur is unknown (Fig. 20). The circumference of the shaft portions preserved with the distal and proximal parts is 105 mm. Osmólska (1996) hypothesized that ~80–90 mm of the shaft is missing, adding up to a total femur length of 310–320 mm.

The proximal part of the femur (Fig. 20A–E) is narrower lateromedially than longer anteroposteriorly. In dorsal view, the femur is L-shaped. The posterior part of the greater trochanter is connected to the femoral head that projects mediodistally, and the anterior part of the greater trochanter is widened anteriorly. In anterior view, the femoral head is positioned higher than the greater trochanter, and they are separated by a broad, shallow depression. The surface of the rounded femoral head is rugose. In posterior view, a wide groove for the capital ligament is present on the femoral head. In medial view, the femoral head is ovoid, and its posterodorsal margin is wider than the anteroventral end. The neck is narrower anteroposteriorly than the head; and the ventral margin of the head is directed downwards before connecting to the neck. The neck extends upwards from the greater trochanter, which is wider lateromedially than the lesser trochanter. The lesser trochanter is almond-shaped in anterior view. The dorsal margin of the femoral trochanters in lateral view is arched; the small, anteriorly positioned lesser trochanter is separated by a shallow groove from the much anteroposteriorly longer greater trochanter. On the lateral surface of the proximal part of the femur, the separation between the lesser and greater trochanter is marked by a shallow and short groove. Below the lesser trochanter, the accessory trochanter (anterior crest sensuOsmólska 1996) is present. It is slightly expanded anteriorly and extends along the preserved part of the proximal shaft. The accessory trochanter keeps a consistent lateromedial width along the preserved proximal part of the shaft. A posterior tubercle is present below the greater trochanter, well visible in anterior and posterior views.

The distal end of the femur is now longitudinally shorter than described by Osmólska (1996), because it has since been thin sectioned. At that time, it measured 105 mm; now, only the distalmost part of the femur including both condyles is present, measuring 52 mm (Fig. 20F–J). The medial condyle is bigger than the lateral condyle, but the lateral condyle extends further distally than the medial condyle. The condyles are distally separated by a deep but narrow notch (the popliteal fossa). Anteriorly and distally, the condyles are separated by shallower and wider depressions (the extensor grooves). The medial condyle is convex, with a slightly rugose surface. The lateral condyle bears an elevation on its distal surface. The tibiofibular crest extends posteromedially. In lateral view, the tibiofibular crest is axe-shaped and projects further posteriorly than the medial condyle.

The accessory trochanter appeared in Tetanurae as a branch of the distal base of the lesser trochanter, and it was reduced in Eumaniraptora. The accessory trochanter is smaller in basal Tetanurae, Carnosauria, basal Coelurosauria, Tyrannosauridae, and Ornithomimosauria than in Caudipteryx spp., Microvenator celerOstrom, 1970, Caenagnathidae, and some Oviraptoridae (Hutchinson 2001). The accessory trochanter of the femur of ZPAL MgD-I/108/1 is clearly demarcated from the lesser trochanter and forms a dorsoventral flange, comparable to that seen in Caenagnathidae, e.g. Elmisaurus rarus ZPAL MgD-I/98, Anzu wylieiLamanna et al., 2014, or Chirostenotes pergracilis Gilmore, 1924 (Currie and Russell 1987). The lesser and greater trochanters are in contact, as in all Caenagnathoidea (Lamanna et al. 2014). The proximal end of the femur further resembles the femur of Elmisaurus rarus ZPAL MgD-I/98 in possessing a cylindrical head positioned higher than the greater trochanter and separated by a depression, which is wider in the larger (ontogenetically older, as indicated by the difference in size between those specimens) ZPAL MgD-I/108/1. Such an embayment is also present in other Caenagnathidae, e.g. Anzu wyliei (see Lamanna et al. 2014), Elmisaurus rarus (see Barsbold et al. 2000), or Ch. pergracilis (see Currie and Russell 1987). A wide groove on the posterior surface of the femoral head for the capital ligament is present in both Elmisaurus rarus ZPAL MgD-I/98 and MgD-I/108/1. Also, similar to Caenagnathidae, the lateral condyle of the femur ZPAL MgD-I/108/1 is positioned more distally than the medial condyle, and the tibiofibular crest is well demarcated [Anzu wyliei (see Lamanna et al. 2014), Elmisaurus rarus (see Barsbold et al. 2000), or Ch. pergracilis (see Currie and Russell 1987)]. The extensor groove is distinct, but shallow, consistent with Elmisaurus rarus (see Barsbold et al. 2000). The proximal and distal ends of the femur ZPAL MgD-I/108/1 are of similar size to the measurements in Elmisaurus rarus (see Barsbold et al. 2000), hence the probable length of the whole bone was similar, ~285 mm.

Bone microstructure of femur

A histological section of the distal part of the shaft, above the condyles, shows a large marrow cavity and thin (~2 mm) cortex (Fig. 20K, L). The external part of the cortex (half of its thickness) is built of parallel-fibred bone, with scattered secondary osteons. The vascularization is laminar, and growth marks are absent. In the section, no definite primary osteons were seen, although the external cortex is poorly preserved, possibly obscuring their presence. The inner cortex is sharply demarcated from the external cortex and built of densely packed secondary osteons: up to four generations are present. Close to the marrow cavity, resorption cavities are present, surrounded by a thick layer of lamellar bone (≤ 0.3 mm). The marrow cavity is surrounded by a thinner layer of lamellar bone (0.15 mm) and filled by slender and elongated bony trabeculae.

The section shows features typical for the metaphyses of long bones: extensive secondary remodelling, lack of growth marks, and numerous resorption cavities. Thus, owing to the lack of any growth record in the section, it is not possible to estimate the growth ratio.

The bone microstructure of the femur in caenagnathids is unknown. Thus far, bone histology of the tibiae of cf. Anzu wyliei (ROM 65884) and Caenagnathidae indet. (UALVP 57349) have been described (Funston and Currie 2018, Cullen et al. 2021). Both, however, represent young individuals, as indicated by their predominant fibrolamellar bone, high vascularity, and limited secondary remodelling (none in UALVP 57349 and ≤ 30% of cortex in OMVP 65884). The predominance of fibrolamellar bone and high vascularity are also seen in the cortices of the femora and fibulae of the oviraptorid O. avarsan, regardless of their ontogenetic age (Funston et al. 2020a). Even in the large-bodied cf. Anzu wyliei (ROM 65884), the section from the tibia revealed a predominately primary tissue, with generally high vascularity and limited secondary remodelling (Cullen et al. 2021). As can be noticed, the section taken from ZPAL MgD-I/108/1 is different from the Caenagnathoidea described before, which is a result of its sectioning at the metaphysis, and not the diaphysis as usually done.

Tibiotarsus

The left tibiotarsus is complete and measures 380 mm (Fig. 21A–F). The bone is slender, slightly bowed laterally (possibly taphonomically exaggerated), and the distal fibula is fused to the distal tibia and calcaneum (Fig. 21A–C). The shaft is elliptical in cross-section (circumference 95 mm) and is compressed anteroposteriorly, possibly as an effect of taphonomical crushing. Proximally, the tibia expands anteriorly and slightly mediolaterally; its anteroposterior depth is 47.5 mm and mediolateral width 59.2 mm. Distally, where the tibia is fused with the astragalocalcaneum distally and the fibula laterally, the tibia expands mediolaterally and measures 57.3 mm.

The cnemial crest condyle (cranial cnemial crest sensuOsmólska 1996) is robust, laterally deflected, and short in anterior view, comprising only ~15% of the maximum proximodistal tibiotarsus length. The fibular condyle (lateral cnemial crest sensuOsmólska 1996) is also robust, slightly curved anteriorly, and shorter mediolaterally and dorsoventrally than the cnemial crest. Between the cnemial crest and fibular condyle, a deep and posteriorly curved incisura tibialis is present (Fig. 21A–E). The medial proximal condyle of the tibiotarsus is long anteroposteriorly; anteriorly, it is smoothly connected with the cnemial crest; posteriorly, it is separated from the fibular condyle by a triangular cleft. The posteromedial edge of the medial proximal condyle of the tibiotarsus is posteriorly extended. Below the fibular condyle, the fibular crest is present. It is tall dorsoventrally, ~20% of the tibiotarsus length. The crest becomes wider mediolaterally and deflects anteriorly; however, the crest is not strongly pronounced. The distal end of the crest is rectangular.

The fibula is fused to the lateral side of the distal end of the tibia, along the distal ~ 23% of the tibiotarsus length (Fig. 21A–C). The distal end of the fibula is partly fused with the calcaneum. The outline of the distal part of the fibula is marked and distinguishable against the remaining bones. The suture between the astragalocalcaneum and distal tibia are clearer in anterior than posterior view; however, that might be a matter of preservation. No suture is visible between the astragalus and calcaneum. The calcaneum shows a lateral depression (lateral epicondylar depression) below the suture with the fibula. The preserved, incomplete ascending process of the astragalus extends along 7.5% of the length of the tibiotarsus. In anterior view, it has subtriangular, pointed medial and lateral processes, separated by a deep depression. At the base of the ascending process, a shallow median depression is present, above which a low, mediolaterally extended ridge is located (Fig. 21A).

A proper tibiotarsus, in which the tibia is fused to the proximal tarsals, is recognized in three non-avian maniraptoran taxa: Alvarezsauridae, Troodontidae, and Avimimidae. Both alvarezsaurids known from the Nemegt Formation (M. olecranus and Nemegtonykus citusLee et al., 2019) have a proximodistally short fibula, which does not reach even the midshaft of the tibiotarsus (Perle et al. 1994, Lee et al. 2019). The presence of a tibiotarsus in the troodontids known from the Nemegt Formation is variable. In the larger species Zanabazar junior (Barsbold, 1974), the astagalocalcaneum is not fused to the tibia (Norell et al. 2009), whereas in the smaller Bo. gracilicrus, a tibiotarsus is present (Osmólska 1987, Cau and Madzia 2021). The hindlimb is unknown in the third troodontid from the Nemegt Formation, Tochisaurus nemegtensisKurzanov & Osmólska, 1991. Only a fragment of proximal right fibula of Bo. gracilicrus is preserved, but the distal end of the tibiotarsus does not show any signs of fusion with the distal fibula, as in other Troodontidae (e.g. Gao et al. 2012). Oviraptoridae and Caenagnathidae show a fused astragalus and calcaneum, but not to the tibia (Currie et al. 2016). Finally, a fused tibiotarsus including the distal end of the fibula is an autapomorphy of Avimimus spp. (Kurzanov 1981, Funston et al. 2018).

However, ZPAL MgD-I/108/1 would be an exceptionally large representative of Avimimus; the largest reported tibiotarsus of Avimimus nemegtensis Funston, Mendonca, Currie & Barsbold, 2018 MPC-D 102/92 measures 282 mm (Funston et al. 2016), and in Avimimus portentosus Kurzanov, 1981 PIN 3907/1 it is 257 mm long (Kurzanov 1981), whereas ZPAL MgD-I/108/1 measures 380 mm, similar to Elmisaurus rarus MPC-D 100/119, i.e. 355 mm. The histological sections of the Iren Dabasu avimimids revealed that the largest sectioned specimens were already adults (Funston et al. 2019). Moreover, three features of the tibia are shared by ZPAL MgD-I/108/1 and Elmisaurus rarus: (i) the medial proximal condyle is more protruded dorsally than in Avimimus spp.; (ii) the fibular crest is longer, and its distal end is rectangular, not arcuate as in Avimimus spp.; and (iii) the medial malleolus protrudes further medially than in Avimimus spp.

Fibula

The left fibula is complete, measuring 340 mm. The distal end is partly fused to the calcaneum and laterally fused to the tibia, along ~33% of the length of the fibula. It is similar to Avimimus portentosus, in which the fibula is fused to the tibia along one-third of its length (Kurzanov 1981). The proximal anteroposterior length of the fibula is 47.1 mm. The proximal end of the fibula is triangular in the lateromedial aspect, only slightly concave medially in dorsal view (Fig. 21G–K), similar to Elmisaurus rarus (MPC-D 100/119; Barsbold et al. 2000) and in contrast to the rectangular proximal end in Avimimus spp. (Kurzanov 1981, Funston et al. 2016). Distal to the expanded proximal end, the fibula narrows anteroposteriorly. On the medial surface is a long (74.2 mm, ~22% of total length) fusiform attachment for the fibular crest of the tibia. At this level, on the anterior surface of the fibula, an elliptical iliofibularis tubercle is present. Distally, the fibula strongly narrows anteroposteriorly and has a triangular cross-section along ~65% of its total length.

Despite the distal part of fibula being fused with the tibia in a similar manner to Avimimus spp., the proximal end of the fibula is more triangular, as in MPC-D 100/119 (Barsbold et al. 2000), than rectangular, as seen in Avimimus spp.

Pedal phalanx

The left phalanx II-2 is 35 mm long (Fig. 19U-Aʹ), and its length to width ratio is two. The proximal articular surface is triangular in posterior view and can be divided into lateral and medial teardrop-shaped concave articular surfaces, separated from each other by a smooth ridge running in the middle of the proximal articular surface. In the lateral and medial views, the proximal articular surface is strongly concave, the plantar margin extends backwards, and the lip-shaped dorsal margin (extensor turbecle) is elevated dorsally and directed posteriorly. The distal articular surface is composed of the lateral condyle and slightly shorter plantodorsally medial condyle, which are separated by a concavity that is shallow dorsally but becomes deeper along the articular surface to its end on the plantar side. In anterior view, the lateral condyle extends further downwards than the medial condyle. The ligament pits are well marked on the both sides of the phalanx; the lateral ligament pit is elongated anteroposteriorly, and the medial ligament pit is circular.

The phalanx II-2 is similar to the corresponding phalanx of Elmisaurus rarus (ZPAL MgD-I/98; length to width ratio of 2.1), especially in the structure of the proximal articular surface, i.e. two teardrop-shaped surfaces separated by a ridge, and the lip-like extensor tubercle. The phalanx II-2 in other theropods from the Nemegt formation differs from the one of ZPAL MgD-I/108/1. This phalanx in Ga. bullatus (ZPAL MgD-I/94) is compressed (the length to width ratio is 1.4), and the extensor tubercle is less pronounced than in ZPAL MgD-I/108/1. The phalanx II-2 is even more compressed in Bo. gracilicrus (ZPAL MgD-I/174; the length to width ratio is 1.2). The proximal articular surface in Ta. bataar (ZPAL MgD-I/3, ZPAL MgD-I/4, ZPAL MgD-I/29, and ZPAL MgD-I/175) is wider than long, in contrast to Elmisaurus rarus, Ga. bullatus, and ZPAL MgD-I/108/1. The length to width ratio of the phalanx II-2 in Ta. bataar is 1.5–1.6, depending on the ontogenetical age.

Results

Size of the surangular foramen in tyrannosaurids

The Pearson correlation coefficient for the measurements of the anteroposterior length of the posterior surangular foramen and skull length is high and statistically significant (.85653; P = .0001), indicating that the anteroposterior diameter of the surangular foramen is dependent on the skull length in tyrannosaurids (Fig. 22A). The regression analysis for all specimens and taxa included shows a negative slope and a high correlation coefficient (slope: 1.32; R² = .8; P = .0001), and the same is true even when juveniles are excluded (slope: 1.2; R² = .68; P = .001). The same is true for Ta. bataar (slope: 0.04; R² = .78; P = .0005), Ty. rex (slope: 0.03; R² = .94; P < .1), Go. libratus (slope: 0.05; R² = .76; P < .01), and Daspletosaurus spp. (slope: 0.08; R² = .88; P < .01; Fig. 22B). This indicates a negative allometry, meaning that the size of the surangular foramen decreases during ontogeny in those tyrannosaurid species. This trend was reported previously for Ty. rex (Carr 2020). Our analysis does not find the surangular foramen of Bi. sealeyi, Daspletosaurus spp., Te. curriei, Q. sinensis, and Alioramus altai as enlarged in comparison to other tyrannosaurids. Also, the surangular foramen of Ta. bataar is not smaller than in other tyrannosaurids as reported previously (Tsuihiji et al. 2011, Voris et al. 2021).

Phylogenetic analysis

The parsimony analysis of the larger, inclusive dataset of coelurosaurian phylogeny (based on Brusatte et al. 2014) resulted in 10 000 most parsimonious trees (the memory limit in TNT) of 3361 steps (Consistency Index = .322, Retention Index = .777). The strict consensus tree (Fig. 23) places Bagaraatan ostromi among tyrannosauroids, as the sister taxon to the clade Dr. aquilunguis + Tyrannosauridae. B. ostromi shares seven synapomorphies with tyrannosauroids or major inclusive clades: (i) extremely reduced retroarticular process of the articular; (ii) pubic tubercle forming a convexity on the anterior margin of the pubis; (iii) mediolateral width of the jaw muscle attachment site of the articular equal to glenoid width (Di. paradoxus + Tyrannosauridae); (iv) surangular shelf prominent; (v) surangular shelf not overhanging the surangular foramen, which abuts the shelf (Eotyrannus lengi + Tyrannosauridae); (vi) cervical vertebrae with hypapophysis (Xiongguanlong baimoensisLi et al., 2010 + Tyrannosauridae); and (viii) thick, laterally offset posterior centrodiapophyseal laminae, which demarcates a deep infradiapophyseal fossa anteriorly in the anterior–middle cervical vertebrae (B. ostromi + Tyrannosauridae). These results strongly support the tyrannosauroid affinity of the B. ostromi holotype. The exact position of B. ostromi outside of Tyrannosauridae, however, might be an artefact of the immature status of the specimen.

Because of the robust tyrannosauroid affinities of the B. ostromi holotype, its more precise relationships within Tyrannosauroidea are better tested with a phylogenetic dataset designed to assess tyrannosauroid ingroup relationships. Thus, when we included it in the tyrannosauroid dataset of Brusatte and Carr (2016), we found 40 most parsimonious trees (769 steps, Consistency Index = .553, Retention Index = .812). The strict consensus tree (Fig. 24) places B. ostromi within a polytomy of derived tyrannosauroids (with Bi. sealeyi, Q. sinensis, Alioramus spp., and Appalachiosaurus montgomeriensis Carr, Williamson, Schwimmer & 2005, Albertosaurinae, and Tyrannosaurinae), which indicates that it is either a member of Tyrannosauridae or a very close relative. Among the synapomorphies of Tyrannosauroidea found also in B. ostromi are: (i) cervical vertebrae with hypapophysis; and (ii) thin, sharp, and deeply inset Mackelian groove on the dentary. The synapomorphies of Tyrannosauridae (or a slightly more inclusive clade, depending on the resolution of the polytomy) present in B. ostromi are: (i) transitional point between the ventral and anterior margins of the dentary positioned below the fourth alveolus; and (ii) the presence of a dentary ‘chin’. Finally, the presence of a small surangular foramen is found to be an autapomorphy of B. ostromi, a reversal from the enlarged condition in closely related taxa.

Because of the juvenile nature of the specimen (see Discussion below), we interpret these results as strong support for the tyrannosaurid (or near-tyrannosaurid) affinities of B. ostromi but do not put much stock into its exact position on the tree, because juveniles often fall out more basally than adults in phylogenetic analysis (Tschopp et al. 2015).

Discussion

The Bagaraatan assemblage ZPAL MgD-I/108

It will remain unknown why Osmólska (1996) refrained from even mentioning some of the remaining bones found together with what she considered to be B. ostromi. Our complete reanalysis of the material collected and catalogued together revealed that ZPAL MgD-I/108 is not a single individual, but an assemblage of two different non-avian dinosaurs, in which only one individual was partly articulated. The presence of an oviraptorosaurian in the chimaera explains why some previous phylogenetic analyses found B. ostromi to be related to Maniraptora (Holtz 1998, Rauhut 2000, Rauhut et al. 2010).

Osmólska (1996) did remove most bones of the Caenagnathidae indet. found as part of ZPAL MgD-I/108 from what she considered to be B. ostromi, possibly owing to the fact that 15 years earlier (Osmólska 1981) she described a caenagnathid from the Nemegt Formation, Elmisaurus rarus, with characteristic elongated manual phalanges. The caenagnathid bones from the assemblage that she did not exclude from the description of B. ostromi are the femur, tibiotarsus, and pedal phalanx II-2 (ZPAL MgD-I/108/1). The femur of the holotype of Elmisaurus rarus is incomplete (e.g. the accessory trochanter is missing), hence Osmólska might have missed the similarities in the structure of the proximal end of the femur of Elmisaurus rarus and the specimen she considered to be B. ostromi. The femur of ZPAL MgD-I/108/1 definitely does not belong to a tyrannosaurid owing to the presence of: (i) a clearly demarcated anterior trochanter, which is proximodistally long; (ii) a saddle-like proximal margin of the femur; (iii) a lateral condyle positioned more distally than the medial condyle; and (iv) a well-demarcated tibiofibular crest. The body size of the individual ZPAL MgD-I/108/1 was similar to Elmisaurus rarus (see Barsbold et al. 2000). The identification of the tibiotarsus is more problematic. The fusion of the astragalocalcaneum with the tibia and the fusion of the distal end of the fibula to the tibia should unambiguously indicate a representative of Avimimidae (Kurzanov 1981, Funston et al. 2018). However, the large (for an avimimid) size of the tibiotarsus and the morphological features of the tibia and fibula are more reminiscent of the Caenagnathidae. In the latter taxon, the fusion usually occurs between the astragalus and calcaneum and less often between the distal tarsals and metatarsals (as proposed by Currie et al. 2016 in mature individuals). There are two caenagnathids thus far known in which the astragalus is fused with the tibia (Atkins-Weltman et al. in review unpublished observations). Therefore, given the large size of the individual, a caenagnathid-like general morphology of the tibia and fibula, and the general tendency towards the fusion of the bones among the heel joint in the Caenagnathoidae, we decided to recognize the tibiotarsus as belonging to a senile or/and atypical (in the terms of bone fusion) caenagnathid. The size of the tibiotarsus fits to the remaining bones of ZPAL MgD-I/108/1, hence they are likely to have belonged to a single individual. Finally, the pes phalanx II-2, as described above, matches the general morphology seen in the caenagnathids. Thus, the hindlimb bones that Osmólska (1996) originally considered a part of B. ostromi could not have been found in articulation with each other, nor with the tyrannosauroid skeleton that we designate as the amended holotype of B. ostromi. Two species of caenagnathids were known from the Nemegt Formation of Mongolia: Elmisaurus rarus and ‘Nomingia gobiensis’; however, the validity of the latter was recently called into question (Funston et al. 2018, 2021), implying that only one caenagnathid lived in the Nemegt Formation (Elmisaurus rarus), hence the specimen ZPAL MgD-I/108/1 is most likely to be a representative of that species.

Finally, the mandible, axial skeleton, and pelvis of the B. ostromi material was found in articulation, as noted by Osmólska (1996) (Fig. 25). This specimen, ZPAL MgD-I/108, is here designated as the holotype of B. ostromi and is referred to Tyrannosauridae indet. owing to the presence of numerous diagnostic features, including a dentary ‘chin’ and the position of the transition between the anterior and ventral edges of the dentary below the fourth alveolus (Fig. 1). Those features were considered synapomorphies of Bi. sealeyi + Appalachiosaurus montgomeriensis + Tyrannosauridae (Brusatte et al. 2010, Brusatte and Carr 2016) and appear early in the ontogeny of tyrannosaurids (Funston et al. 2020b). The dorsoventrally narrow Meckelian groove deeply inset into the medial side of the dentary is a tyrannosauroid synapomorphy, also present already in hatchlings (Funston et al. 2020b). The extremely reduced retroarticular process of the articular and prominent surangular shelf present in B. ostromi are also typical for tyrannosauroids (Brusatte et al. 2014). The features of the postcranial skeleton, such as the convexity on the anterior margin of the pubis (present in all Tyrannosauroidea), cervical vertebrae with a hypapophysis (present in X. baimoensis + Tyrannosauroidea), and thick posterior centrodiapophyseal laminae (present in Dr. aquilunguis + Tyrannosauroidea) further support the close affinities of this associated holotype skeleton of B. ostromi with tyrannosaurids.

Only three clearly valid tyrannosaurid species are currently known from the Nemegt Formation in Mongolia: the large-bodied, common Ta. bataar, and the much smaller and uncommon long-snouted Alioramus remotus Kurzanov, 1976 and Alioramus altai. Another potentially valid taxon, R. kriegsteini (Sereno et al. 2009, Carr 2023), also most probably comes from the Nemegt Formation (Fowler et al. 2011), but given that its holotype is a very small juvenile, its taxonomic status is difficult to determine and awaits full description of the material. Bagaraatan ostromi is much smaller than all of them (Raptorex notwithstanding), with an estimated body length not larger than 3 m, vs. 10 m in Ta. bataar and 5–6 m in the known subadult skeletons of Alioramus spp., which admittedly might have been larger as adults (Brusatte et al. 2009, 2012). The size of the mandible of Bagaraatan is slightly larger than those of ‘Shanshanosaurus huoyanshanensis’, which was once considered a valid species from the Chinese Subashi Formation (contemporary to the Nemegt Formation; Wan et al. 2007) but is now regarded a juvenile tyrannosaurid (Currie and Dong 2001). The B. ostromi mandible, however, is similar in size and general structure to the mandible of the juvenile Ta. bataar MPC-D 107/7 described by Tsuihiji et al. (2011). This raises a question: does B. ostromi represent a juvenile of one of the already known Nemegt tyrannosaurid species?

Is B. ostromi a valid taxon?

Osmólska (1996) listed eight diagnostic features for B. ostromi: (i) two surangular foramina [also considered by Holtz (2004) as an autapomorphy of B. ostromi]; (ii) articular with an oblique posterior surface and a short retroarticular surface; (iii) caudal vertebrae with thin-walled centra; (iv) hyposphene–hypantrum articulations in at least the first 16 caudal vertebrae [also listed by Holtz (2004) as an autapomorphy of B. ostromi]; (v) prezygapophyses in proximal caudal vertebrae with ridges on the lateral surfaces; (vi) ilium with two deep depressions; (vii) femur with the anterior trochanter (anterior crest sensu Osmólska); and (viii) tibia and fibula fused distally. The status of those features is briefly discussed below. Given that we have now re-identified the hindlimb bones as belonging to other non-tyrannosaurid taxa, those features regarding the hindlimb were discussed above, hence they will be omitted in this section.

The ilium with a distinct ridge on the lateral surface of the postacetabular process, demarcated anteriorly, medially, and posteriorly by depressions, is striking (Fig. 16). It occurs symmetrically on both ilia, and better preserved on the left, which is less compressed. Such ridges are not found in other theropods, to our knowledge, and are not present in juvenile Ta. bataar (MPC-D 107/7) nor the other tyrannosaurid juvenile, R. kriegsteini (Sereno et al. 2009). Therefore, they might be a diagnostic feature of B. ostromi.

The ridges on the lateral surfaces of the prezygapophyses are found also in the proximal caudal vertebrae of Ty. rex, running from the anterior margin of the transverse process to the prezygapopysis (Brochu 2003). Similar ridges on the prezygapophyses of anterior caudal vertebrae are also present in Ta. bataar (ZPAL MgD-I/176). Osmólska (1996) did not quantify how thin-walled the caudal vertebrae centra of B. ostromi are in comparison to other theropods. We do not recognize any clear difference between the centrum thickness of B. ostromi and other theropods. The stout hyposphene–hypantrum articulations in at least the first 16 caudal vertebrae were considered an autapomorphy of B. ostromi by Osmólska (1996) and Holtz (2004). The presence of the hyposphene–hypantrum articulation is seen in many archosaurs, is strongly correlated with body size, and is often already present at a young age, before the articulation is necessary to support the large mass of the fully grown animal (Stefanic and Nesbitt 2019). The hyposphene–hypantrum articulations are common in theropods, and for instance, are present in the caudal vertebrae of medium-sized Ta. bataar (ZPAL MgD-I/176). The oblique posterior surface and short retroarticular surface of the articular, also listed by Osmólska (1996), are tyrannosauroid synapomorphies (Brusatte et al. 2014).

The presence of two surangular foramina and the ridge on the lateral surface of the postacetabular process of ilium seem to be the only two features listed by Osmólska (1996) that distinguish B. ostromi from other tyrannosaurids. The two surangular foramina were also later listed by Holtz (2004) as unique for B. ostromi in comparison to other theropods (Currie et al. 2003). There is some confusion in the literature about the size of the surangular foramen in tyrannosauroids and its phylogenetic significance and ontogenetic and individual variation. In their phylogenetic dataset of tyrannosauroids, Brusatte and Carr (2016) used a character that simply divided the size of the foramen into two states: those with a dorsoventral depth < 30% or > 30% of the depth of the posterior end of the surangular. This was based on earlier characters used by Sereno et al. (2009), Carr and Williamson (2010), and Brusatte et al. (2010). The enlarged condition was found to be synapomorphic of a clade consisting of Dryptosaurus + Tyrannosauridae, whereas the primitive smaller foramen is seen in more basal tyrannosauroids, such as Suskityrannus, Eotyrannus, Dilong, and proceratosaurids.

Other authors, however, have considered the foramen differently. The size of the surangular foramen in tyrannosaurids was divided into moderate (Go. libratus, Albertosaurus sarcophagusOsborn, 1905, Ty. rex, and Ta. bataar) or enlarged (Bi. sealeyi, Daspletosaurus spp., Te. curriei, Thanatotheristes degrootorumVoris et al., 2020, Q. sinensis, and Alioramus altai) states by Voris et al. (2021). However, the surangular foramen in Ty. rex and Albertosaurus sarcophagus also used to be classified as smaller than in other tyrannosaurids (Carr and Williamson 2004). Other authors reported that the surangular foramen in Ta. bataar is smaller than in other tyrannosaurids and invariant during ontogeny (Tsuihiji et al. 2011, Voris et al. 2021). Also, the surangular foramen in ‘Shanshanosaurus huoyanshanensis’ was described as large (Currie and Dong 2001), but later as small (Tsuihiji et al. 2011). For Ty. rex, the size of the surangular foramen was first reported as increasing (Carr 1999) but later as decreasing in size through ontogeny (Carr 2020).

Because of this confusion, we built a dataset to examine the size of foramina in a quantitative context. In tyrannosaurids, growth of the mandible, skull, and femur is isometric and related to the body size of the individual (Currie 2003b). Thus, we assessed the relationship between the size of the surangular foramen and skull length (as a proxy for body size). Our results (Fig. 22) show that in all taxa the size of the surangular foramen decreases during ontogeny (negative allometry) and is strongly correlated with the length of the skull. Thus, e.g. Alioramus altai (IGM 100/1844) and the similar-sized Go. libratus (TMP 1991.36.500) have surangular foramina of proportionally the same size. Although the surangular foramen–skull length correlation is statistically significant, variability in surangular foramen size is also apparent, especially in Go. libratus and Ta. bataar, for which the data are less fitted to the trend than for the other species (R² = .76–.78, vs. R² > .88–.94 in Daspletosaurus spp. and Ty. rex; Fig. 22). Indeed, although the surangular foramen is rather enlarged in Tarbosaurus individuals (as in other tyrannosaurids; Fig. 22) bigger than MPC-D 100/66 (skull length: 45 cm), few exceptions were found within the hypodigm. The surangular foramen of the medium-sized specimen MPC-D 107/14 is exceptionally small in comparison to other Ta. bataar individuals of similar size (e.g. ZPAL MgD-I/3 and MPC-D 107/5; Fig. 5). Moreover, a specimen larger than those listed above, MPC-D 100/60, shows asymmetrical surangular foramina: the left one is smaller (anteroposterior length: 23 mm) and the right one larger (anteroposterior length: 40 mm). The smaller surangular foramen of the left mandible can be noticed as an outlier in the Figure 22. It would appear that there was some variability in the timing of the surangular foramen enlargement, at least in Ta. bataar. The size of the surangular foramen in B. ostromi, regardless of whether it is measured for the single (posterior only) or double (posterior + anterior) foramina, falls into the overall variability of surangular size in the tyrannosaurids generally and Ta. bataar specifically. The position of the surangular foramen in ‘Shanshanosaurus’ and Ta. bataar MPC-D 107/7 is similar to the position of the posterior opening in the surangular of Bagaraatan, and those individuals cluster together on the plot. In turn, if the length of the area of both surangular foramina is measured for B. ostromi, it clusters with R. kriegsteini, the surangular foramen of which was previously reported to be ‘enlarged’ (Fowler et al. 2011).

What might explain the strange double set of foramina in Bagaraatan? The bone between the anterior and posterior surangular foramina in B. ostromi is very thin, and the relative position of this area and both foramina matches the surangular foramen of R. kriegsteini and Ta. bataar (ZPAL MgD-I/31). The surangular in tyrannosaurids during the early years of life was invaded by a pneumatic diverticulum (Gold et al. 2013), which pneumatized the bone and formed the enlarged surangular foramen, bordered by a pneumatic pocket posterodorsal to it. Given that more basal tyrannosauroids have a small foramen without a pneumatic pocket, it is not clear whether there was any pneumatic diverticulum in this region in these species. Owing to the fact that pneumatic diverticula induce bone resorption when they contact bone (Bremer 1940, Witmer 1997, Wedel 2007), we propose that the mandible of B. ostromi exhibits local bone resorption, induced by the pneumatic diverticula, that would explain the extremely thin bone between the anterior and posterior foramen. We hypothesize that if the pneumatization process continued slightly longer, the two foramina might have merged into a single large foramen, which is the common condition in Dryptosaurus + Tyrannosauridae (Brusatte and Carr 2016). This proposal is supported by the fact that the posterior surangular foramen in B. ostromi is similar in length and positioned in a similar place as in the smaller ‘Shanshanosaurus’ and Ta. bataar MPC-D 107/7 (skull length ~29 cm) specimens. Furthermore, the area of the surangular containing the posterior and anterior surangular foramina and the thinned bone between them matches the length and position of the surangular foramen in Raptorex (skull length ~30 cm). Therefore, B. ostromi (skull also ~30 cm long) possibly captures the precise moment of ongoing bone resorption and perforation attributable to the pneumatic diverticulum. This process probably occurred early in ontogeny, in specimens 2–3 m long, which were probably 2–3 years old at the time of death (as indicated for Ta. bataar MPC-D 107/7 by Tsuihiji et al. 2011). Apparently, around this growth stage the pneumatic diverticulum invaded the bone, and thus variability in the size, shape, and even the number of foramina is to be expected.

Juvenile Tyrannosauridae indet.

Owing to its small body size and similarity to other juvenile tyrannosaurid specimens from the Nemegt, it is likely that B. ostromi is a juvenile tyrannosaurid. We tested this hypothesis further by determining whether B. ostromi shows juvenile features that have been well documented in Ty. rex, whose ontogenetic osteological changes have been chronicled in detail by Carr (2020). We recognized that B. ostromi shows 22 juvenile and only five adult mandible features found in Ty. rex by Carr (2020). Four of the 22 juvenile features were found only in early juveniles, and the remaining 18 in late juveniles.

The mandible characters recognized both in B. ostromi and juveniles of Ty. rex are as follows: (i) size of the first three alveoli of the dentary increasing posteriorly; (ii) shallow dentary in lateral view; (iii) shallow coronoid region of the surangular; (iv) no ridge delimiting the caudoventral fossa of the angular caudal process; (v) first two dentary alveoli much smaller than the latter alveoli; (vi) eighth tooth is the mesiodistally longest in the dentary; (vii) the alveoli decreasing posteriorly in mesiodistal length from the sixth to seventh alveolus; (viii) single large pit medial to the Meckelian fossa; (ix) low angle of the ‘chin’; (x) lightly textured ‘chin’ region; (xi) distance of the ventralmost dentary foramen from the dorsal margin of the dentary to the total height of the bone > 40%; (xii) the lateral extension of the surangular shelf horizontal; (xiii) surangular shelf not slanted; (xiv) small surangular foramen; (xv) caudal extent of the coronoid process declining before it reaches the glenoid; (xvi) presence of an embayment on the caudal margin of the surangular foramen; (xvii) cleft between the caudal glenoid process dorsoventrally short and shallow; (xvii) caudal end of the surangular shelf fading below the glenoid region; (xix) lateral scar on the surangular present; (xx) caudal glenoid process as tall as the rostral process; (xxi) lateral scar on the surangular rugose and shallow; and (xxii) dorsal orientation of the anterior glenoid foramen (Carr 2020).

The prevalence of features shared by B. ostromi and juvenile Ty. rex supports the identification of B. ostromi as a young tyrannosaurid. The less numerous adult Ty. rex features (Carr 2020) found in B. ostromi are listed below, with comments regarding variability within Ta. bataar. First, the second dentary tooth is > 75% of the mesiodistal length of the third dentary tooth. The proportions between the first three dentary teeth in Ta. bataar seem to be variable. In the subadult ZPAL MgD-I/175, the second dentary tooth is < 75% of the mesiodistal length of the third dentary tooth, and in the adults (ZPAL MgD-I/5) it is between 70 and 78%. Second, the combined mesiodistal lengths of the first two alveoli of the dentary are greater than the mesiodistal length of the third alveolus, as in all examined individuals of Ta. bataar (subadults ZPAL MgD-I/45, ZPAL MgD-I/46, and ZPAL MgD-I/175; and adults ZPAL MgD-I/4 and ZPAL MgD-I/5; Table 1). However, the difference in all Ta. bataar specimens is greater (~4 cm) than in B. ostromi (1 cm). Third, there is no deviation in the ‘chin’ region, which is not recognized in any examined Ta. bataar specimen (ZPAL MgD-I/45, ZPAL MgD-I/46, ZPAL MgD-I/175, ZPAL MgD-I/4, and ZPAL MgD-I/5), nor it has been described in juvenile Ta. bataar MPC-D 107/7 (Tsuihiji et al. 2011). Fourth, the caudal surangular foramen is positioned far anteriorly to the glenoid, as in all Ta. bataar individuals (ZPAL MgD-I/4, ZPAL MgD-I/5, and ZPAL MgD-I/31), including the juvenile MPC-D 107/7 (Tsuihiji et al. 2011). Fifth, the glenoid fossa is short and deep in adult Ty. rex and B. ostromi. A shallow and long glenoid fossa can be recognized in ‘Shanshanosaurus huoyanshanensis’ see (Currie and Dong 2001), but already in the slightly larger MPC-D 107/7 and B. ostromi, as in young and adult Ta. bataar (ZPAL MgD-I/4, ZPAL MgD-I/5, and ZPAL MgD-I/31), it is narrow and deep. Those features possibly indicate some species-dependent variability, similar to the proportion of the antorbital fenestra, which does not shorten as much during the ontogeny of Ta. bataar as it does in Ty. rex (Tsuihiji et al. 2011).

As it is clear that the B. ostromi holotype belongs to a juvenile tyrannosaurid, the question becomes: can we identify which species it belonged to? We can first make comparisons with the other Nemegt tyrannosaurids: Ta. bataar and Alioramus spp. The mandible of B. ostromi is generally similar to juvenile Ta. bataar or putative juveniles of that species, like ‘S. huoyanshanensis’ and R. kriegsteini (see Currie and Dong 2001, Sereno et al. 2009, Fowler et al. 2011, Tsuihiji et al. 2011). The dentary is straight in the dorsal and ventral view, shallow, slender, and thickens and tapers dorsally at the anterior end. Bagaraatan ostromi, like MPC-D 107/7, but in contrast to Ta. bataar specimens and Alioramus altai, does not show any pneumatic pocket behind the surangular foramen (Tsuihiji et al. 2011, Brusatte et al. 2012). The cervical vertebrae of B. ostromi strongly resemble the middle or posterior cervical vertebrae of ‘S. huoyanshanensis’. They share the posterodorsal rather than dorsal inclination of the neural spines, and have less flexed centra than in adult, large tyrannosaurids (Currie and Zhiming 2001). The fusion of some bones occurred early in the ontogeny of tyrannosaurids, e.g. the juvenile Ta. bataar already has fused nasals (Tsuihiji et al. 2011). However, the articular remains unfused with the surangular in B. ostromi, similar to ‘S. huoyanshanensis’, juvenile Ta. bataar (MPC-D 107/7), and Alioramus altai (see Brusatte et al. 2012). In contrast, in Raptorex and larger Ta. bataar individuals the articular is fused to the surangular. Moreover, early in ontogeny partial fusion of the pelvis was reported in Raptorex (see Fowler et al. 2011) and young Ty. rex (BMR P2002.4.1, ‘Jane’; Parrish et al. 2013). In contrast, the pelvic bones are unfused in B. ostromi, juvenile Ta. bataar (MPC-D 107/7), and subadult Alioramus altai (see Brusatte et al. 2012) However, an early fusion of the cranial sutures might not necessarily be associated with an early fusion of postcrania, because these functional units could be subjected to developmental plasticity or separate evolutionary pressure depending on the ecology and preferred or available diet. The co-ossification of postcranial sutures and fusion between bones among tyranosaurids require further study. However, owing to their high variability, also in juveniles, we do not find them to be an adequate indicator for growth stage in tyrannosaurids.

We can more thoroughly compare B. ostromi with young juvenile Ta. bataar, because no young juveniles of Alioramus spp. are known thus far. Given that both tyrannosaurids occur in the Nemegt Formation and that B. ostromi lacks diagnostic features of either Alioramus spp. (see Brusatte et al. 2012) or Ta. bataar (see Hurum and Sabath 2003), which is mostly attributable to the fragmentary nature of the holotype skeleton, we cannot assign ZPAL MgD-I/108 to any particular species. Some subtle features suggest that B. ostromi might be a juvenile of Ta. bataar: e.g. (i) already strongly expanded anterior end of the dentary; (ii) ‘chin’ well demarcated; and (iii) lack of the pneumatic pocket next to the surangular foramen. However, because those features might potentially be a result of intraspecific variability or be more widespread among juvenile tyrannosaurids than currently suspected, we cannot clearly determine whether B. ostromi is a juvenile of Ta. bataar or Alioramus spp. Thus, we consider B. ostromi to be an indeterminate juvenile representative of the Tyrannosauridae. This assessment might be modified in the future, when more juvenile individuals of tyrannosaurid taxa are known (particularly young individuals of Alioramus spp.) and when the growth series and variability at early life stages are better understood.

Conclusions

The enigmatic theropod described by Osmólska (1996) from the Late Cretaceous Nemegt Formation of Mongolia, B. ostromi, is a chimaera of two non-avian dinosaurs. First, the femur, tibiotarsus, and pedal phalanx II-2, together with several other bones not described by Osmólska but catalogued under the same specimen number (manus phalanges, caudal vertebrae, and rib), are referred here to the Caenagnathidae. Second, the mandible, axial skeleton, pelvis, and pedal phalanx IV-1 (which together form an associated skeleton) are considered here as the holotype of B. ostromi and identified as Tyrannosauridae indet. (Fig. 25). Owing to the presence of numerous juvenile features, which are seen in young individuals of Ty. rex, and many shared similarities with juvenile Ta. bataar and a putative representative of that species (‘Shanshanosaurus’), we propose that B. ostromi is a juvenile tyrannosaurid. There are two features distinguishing B. ostromi from other tyrannosaurids: (i) the ridge separating two depressions on the lateral surface of the postacetabular process; and (ii) the presence of a double surangular foramen. The latter we interpret as a result of ongoing bone resorption owing to the entrance of the pneumatic diverticulum into the surangular. Three (or four) previously known tyrannosaurids co-occur in the Nemegt Formation: Ta. bataar, Alioramus altai, and Alioramus remotus (and maybe R. kriegsteini), and the morphology of the mandible suggests that B. ostromi is most similar to juvenile Ta. bataar. However, owing to the lack of diagnostic features in the preserved material, which would clearly indicate its taxonomic identity, and the confounding issues of individual and ontogenetic variation in tyrannosaurids, we simply conclude at the moment that Bagaraatan is a juvenile tyrannosaurid of uncertain species placement. Yet, we do not elect formally to sink B. ostromi as a name, because we have identified some potentially diagnostic features at the species level and because tyrannosaurid specimens are commonly found in the Nemegt Formation, meaning that future discoveries should help to clarify this taxonomic issue.

Acknowledgements

The presented material was collected during the Polish–Mongolian Paleontological Expeditions organized by the Mongolian and Polish academies of sciences in the late 1960s and early 1970s. This research was possible thanks to the work of all members of the expeditions, whom we would like to acknowledge. We are grateful to Khishigjav Tsogtbaatar for the possibility to work on the Tarbosaurus material housed in the Institute of Paleontology of the Mongolian Academy of Sciences. J.S.-M. thanks Sanjaadash Ulziitseren, Damidansuren Idersaikhan, Zorig Enkhtaivan, Chagnaadorj Bayardorj, and Bat-Erdene Erdenekhuyag for all their help during the research visit. The manuscript benefitted from input by Greg Funston and one anonymous reviewer. We thank the Editor for processing the manuscript. We would like to thank Krzysztof Owocki for sharing the thin section taken from the femur of ZPAL MgD-I/108 and Katarzyna Przestrzelska for the accurate preparation of the caenagnathid vertebra. We also thank Boris Morkovin and Andrey Podlesnov for sharing the photographs of Avimimus portentosusPIN 3907/1 and 3907/6. S.L.B. thanks Peter Makovicky for the discussion on Bagaraatan while doing his PhD work, and acknowledges that he also independently came to the conclusion that the material represented a chimera of multiple theropods. S.L.B. also thanks Grzegorz Niedźwiedzki, Tomasz Sulej, Jerzy Dzik, Magdalena Borsuk-Białynicka, and Zofia Kielan-Jaworowska for hosting him at the ZPAL collections over the years and for their friendship and hospitality. We thank Jakub Zalewski, who prepared the life reconstruction of Bagaraatan ostromi.

DATA AVAILABILITY

Relevant data are published alongside the paper as supplementary files. Further requests and enquires are welcome to be directed to the corresponding author.

Author contributions

J.S. conceived the study, wrote the initial manuscript, and prepared the three-dimensional models and figures; S.L.B. performed the phylogenetic analyses and revised the manuscript; T.S. helped in the taxonomical interpretation of the data, photographed the specimens, and revised the manuscript. All authors discussed the results and contributed to the final manuscript.

Conflict of interest

None declared.

Funding

This work was supported by the National Science Centre, Poland (grant no. 2019/35/B/NZ8/02292). S.L.B.’s work on Nemegt theropods stems from his PhD at Columbia University and the American Museum of Natural History, supervised by Mark Norell, whom he thanks for his friendship and mentorship over the years. S.L.B. was supported by a National Science Foundation Graduation Research Fellowship, National Science Foundation Doctoral Dissertation Improvement Grant (DEB 1110357), Columbia University, and American Museum of Natural History.

References