A new two-fingered dinosaur sheds light on the radiation of Oviraptorosauria

Late Cretaceous trends in Asian dinosaur diversity are poorly understood, but recent discoveries have documented a radiation of oviraptorosaur theropods in China and Mongolia. However, little work has addressed the factors that facilitated this diversification. A new oviraptorid from the Late Cretaceous of Mongolia sheds light on the evolution of the forelimb, which appears to have played a role in the radiation of oviraptorosaurs. Surprisingly, the reduced arm has only two functional digits, highlighting a previously unrecognized occurrence of digit loss in theropods. Phylogenetic analysis shows that the onset of this reduction coincides with the radiation of heyuannine oviraptorids, following dispersal from southern China into the Gobi region. This suggests expansion into a new niche in the Gobi region, which relied less on the elongate, grasping forelimbs inherited by oviraptorosaurs. Variation in forelimb length and manus morphology provides another example of niche partitioning in oviraptorosaurs, which may have made possible their incredible diversity in the latest Cretaceous of Asia.


Expanded Diagnosis 24
Oksoko avarsan can be distinguished from citipatiine oviraptorids by the enlarged 25 first manual digit and reduced second and third manual digits. It can be distinguished 26 from most heyuanniine oviraptorids by the presence of a cranial crest (Fig. S1). Two 27 heyuanniines are known which possess a cranial crest: Nemegtomaia barsboldi and Banji 28 long. In both of these taxa, the cranial crest is composed primarily of the nasals and fibers (Fig. S6a). On the medial side, there is primary osteonal bone towards the 70 endosteal surface and avascular parallel-fibred bone near the periosteal surface. The 71 medullary cavity and endosteal lamellae are both well developed. Like the femur, there is 72 an annulus developed towards the periosteal surface that probably represents the first 73 growth mark. Based on these data, MPC-D 102/11 is best regarded as a young juvenile 74 approximately one year of age. 75 MPC-D 102/110.a-The fibula ( Supplementary Fig. S5) is composed mostly of 76 primary fibrolamellar bone, but there are some secondary osteons on the medial side. 77 There is a high proportion of woven bone with dense, plump osteocyte lacunae (Fig. 78 S6b). The medullary cavity is small but has defined edges and in some places, there are 79 endosteal lamellae. There are no obvious zones of parallel-fibred bone or lines of arrested 80 growth. Towards the periosteal surface on the anterior side of the fibula there is a region 81 of reduced vasculature, but no obvious annulus is present and this zone cannot be traced 82 around the entire cortex. This may be incipient development of the annulus recorded in 83 the other individuals. It is unlikely that a growth mark has been obscured by secondary 84 remodeling because secondary osteons are sparse and primary bone is visible between 85 them. However, the sampling location may have affected the preservation of growth 86 marks. The predominance of primary bone and absence of a growth mark suggest this 87 individual is a young juvenile, but no age assessment can be made. both have similar histological texture. Neither fragment shows the medullary cavity, but 90 this may be because the samples were taken distally. Both are composed predominantly 91 of primary fibrolamellar bone, but each has some secondary osteons endosteally. 92 Vasculature is mostly longitudinal, but there are some small zones of reticular canals. 93 Near the periosteal surface there is a zone of avascular, parallel-fibred bone similar to that 94 in MPC-D 102/11. It likely represents the first growth mark, although no distinct rest line 95 is visible. Like MPC-D 102/11, this specimen was likely a young juvenile approximately 96 one year of age. 97 MPC-D 102/12-The fibula (Fig. S5) and a fragment of the femur were 98 sectioned. The fibula has more secondary remodeling, and therefore some of the growth 99 record has been erased (Fig. S6c). The medullary cavity is well formed and lined by 100 multiple generations of endosteal lamellae. Several other erosive cavities also excavate 101 the cortex; these are separated by trabeculae or endosteal lamellae. Vasculature is 102 longitudinally oriented, and towards the medial surface these canals are arranged into 103 circumferential rows. At least three lines of arrested growth are recorded on the medial 104 surface, but it is likely that more have been obscured by secondary remodeling and 105 expansion of the medullary cavity. 106 The femur exhibits less secondary remodeling, all of which is concentrated in a 107 vertical column extending perpendicular to the periosteal surface (Fig. S6d). The primary 108 bone in this column contains larger, more densely packed osteocyte lacunae, and well-109 developed Sharpey's fibers towards the periosteal surface (Fig. S6d). Accordingly, it 110 likely represents a zone of muscle insertion. A small strip of endosteal lamellae marks the 111 edge of the medullary cavity, indicating that the entire cortex is preserved. Vasculature 112 changes from predominantly reticular endosteally through plexiform towards a zone of 113 parallel-fibred bone about halfway through the cortex. A faint cement line is visible 114 within this zone of parallel-fibred bone, indicating that it represents a growth mark in the 115 form of an annulus. External to this annulus, vasculature is laminar and decreases towards 116 the periosteal surface. At least four more annuli are visible in the external part of the 117 cortex, and have decreased spacing periosteally. Whereas a zone of fibrolamellar osteonal 118 bone separates the first, second, and third annuli, the spaces between the third, fourth, and 119 fifth annuli are composed of parallel-fibred bone. This forms a continuous band of 120 parallel-fibred bone on the periosteal surface of the cortex. Although no distinct lines of 121 arrested growth can be distinguished in this area, this is likely the result of the light 122 colour of the bone and the reduced thickness of the slide. In any case, the dominance of 123 parallel-fibred bone at the periosteal surface indicates that this individual was growing 124 slowly 6-10 . The presence of five annuli and the low growth rate suggest that this 125 individual was an adult at least five years old and was approaching maximum body size. 126 127 128

Expanded Statistical Methods 129
To trace the patterns of digit and limb reduction in oviraptorids, a proxy for digit 130 size was developed. We follow the conventional nomenclature for manual digits of digits 131 I, II, and III, because those are the phenotypes expressed, although we realize these may 132 be homologous to digits II, III, and IV 11 . Measurements for 73 complete digits (digit I, n 133 = 32; digit II, n = 29; digit III, n = 12) of 31 oviraptorosaur specimens (15 taxa) showed 134 that straight-line length of the ungual was strongly correlated to the pre-ungual length of 135 that digit (0.81<R 2 >0.88; Fig. S7). Therefore, the ratio of two ungual lengths could be 136 used to determine the relative proportions of two digits, even where those digits were not 137 complete. Based on these correlations, the ratio of ungual III-4 to I-2 was used because it 138 maximized the availability of data: 23 of 39 oviraptorosaur taxa (totaling 29 specimens) 139 have both ungual I-2 and III-4, whereas only 9 of more than 100 measured oviraptorosaur 140 specimens have complete first and third digits. To improve estimation of the root 141 condition for maximum likelihood analysis, additional outgroups representing a broader 142 array of coelurosaurs were grafted to the majority-rules phylogeny (following the 143 topology of Hendrickx et al. 12 ) to improve estimation of the root condition (Figs. S8, 9).

144
Archaeopteryx was removed as an outgroup because its forelimb and manual proportions 145 are unlikely to be representative of the basal coelurosaurian condition. This was 146 necessary because the forelimbs of Incisivosaurus are not known, which overweighed the 147 influence of Archaeopteryx on the root condition. 148 Forelimb length for oviraptorosaurs was calculated as the sum of the length of the 149 humerus, ulna, and metacarpal II, but not including the lengths of the digits. Forelimb 150 length was divided by femoral length and mapped as a continuous character in the same 151 way as the ratio of ungual III-4 to I-2. Allometry of the forelimb in oviraptorosaurs was 152 tested using bivariate plots against log-transformed femur length (Main Text: Fig. 5).

153
In addition to mapping the ratio of ungual III-4 to I-2 on a phylogeny, the relative 154 proportions of the digits for 17 oviraptorosaurs (cyan) with complete hands or scaled 155 composites and 67 theropods (grey) were plotted in a ternary plot (Main Text: Fig. 5).

156
Scaled composites were created by scaling overlapping elements to the same length to 157 estimate missing measurements where a missing element is known from another side or 158 another specimen. Composites were created in this way for Chirostenotes, Nemegtomaia, 159 and Oviraptor. Digit length, including the ungual, was calculated from straight-line 160 measurements of the long axes of each phalanx. The length of each digit was divided by 161 the sum of the lengths of all digits and plotted as a percentage. 162 The history of oviraptorosaur biogeography was stochastically simulated based on 163 the tip data for 1000 replicates using the make.simmap function of phytools v0.6-44, 164 using a continuous-time reversible Markov model with equal rates of transformation. 165 This allowed for estimation of ancestral biogeographic states and the posterior probability 166 of each state at each node. This model is necessarily simplistic, because of the limited 167 information on the stratigraphy and geographic ranges of each taxon. In the future, it may 168 be possible to increase the precision of the Markov model by scaling transformation rates 169 based on biogeographic information, but this was not feasible for the present study. The 170 results of the analysis were confirmed by S-DIVA analysis in RASP, which allows nodes 171 and tips to be present in more than one area. "Macrophalangia canadensis" and "Caenagnathus sternbergi" were subsumed into 180 Chirostenotes pergracilis, which was updated based on new specimens 1 . Characters 181 scored for "Alberta Dentary Morph 3" were added to Citipes elegans, and the former 182 operational taxonomic unit (OTU) was removed. Ojoraptorsaurus boerei was also 183 removed because it could be coded for relatively few characters (approximately 1%).

184
Leptorhynchos gaddisi was removed because it is represented only by the mandible and is 185 therefore provides little data. Oksoko avarsan was added to the matrix and could be 186 coded for nearly all of the characters (97.5%). Despite their early ontogenetic stages, 187 Banji long, Microvenator celer, and Yulong mini were included in the analysis. Juvenile 188 OTUs are typically recovered more basal than they should be, but this can be corrected 189 by not scoring ontogenetically variable characters for these OTUs 16 . Although 190 caenagnathid ontogeny is more poorly known, analyses of oviraptorid ontogeny (G.F.F., 191 P.J.C. pers. obs.) indicate that few-if any-characters in the matrix are ontogenetically 192 variable. Accordingly, these juvenile OTUs are likely to provide at least some important 193 information.

194
Several characters were dropped from previous analyses because they were Funston and Currie 15 , because these were poorly constructed and overweighted the 201 influence of the manual phalanges and metatarsus. These characters were initially 202 introduced to provide more information on the relationships of caenagnathids, but 203 improved sampling of the taxa accomplished the same task more rigorously. In addition 204 to removing characters, all of the characters were treated as unordered. Previously 205 ordered characters are highlighted in the Character List, but because they applied only to 206 a few taxa, they served to force relationships artificially and were removed for this 207 analysis. 208 Three outgroups were included in the anaylsis: Herrerasaurus ischigualastensis, 209 Velociraptor mongoliensis, and Archaeopteryx lithographica. Besides being the taxa 210 already used by the matrix of Osmólska et al. 13 , these outgroups are appropriate because 211 they polarize primitive characters within both Saurischia and Coelurosauria. 212 Herrerasaurus is typically recovered as a basal saurischian or theropod, and therefore 213 provides information on the ancestral characters of all theropods. Velociraptor and 214 Archaeopteryx were both included in the matrix of Maryañska et al. 17 and Osmólska et 215 al. 13 to test whether oviraptorosaurs were more closely related to birds than other 216 theropods. Nonetheless, they each provide an appropriate polarization for paravian 217 characters. 218 The resulting matrix had 42 taxa and 246 characters, which were a mix of binary 219 and multistate characters. All multistate characters were treated as unordered. The matrix 220 has a relatively high proportion of missing data: 51.7% of the characters could not be 221 coded. As expected, much of this poor data quality comes from the caenagnathid portion 222 of the tree, where only about one-third of the characters could be coded (68.7% missing 223 data). This likely accounts for their volatility in previous analyses, typically resulting in a 224 13-tomy of caenagnathids more derived than Gigantoraptor erlianensis. Regardless, this 225 is still a drastic improvement over previous analyses-for example, the caenagnathid 226 portion of the matrix of Lü et al. 2 has 80.6% missing data. Oviraptorids, in contrast, could 227 be coded for more than half of the characters (46.0% missing data) and have been 228 relatively stable in most analyses. However, some analyses have differed in the 229 membership of each subfamily, and therefore better resolution in the oviraptorid portion 230 of the tree is still important. 231 Tree search-A parsimony-based heuristic tree-search was run in TNT v.1.1 232 using 10000 replications of Wagner trees followed by tree bisection-reconnection (TBR) 233 branch swapping, holding up to 10 trees each replication. A final round of TBR branch 234 swapping was used on the resulting trees to find additional most parsimonious trees. 235 Bremer support values were calculated using the Bremer.run package included with TNT 236 v1.1. The analysis produced nine most parsimonious trees of 641 steps, with relatively 237 strong Bremer support for the major clades of Oviraptorosauria. The strict consensus tree 238 (CI: 0.41, RI: 0.63) has a polytomy at the base of Oviraptorosauria within Caudipteridae, 239 and a second polytomy within Oviraptoridae between non-heyuannine oviraptorids. 240 Otherwise, the phylogeny is completely resolved. The majority-rules consensus tree (CI: 241 0.43, RI: 0.66) is completely resolved and was therefore used for the subsequent analyses. 242 Statistical methods-The phylogeny was time-scaled using age ranges published 243 from the literature. Although the ages of most taxa could be determined relatively 244 precisely, the stratigraphic ranges of oviraptorids from southern China are poorly 245 constrained. In these cases, stratigraphic ranges were taken from published estimates of 246 the ages of the formations where the specimens were found. Time-scaling was done using 247 the paleotree v3.3.0 package in R statistical package. Branch lengths were calculated 248 using the equal dating method of Brusatte et al. 18 .

250
Results 251 The cladistic analysis produced nine most parsimonious trees of 641 steps (all 252 trees are available in the included Nexus file). The strict consensus tree has polytomies 253 near the base of Oviraptorosauria and within oviraptorids, but is otherwise well resolved 254 (Fig. S10). The majority-rules tree is completely dichotomous (Figs. S8, 9), and this exact 255 topology is also recovered among the most parsimonious trees (Tree 3 of the 256 supplementary Nexus file). Although the majority-rules tree does not inherently add or 257 resolve information to the analysis, most downstream analyses require fully dichotomous 258 trees. This topology was chosen for subsequent analyses, rather than randomly resolving 259 dichotomies, because this topology is among the most parsimonious trees. In other words, 260 it is a well-supported choice for a preferred topology. 261 Incisivosaurus gauthieri is recovered as the earliest-diverging oviraptorosaur, 262 followed by a paraphyletic Caudipteridae where Similicaudipteryx yixianensis is an 263 outgroup to the sister taxa Caudipteryx dongi and Caudipteryx zoui. These species of 264 Caudipteryx are sisters to a well-supported (decay index 2) group of Avimimus and 265 Caenagnathoidea, referred to here as Edentoraptora based on the ubiquitous absence of 266 teeth in these animals. The species of Avimimus are recovered as sister taxa to each other 267 and together are sister to Caenagnathoidea. Caenagnathoidea is well-supported (decay 268 index 3) and is divided at its base into Caenagnathidae and Oviraptoridae. 269 The most primitive caenagnathid is Microvenator celer, followed by 270 Gigantoraptor erlianensis, which is sister to the so-called 'derived caenagnathids'. These 271 taxa are characterized by relatively small size and complexly textured occlusal surfaces 272 of the fused dentaries. This group is relatively well-supported (decay index 2), and 273 therefore referred to here as Caenagnathinae for clarity. To date, four specimens totaling seven individuals of Oksoko avarsan are known. 347 The first known specimen of this taxon (MPC-D 100/33) was collected by the 1974 348 Soviet involved stabilizing the poorly-made field jacket, which was too thin and included plastic 371 shopping bags as a separating layer. The fragile nature of the specimen and the 372 association of the skeletons means the specimens were not disarticulated. Some photos 373 were taken by the last author during preparation in 2007, which confirm that the 374 specimens were not artificially associated (Fig. S11). Additional preparation and 375 disarticulation of the left hand of MPC-D 102/110-A was carried out by the lead author. 376 All three hands preserved in the block were fully encased in matrix when prepared, and 377 therefore the phalanges of the third digit could not have gone missing through erosion or 378 during collection. The preservation of delicate structures like the scleral ossicles and 379 complete articulation of the skeletons suggests that the distal phalanges are not absent 380 because of disarticulation or transport before burial. The provenance of MPC-D 102/11 381 and MPC-D 102/110 were not recorded when the specimens were returned to the MPC, 382 but were determined using geochemical approaches 24 (see section on 'Provenance').

383
MPC-D 102/11 shares many taphonomic features with MPC-D 102/110, 384 including the crouched posture (Fig. S12), the nature of the remaining matrix, and the 385 fine quality of preservation. Furthermore, it is unlikely that they were returned on the 386 same date coincidentally. The specimens were probably excavated by and confiscated 387 from the same poachers, and were likely collected from the same deposit or, at least, 388 nearby sites. It is likely that the extraordinary postures of both specimens are the result of 389 Furthermore, it differs from Heyuannia yanshini in that the third metacarpal is shorter 412 than the second metacarpal, and is reduced in width. All of these features are identical to 413 the states in Oksoko avarsan, and its referral to this taxon reduces variation in Heyuannia 414 yanshini. Furthermore, their analysis suggested that the specimen was from the Nemegt locality 453 with 90% certainty, although they did not include the western Nemegt Basin sites Bügiin 454 Tsav and Guriliin Tsav in their analysis. Regardless, these data strongly support the 455 Nemegt Formation provenance of MPC-D 102/110. 456 These conclusions are also supported by the specimens of known provenance, 457 MPC-D 100/33 and MPC-D 102/12. These specimens are from Bügiin Tsav and Guriliin 458 Tsav, respectively, and confirm that Oksoko was from the Nemegt Formation. 459 Provenance of the poached specimens could be confirmed by future relocation of the 460 quarry. clearly resting on a hard surface, rather than being trapped in soft sediment. The cause of 478 death cannot be directly ascertained, but the tight huddle of the specimens may point to 479 exposure as a possible cause. Regardless, the cause of death must have been non-violent, 480 because the specimens were resting when killed and subsequently buried. 481 482 9.  premaxilla (1). 520 16. Accessory maxillary fenestrae: absent (0); at least one accessory fenestra present (1). 521 17. Nasal along midline: longer than frontal (0); shorter than or as long as the frontal 522 (1). 523 18. Nasals: separate (0); fused (1). 524 19. Subnarial process of the nasal: long (0); short (1). 525 20. Shape of the narial opening: longitudinally oval (0); teardrop-shaped, slightly longer 526 than wide (1); much longer than wide (2). (ORDERED) 527 21. Nasal recesses: absent (0); present (1).