New Specimens of the Crested Theropod Dinosaur Elmisaurus rarus from Mongolia

New specimens of Elmisaurus rarus from the Upper Cretaceous of Mongolia (Nemegt Formation) preserve bones not previously found in “elmisaurids” that help elucidate their relationships to Leptorhynchos elegans and other oviraptorosaurs. Elmisaurus rarus and the North American Leptorhynchos elegans are known from numerous but incomplete specimens that are closely related to, but nevertheless clearly distinguished from, Chirostenotes pergracilis and Epichirostenotes curriei. These specimens include the first known cranial bone attributed to Elmisaurus, the frontal, which clearly shows this animal had a cranial crest (most of which would have been formed by the nasal bones). The first vertebrae, scapula, femora, and tibiae from Elmisaurus are also described. The Elmisaurinae can be distinguished from the Caenagnathinae by the coossification of the tarsometatarsus and smaller size at maturity. Examination of oviraptorosaur hindlimbs reveals four distinct morphotypes, possibly attributable to paleoecological differences.


Introduction
Discovered originally by the Polish-Mongolian Palaeontological Expedition of 1970, the remains of Elmisaurus rarus Osmólska, 1981 consisted mostly of hand and foot elements derived from three specimens (Osmólska 1981). Some additional skeletal elements, mostly fragmentary, were not included in the original description because they were not considered diagnostic when the original paper was written. Osmólska (1981) noted that the manual elements of Elmisaurus show strong similarities with Chirostenotes pergracilis Gilmore, 1924, and the foot elements with Macrophalangia canadensis Sternberg, 1932. This suggested to her that Chirostenotes and Macrophalangia were congeneric, a suspicion confirmed later by the discovery of a specimen in Dinosaur Provincial Park (Currie and Russell 1988).
Elmisaurus remains were subsequently recognized in North America (Currie 1989), although they were attributed to a different species ("Elmisaurus" elegans). These species, along with Chirostenotes pergracilis, were assigned (Currie 1990(Currie , 1997 to the family Elmisauridae, which Osmólska (1981) had established. The discovery in Asia of small mandibles comparable with North American caenagnathids prompted  to speculate that they were from an animal like Elmisaurus. This implied that Elmisauridae and Caenagnathidae might be synonymous. The inclusion of Chirostenotes in the Elmisauridae was shown to be incorrect when it was established that this genus was congeneric with the caenagnathid oviraptorosaur Caenagnathus (Sues 1997), which has left open the question of whether or not the family Elmisauridae is synonymous with Caenagnathidae. Osmólska, 1981  Description.-An isolated frontal (50.5 mm long) of Elmisaurus rarus was found with MPC-D 102/007 (Fig. 1). The identification is based on the presence of ventral depressions for the orbit and brain cavity, and sutures for the laterosphenoid, parietal and postorbital. The bone is less than 2 mm thick and does not appear to be pneumatized. The interfrontal suture is dorsoventrally thin posteriorly (4 mm), dorsoventrally deep (15.2 mm) anteriorly, and lightly striated longitudinally. The nasal contact is a large, deep slot between the interfrontal suture and the orbital margin; the nasal extended posteriorly to a position between the supratemporal openings. There are no obvious sutures for either the prefrontal or the lacrimal, which suggests that the large nasal contact was also at least partially filled by the lacrimal. Unfortunately, the anterolateral tip of the frontal is incomplete. The slot ("nas" on Fig. 1A) is open ventrally almost to the back of the orbit, posterior to which it becomes a depression in the dorsal surface of the frontal. The frontals formed a flat dorsal surface between the upper temporal fenestra. There is a strong suture (7.5 mm long anteroposteriorly; "par" on Fig. 1E) that overlapped the front of the parietal. In dorsal view, the suture shows that the most anterior margin of the parietal is on the midline. However, a thin process extended anterolaterally along the margin of the frontal to contact the laterosphenoid and postorbital as in most theropods; the suture is only visible in lateral and ventral views. The postorbital process of the frontal is almost perpendicular to the medial orbital margin. This is the widest part of the bone. As in other theropods, the dorsal surface of this process slopes posteroventrally into the supratemporal fenestra. There is a distinct suture posterolaterally with the postorbital bone. The ventral margin of the postorbital process has a transverse groove for the postorbital process of the laterosphenoid. Overall, the postorbital process is similar to those of dromaeosaurids. The domed, ventral surface over the brain has impressions of blood vessels, suggesting the animal was highly encephalized (Osmólska 2004).

Elmisaurus rarus
Three vertebrae were recovered with MPC-D 102/007 (Fig. 2). An anterior to mid-cervical neural arch is relatively small (21.4 mm long) and tapered, and resembles the third to fifth cervical vertebrae of dromaeosaurids and oviraptorosaurs. The second preserved vertebra is a posterior cervical or anterior dorsal that tapers ventrally into a midline keel. The 26.4 mm long centrum is pierced laterally by a single pleurocoel (sometimes termed a lateral excavation), close to the posterior margin. There is no sign of the suture between the neural arch and centrum, and the coosification suggests that the specimen was mature at the time of death. Only the base of the neural arch is preserved, but on the right side it forms the margin of a large infradiapophysial fossa.
The third vertebra of MPC-D 102/007 is an anterior dorsal that lacks the neural spine and transverse processes (Fig. 2). The 27.7 mm long centrum has a large (5 mm) pleurocoel on each side. The anterior edge of the centrum extends ventrally far below the ventral margin of the vertebra (hypapophysis), and as a consequence the centrum is 22.5 mm high anteriorly. The ventral surface of the hypapophysis is unfinished and suggests that this is a cervicodorsal vertebra, of which there are two or three in oviraptorids and alvarezsaurids. The parapophysis is positioned on the lateral surface at the anteroventral margin of the neural arch, which is consistent with the identification of this vertebra as an anterior dorsal. The neurocentral suture is fused but still discernible for most of its length.
An intricate system of laminae and ridges outline the infraprezygapophysial, infradiapophysial, and infrapostzygapophysial pneumatic fossae (Britt 1993). In addition to the highly angled zygapophyses, there are paired hyposphene and hypantrum articulations.
Two fragmentary mid-sacral vertebrae (ZPAL MgD-I/98) show that the ventral surfaces of these vertebrae are almost flat, and that there were no longitudinal furrows along the ventral surfaces of the mid-sacral centra. Such a furrow has been reported in some other maniraptorans (Currie and Russell 1988;Norell and Makovicky 1997), although it is usually present in the more posterior vertebrae. On each of the two sacrals a deep, elongate pleurocoel is present ventrolaterally as in caenagnathids (Currie and Russell 1988), oviraptorids (Balanoff and Norell 2012), tyrannosaurids (Brochu 2002), dromaeosaurs, megaraptorines, ornithomimids, therizinosaurs, and birds (PJC personal observation).
Three of the rib heads, all from the left side, are reasonably well preserved ( There is a fragment of scapula preserved in specimen ZPAL MgD-I/98 (Fig. 4). It represents a proximal portion of the bone that has an almost complete glenoid surface and most of the basal portion of the acromial process. The latter is thick across the base and inclines laterally, which suggests that its missing distal part might provide an attachment surface for the clavicular epicleideum. The supraglenoid ridge is not pronounced. Just above the margin of the glenoid and close to the coracoscapular suture, there is a small irregular depression on the lateral surface of the scapula with two to three tiny foramina.
On the ventromedial side of the scapular fragment, there is a small nutrient foramen positioned close to the glenoidal surface. The portion of the ventral margin adjoining to the glenoid is rough and thick where it forms an elongate tuberosity.
As only the proximal portion of the bone is preserved (Fig. 4A), the relationship of the scapula to the rib cage, and orientation of the glenoid, whether ventral or lateral, cannot be shown beyond doubt. In theropods with ventrally (or ventroposteriorly) facing glenoids, the scapular portion of the glenoid faces anteroventrally as in Gallimimus bullatus Osmólska, Roniewicz and Barsbold 1972, and the oviraptorids Citipati sp. nov. Clark, Norell, and Rowe, 2002 (MPC-D 100/42), Conchoraptor gracilis Barsbold, 1986 (ZPAL MgD-I/099), and Ingenia yanshini Barsbold, 1981 (MPC-D 100/30). In this specimen of Elmisaurus rarus (ZPAL MgD-I/98), the glenoid faces somewhat lateroventrally as in dromaeosaurids (Jasinoski et al. 2006), Epichirostenotes (Funston and Currie, in press) and some therizinosaurs, which clearly have more lateral orientations to their glenoids (Perle 1979;Kirkland and Wolfe 2001). The angle to the lateral surface of the blade is approximately 110-120° in Elmisaurus rarus, whereas it is around 90° in most other theropods.
In addition to the incomplete manus described by Osmólska (1981), four elements of the hand were recovered with MPC-D 102/007 (Fig. 5). Phalanx II-2 lacks the proximal end, so the total length is unknown. The shaft is 7.3 mm deep and 5.4 mm wide.
The distal width is 8.5 mm, which is less than the 10 mm width of ZPAL MgD-I/98 (Osmólska 1981). Like all theropod penultimate phalanges, the collateral ligament pits (foveae ligamentosae) are deep on both sides, and are positioned high on the medial and lateral surfaces of the distal expansion. One ungual phalanx was recovered and lacks only the distal tip. It is about 10% smaller than the ungual described for ZPAL MgD-I/98. The absence of a proximodorsal "lip" is due to breakage (Fig. 5F). The smaller size of MPC-D 102/007 plus the more open curvature, the more distally positioned flexor tubercle and the fact that it articulates well with II-2 indicate that this ungual is phalanx II-3. The ungual described for ZPAL MgD-I/98 (Osmólska 1981) is almost certainly I-2. Phalanges III-1 and III-3 are very slender elements as in other elmisaurids (Osmólska 1981;Currie 1990), and have shaft diameters that are less than two-thirds the width of phalanx II-2.
The smaller of the two phalanges has a shallow concave proximal articulation that indicates it is phalanx III-1. The longer phalanx has the high collateral ligament pits that identify it as phalanx III-3. The lateral pit is deeper than the medial one in Elmisaurus (Osmólska 1981) and other theropods, indicating that this element is from the left hand.
The preserved proximal part of the right pubis of ZPAL MgD-I/98 is associated with the adjoining portion of the ischium forming the ventral margin of the acetabulum.
The iliac process of the pubis is broken off, and the ischiadic process is reduced to a thickened lip. Judging by the preserved proximal portion of the left pubic shaft, the shaft may have been somewhat concave anteriorly.
In addition to the proximal part of the right ischium (ZPAL MgD-I/98), there are incomplete shafts of both ischia. They show that the shaft was relatively long and massive, and was only slightly flattened mediolaterally. The distal ends of the shafts are missing, and only the bases of the thin obturator processes are preserved (Fig. 4B).
The head of the left femur (ZPAL MgD-I/98) is cylindrical, somewhat higher than the greater trochanter, and separated from the latter by a broad, shallow depression ( Fig.   4). On the posterior surface of the head, there is a wide groove for the capital ligament.
The anterior (lesser) trochanter is not complete, but is prolonged into a mediolaterally flattened and anteriorly extended ridge. The preserved thickness of the trochanter is 8 mm. Its shape and orientation suggests that the anterior trochanter was a wing-like rather than a finger-like structure. On the medial side of the shaft, a short distance below the head, there is a shallow, longitudinally oval depression. It corresponds to similarly positioned depressions (or a scars) in ornithurine birds, oviraptorids and ornithomimids (Gallimimus bullatus). On the posterolateral surface of the proximal end of the femur, some distance (25-30 mm) below the upper margin of the greater trochanter, there is a large but low protuberance resembling in its position the "posterior trochanter" of Deinonychus antirrhopus (Ostrom 1976). Similar protuberances seem be present in several other theropods, including Bagaraatan ostromi (Osmólska 1996) and Gallimimus bullatus (Osmólska et al. 1972) The preserved shaft of the left femur of MPC-D 102/007 has a circumference of 70 mm. A formula (0.8685x + 0.7654, where x and y are the log transformed values of shaft circumference and femoral length respectively; n is 106, and the r 2 value is 0.9808) comparing the femur shaft circumference with femur length in 106 coelurosaurs (Currie 2003), can be used to estimate the length of the femur of MPC-D 102/007 as 233 mm.
Using transverse shaft width produces a slightly different result where the estimated value of the femur is 236 mm. Finally, comparison of femoral versus tibial lengths in oviraptorosaurs produces an estimate of 270 mm. The last comparison is tightly constrained (0.9377x + 0.229, where x and y are the log transformed values of femoral and tibial length respectively; n is 43, and the r 2 value is 0.99385). The three measurements were averaged for an estimated femur length of 246 mm. The head of the femur is badly eroded but shows a few characters of interest. Like the head of ZPAL MgD-I/98, the lesser trochanter appears to have been a tall, winglike structure as in most theropods; it does not seem to have been the closely appressed, fingerlike lesser trochanter that is seen in oviraptorids. The shaft circumference suggests that this elmisaurid weighed about 18 kg, using the Anderson et al. (1985) method. Tibiae (Fig. 6), previously undescribed for Elmisaurus, were recovered in two specimens (MPC-D 102/007, MPC-D 102/010). The average length (323 mm) of the right and left tibiae of MPC-D 102/007 suggests that this animal stood approximately 75-80 cm high at the hips. The tibia is somewhat more gracile and elongate than those of similar sized oviraptorids. For example, MPC-D 102/007 (Elmisaurus) has a femur slightly longer (an estimated length of 246 mm) than that (an estimated length of 242 mm) of MPC-D 102/011 (cf. Ingenia), but the tibia of MPC-D 102/007 is 25% longer, even though the shaft circumference is 10% less. The unfinished anterior surface of the prominent cnemial crest slopes anteroventrally from the articulation for the medial condyle of the femur (Fig. 6). The cnemial crest is separated from the outer (fibular) condyle by a deep incisura tibialis (Fig. 6D, E). The outer condyle would have contacted more than half the anteroposterior length of the proximal end of the fibula. Its lateral articular surface for the fibula is oriented posterolaterally. This surface is separated from a low ridge extending from the fibular condyle by a shallow concavity. The groove separating the outer and inner condyles is shallow, whereas it tends to be more pronounced in oviraptorids. The rugose edge for the interosseum tibiofibular ligament attachment (Fig. 6G) is more than 4 cm long and oriented anterolaterally. This attachment is 35% of the way down the shaft of the tibia from the proximal end, whereas it is 41% in closely appressed distally to the lateral edge of the anterior surface of the tibia. A more medial impression shows that the ascending process was at least 65 mm high (20% tibial length). Overall, the anatomy of the tibia is virtually the same as an oviraptorid tibia.
Whereas the tibia of MPC-D 102/007 is only 25% longer than that of an oviraptorid of equivalent size, the tarsometatarsus is relatively longer in Elmisaurus; the tarsometatarsus is 70% of femur length in MPC-D 102/007, whereas the average in 22 oviraptorid specimens is only 55% (with a maximum of 60% in the chicken-sized Yulong mini [Lu et al. 2013]).  (Fig. 7A), and to metatarsals III, IV, and V. As pointed out by Osmólska (1981), a process projects dorsally from the posterolateral corner of the bone. Unlike any other theropods, the process arches distally to meet the upper end of the fifth metatarsal (Fig. 7G).
The tarsometatarsus in MPC-D 102/006 (Fig. 8A, Table 1), is 15% longer than ZPAL MgD-I/172 but is not as completely fused. Nevertheless in MPC-D 102/006, the third and fourth distal tarsals are fused to each other, and are fused posteriorly to the second to fifth metatarsals. Their anterior margins are still distinct and had not coossified with the second to fourth metatarsals. The second to fourth metatarsals are also coossified posteriorly, but remain distinct in anterior view, which is similar to ZPAL MgD-I/172 (Osmólska 1981). The fifth metatarsal is closely appressed to the fourth but is not fused to it (Fig. 7G). The isolated fourth metatarsal (MPC-D 102/008; Fig. 8C) is 164 mm long (12% longer than ZPAL MgD-I/172) but was not found with either the distal tarsals or the other metatarsals. The posterior part of the proximal surface was damaged when the specimen was found, suggesting that it had been fused posteriorly to the rest of the tarsometatarsus but had broken away before burial and fossilization. The metatarsals of the associated specimen MPC-D 102/007 are about 15% larger than the equivalent parts of ZPAL MgD-I/172, and show no signs of fusion to each other. However, the third distal tarsal had remained associated with the head of the second metatarsal, suggesting that coossification had begun between these elements. There is therefore some variability in the onset of tarsometatarsal coossification in Elmisaurus rarus.
The metatarsus of MPC-D 102/007 (Fig. 8B) is 53% the length of the tibia, which is more elongate than in most oviraptorosaurs (43-48%), but is less than most arctometatarsalian theropods. A section across the mid-length of the metatarsus is deeply concave on the plantar side (Fig. 9). This is because metatarsals II and IV are deep The inner sides of metatarsals II and IV form steep and deep walls (lateral and medial respectively) of a plantar trough (Fig. 9), the narrow bottom of which is formed by the weakly concave surface of metatarsal III. The anterior (dorsal) surface of the metatarsus is concave along about its proximal half, but much less than the plantar surface.
Proximally, there is a short slit between the otherwise tightly connected (or fused) metatarsals III and IV (Osmólska 1981). This slit occurs in ZPAL specimens Mg-D I/127, 98 (Osmólska 1981: pls. 20: 2; 21: 1) and MPC-D 102/007 (Fig. 8B). Its position is comparable to that of the lateral proximal vascular foramen in modern birds, and Confuciusornis sanctus (Chiappe 1999); it probably transmitted a. tarsalis plantaris to the plantar aspect of the foot. On all adequately preserved metatarsi, there is a second, narrower slit located between metatarsal II and the somewhat medially incised metatarsal III (Fig. 8B). This slit is placed more distally than the former, but also cuts through the metatarsus. A similarly located slit is also present in species of Velociraptor (HO personal observation). On the anterior (dorsal) surfaces of metatarsals II-IV (Fig. 8A), there is a tripartite protuberance proximally (Osmólska 1981). A similarly placed tubercle is present on metatarsal II in an undetermined dromaeosaurid specimen, probably a species of Velociraptor (Norell and Makovicky 1997) and in Confuciusornis sanctus (Chiappe et al. 1999). These authors suggest that the tubercle probably marks the insertion of M. tibialis cranialis.
Metatarsals II, III, and IV are the same in the new specimens as they are in ZPAL MgD-I/127 (Osmólska 1981) and little needs to be added to their description. Because the elements of MPC-D 102/007 are separate, some of the contacts between the metatarsals can be seen well. Relatively small (12 mm high, 7 mm anteroposteriorly) facets on metatarsals II and IV contact each other anterior to the proximal end of the third metatarsal (Fig. 8E). The posterior surface of the third metatarsal has two longitudinal ridges (cruciate ridges) that are continuous with the posterior articular ridges. These ridges cross distally, forming a distinctive chiasmata (Fig. 10), as in Leptorhynchos elegans from Alberta (Funston et al. in press).
The fifth metatarsal extends proximally beyond the proximal surfaces of the distal tarsals. When found, it formed an ossified arch with the top of the distinctive protuberance (Osmólska 1981) of the fourth distal tarsal. Unfortunately, this fragile arched structure was damaged during collection, although the outline is still visible and it is clear that the fifth metatarsal was fused into the tarsometatarsus as in Avimimus. The fifth metatarsal is 70.3 mm long, which is 38% the length of the third. The long tapering fifth metatarsal contacts most of the margin of the ridge along the posterolateral edge of the fourth metatarsal (Fig. 8B) rather than arching away from it as it does in other theropods (Currie and Peng 1993).
Although the first metatarsal was not recovered with MPC-D 102/007, the rest of the digit from the right side was (Fig. 11). The medial collateral ligament pit is deeper than the lateral one in pedal phalanx I-1. In both size and shape, it is similar to I-1 of ZPAL MgD-I/98.
Discussion.-Assignment of Elmisaurus rarus to Maniraptora is strongly supported by manus structure (Currie 1990). However, elmisaurines differ from other maniraptorans (except Chirostenotes pergracilis) in their slender and elongate first metacarpals, which are short and rather massive in dromaeosaurids, oviraptorids and troodontids (Table 2).
Within the third digit, phalanx III-3 is distinctly longer than the more proximal phalanges, and, peculiarly, phalanx III-1 is the most gracile. The manus in Elmisaurus rarus and Chirostenotes pergracilis are similar in size, and are practically identical in relative proportions (Table 3). The only differences between these taxa concern the more slender first digit and the shapes of the unguals in Chirostenotes pergracilis (Zanno and Sampson 2005: fig. 5a-c), which are more slender, less curved, and have much stronger dorsoproximal lips.
There are also some features in the pes that are exceptional amongst non-avian maniraptorans; the plantar surface of the metatarsus is deeply concave, which results in an arched cross-section in the middle of the metatarsus (Fig. 9). Among other unique features, the tightly attached (or fused) metatarsals III and IV, the large, compound proximal protuberance on the posterior (plantar) surfaces of metatarsals II-IV, and the proximal tuber formed by the fourth distal tarsal and the fifth metatarsal should be mentioned.
A suite of manual and pedal characters ties the elmisaurines to Caenagnathidae, including the elongate metacarpal I, the proximodorsal "lip" on the unguals, and the flat, proximally pinched metatarsal III. Despite these similarities, Elmisaurus rarus and Leptorhynchos elegans share a number of features that unite them in the Elmisaurinae (Funston et al. in press). The paired cruciate ridges on the posterior surface of metatarsal III (Fig. 10), the fusion of the distal tarsals to each other, and their coossification with metatarsals II, III, and IV, are unique to elmisaurines. Sues (1997) suggested that material assigned to Elmisaurus by Currie (1989) was probably synonymous with Chirostenotes.
The abundance of additional material described here shows that Elmisaurus and Based on foot and leg structure, Oviraptorosauria can be divided up into four major clades (Lamanna et al. 2014). Caudipteryx has elongate arctometatarsalian feet showing no fusion of the calcaneum with astragalus, or distal tarsals with metatarsals.
Avimimus, recovered as the sister taxon to Caenagnathoidea sensu Lamanna et al. (2014), has an astragalocalcaneum fused to the tibia, and the distal tarsals fused to the arctometatarsalian metatarsus (Kurzanov 1981). Caenagnathids have a deep depression on the medial surface of the proximal end of the fibula, fused astragalocalcaneum, elongate arctometatarsalian feet and in the case of elmisaurines at least, the coossified distal tarsals and metatarsals. Oviraptorids lack the depression on the medial surface of the proximal end of the fibula, retain separate astragali and calcanei, and have relatively shorter feet that are not arctometatarsalian. Within each of these groups, proportions of the hindlimb track the changes in morphology. In spite of the elongate, arctometatarsalian nature of the foot in caenagnathids, the tibia is double the length of the metatarsus, suggesting a cursorial lifestyle. In both Avimimus and caenagnathids, elongation of the hindlimb is accompanied by some degree of fusion. In contrast, oviraptorids have short tibiae compared to the femora and a short metatarsi compared to the tibiae, with no fusion of hindlimb elements. This, combined with the non-arctometatarsalian metatarsus, argues against extensive cursoriality in oviraptorids.
It is possible that the morphology of the hindlimb reflects the dietary adaptations of each of these clades. There are some indications of herbivory in members of the Oviraptorosauria, such as the gastroliths in Caudipteryx (Ji et al. 1998) and stable isotopes in oviraptorid eggshell (Montanari et al. 2013). Caenagnathids are likely omnivorous (Funston and Currie 2014), and may have retained long legs and elongate hands for prey capture. In contrast, oviraptorids, which may have engaged in crushing (Barsbold 1977(Barsbold , 1983 or strict herbivory (Smith 1992;Montanari et al. 2013), have reduced distal limbs. Longrich et al. (2013) assigned three partial mandibles to Leptorhynchos, based on size. These mandibles are similar to those of Chirostenotes pergracilis (sensu Longrich et al. 2013;Funston and Currie 2014) and would be equally capable of an omnivorous diet. Although there are numerous problems with these assignments (Funston et al. in press), if the assignment of mandibular material to Leptorhynchos elegans is correct, then it is reasonable to assume that Elmisaurus rarus possessed a similar mandible. In this case, the elongation of the hindlimbs in elmisaurines may be the result of ecological pressures for prey capture and herbivory. Institute. This paper is dedicated to the memories of Halszka Osmólska and Demchig Badamgarav, two great women who made significant contributions to this study.