Retrodeformation and muscular reconstruction of ornithomimosaurian dinosaur crania

Specimens

Few well preserved, three-dimensional ornithomimosaur skulls are known. Here we focus on crania from three taxa: Garudimimus brevipes, Struthiomimus altus and Ornithomimus edmontonicus. Garudimimus is known from only a single specimen. Our chosen specimens of S. altus and O. edmontonicus represent the best prepared material for either taxon. There are other cranial remains, but most are badly crushed, encased within matrix prohibiting detailed observation, or remain taxonomically contentious. A number of specimens were examined first hand (see Appendix S1) and information from the published literature on the well preserved skulls of Gallimimus (Osmólska, Roniewicz & Barsbold, 1972), Deinocheirus (Lee et al., 2014), and Sinornithomimus (Kobayashi & Lü, 2003), was used for comparison where possible and inform on the retrodeformation process.

The specimen of Garudimimus brevipes (GIN 100/13, described by Barsbold (1981) and Kobayashi & Barsbold (2005)) was scanned at the University of Texas using a P250D scanner at 419 kV, 1.8 mA, aluminum filter, slice thickness = 0.5 mm, total slices = 517. The Ornithomimus edmontonicus specimen (RTMP 1995.110.0001) was scanned along the coronal axis for a total of 420 slices (0.63 mm thickness) with a GE LightSpeed Plus CT scanner (Tahara & Larsson, 2011). The Struthiomimus altus specimen (RTMP 1990.026.0001) was scanned using the same parameters as the Ornithomimus specimen, creating a dataset of 416 slices along the coronal axis. For both Ornithomimus and Struthiomimus the scans are of relatively low quality. To provide better detail, the scans were upsampled in Avizo 7.0 (FEI Visualization Sciences Group, USA). This process creates interpolations between each of the original CT slices to provide twice the number of slices in every axis for smoother reconstructions, but not providing any further resolution. The Garudimimus CT dataset was not resampled.

Reconstructions

The CT datasets were loaded into the visualisation and analysis package Avizo 7.0. Segmentation and isolation of each individual cranial bone was performed, as far as the deformed, and in some places incomplete, datasets permitted. As all of the specimens suffered deformation, it was necessary to undertake retrodeformation to provide a complete undeformed skull for each species on which the soft tissue reconstructions could be based. Notably, the nature and magnitude of deformation differed in each taxon, and hence specimen-specific retrodeformation processes were applied to each specimen. Furthermore, there is no known undistorted skull for any of the taxa studied. The process of deformation was therefore informed by the topographic relationships of the individual cranial elements in the 3D dataset, evidence of breakage and cracks revealed from direct observation of specimens and the CT scan data, and information gathered from related ornithomimosaur material from museum collections and the literature (as outlined above, and see Appendix S1). Where possible, a set of criteria were employed to perform and constrain the process. As outlined in Arbour & Currie (2012), the shape of the orbit was used a proxy to determine the degree of deformation. Orbital retrodeformation was therefore employed to reconstruct the arrangement of the surrounding facial bones. In all studied ornithomimosaurs, both actual specimens and literature study, the pattern of breakage and deformation to the bones of the orbital region suggest that the orbits in undeformed taxa should be approximately circular. As such, this was the first correction applied to the Garudimimus and Struthiomimus skulls. In Garudimimus, the individual bones were segmented from the CT scan datasets and the bones surrounding the orbit were rotated into position using the editing tools in Avizo (sensu Lautenschlager et al., 2013; Button, Rayfield & Barrett, 2014). This process was sequentially repeated with bones further from the orbits, until all of the right side of the specimen was reconstructed with the original material (no obvious plastic deformation was seen in this specimen except for the posterior margin of the maxilla). In the Struthiomimus skull, the orbital region was dorsally shifted by translating the bones within the “edit label field” function in Avizo until a circular orbit was restored. This process was continued anteriorly and posteriorly until a smooth cranial roof was created. The orbit was then measured in anteroposterior and dorsoventral axes with the “measure” tool within Avizo to check whether a near circular structure has been achieved via the retrodeformation process.

In ankylosaur skulls, it was noted that the bones of the palate suffered little deformation (Arbour & Currie, 2012). This was also true for the specimens studied here, although the palatines and pterygoids in Ornithomimus were mediolaterally displaced and overlapped. As such, palatal morphology and width were used as a marker to determine the mediolateral dimensions and required expansion of the skulls. For Ornithomimus the palatal bones were separated and aligned, and the remainder of the skull expanded mediolaterally to fit the palate. The palatal morphology of observed and well preserved specimens in the literature was used to inform on this procedure.

For the remaining bones it was possible to determine whether cortical bone had collapsed or was damaged using the CT scan data, so that the surface topography of the bones could be reconstructed using the paintbrush region-selecting tool within Avizo to match that seen in other specimens or ornithomimosaur taxa (e.g., the jugals in Ornithomimus). In some places bone was so badly damaged that full reconstruction required material from the other scans and digital manipulation using the paintbrush tools to create “new bone”. This was always informed by the individuals studied here as well as other specimens and taxa from museum collections and the literature. For example, in the anterior portion of the jugal in Garudimimus, the bone is broken and partially missing, but should overlap the posterior ramus of the maxilla and contact the lacrimal. The maxillary ramus was therefore ventrally displaced to bring it into alignment with the preserved remains of the jugal (as described for the orbit of Struthiomimus), and the jugal was extended using the paintbrush tool in three dimensions to provide the required contact whilst maintaining the shape seen in the other scanned and observed ornithomimosaurs. In Garudimimus the right side of the skull was better preserved than the left, whilst the opposite was true in Struthiomimus. Bones of the better preserved sides, once aligned and reconstructed, were mirrored about the sagittal midline of the skull, using the mirror function in Avizo (Lautenschlager, 2013).

Ornithomimosaur myology

Following methods of Holliday (2009), Lautenschlager (2013) and Button, Rayfield & Barrett (2014), the individual insertions and origination sites for the adductor muscles were digitally mapped onto the 3D ornithomimosaur skull reconstructions. Where there was a lack of osteological correlates on the bones in either the CT scans or the actual specimens, phylogenetic bracketing was used to ascertain likely insertion and origination locations. These originations and insertions were demarcated on the skull and mandible. As there was no scanned Garudimimus jaw, the Struthiomimus jaw was used (scaled and rotated into place) to ascertain muscle orientation as it was the closest in morphology of the two ornithomimids.

For each of the individual muscles, a number of simple rods were used to connect the limits of the origins and insertions (following Curtis et al., 2008). This process was used to assess the margins of the muscles and ensure there was no overlap with either the bone or other muscle bodies. In places, these rods were manually wrapped around the bones within Avizo. In other reconstructions, the neurovascular system also has been used (Lautenschlager, 2013), but its canals were not readily traceable in-silico from our lower quality CT scans. In museum specimens with matrix-obscured neurocrania, these canals were not visible either. Muscles were fully ‘fleshed’ by connecting all of the rods belonging to the same muscles until they were all merged to form a single “muscle” (Lautenschlager, 2013; Button, Rayfield & Barrett, 2014). This process was repeated for all adductor muscles. All of the fleshed out muscles were then enlarged until they occupied the maximum amount of space within the chambers without intersecting in three-dimensional space, which Avizo can prevent. The expanded muscle bodies were then digitally smoothed using tools in Avizo.

Muscle forces were estimated using the dry skull method (Thomason, 1991), where force (F_mus) equals the cross-sectional area (CSA) multiplied by the isometric muscle stress (σ here taken as 0.3 N mm⁻²: Weijs & Hillen, 1985; Thomason, 1991): ${\displaystyle F_{\mathrm{mus}}=\mathrm{CSA}\times \sigma .}$ The CSA is calculated in Avizo, using the ‘clipping plane’ tool to define the cross section and the ‘material statistics’ module which calculates the surface area. This was done for each muscle at its widest location to give the maximum CSA and thus maximum estimated force. As this method fails to take into account pennation angle of muscle fibres, the forces were multiplied by a scale factor (calculated from experimental comparisons between modelled and actual data (Thomason, 1991)) of 1.5 to compensate. Given the arrangement of muscle bodies, the total muscle force is the resultant of anteroposterior, dorsoventral and mediolateral force components. Mediolaterally orientated muscle force has limited influence on jaw closing due to the almost vertical orientation of the muscle lines of action. As such the dorsoventral component is studied for bite force lever mechanics (as in Lautenschlager, 2013). The force of each muscle (F_mus: Table 2) can be multiplied by the perpendicular distance of the muscle centroid from the jaw joint (measured in Avizo) to provide a muscle moment: ${\displaystyle F_{\mathrm{in}}=F_{\mathrm{mus}}\times \text{perpendicular distance from joint}.}$ The sum of each of the muscle input moments can then be used to calculate bite forces (F_bf) at individual locations along the skull (Table 2): ${\displaystyle \text{Total}{F}_{\mathrm{in}}=\Sigma \left(F_{\mathrm{mus}}\times \text{perpendicular distance from joint}\right)}$ ${\displaystyle \mathrm{Total}{F}_{\mathrm{in}}=\mathrm{Total}{F}_{\mathrm{out}}}$ ${\displaystyle \mathrm{Total}{F}_{\mathrm{out}}=F_{\mathrm{bf}}\times \text{perpendicular distance from joint}.}$

Rhamphothecae

Foramina are regularly cited as evidence for a keratinous rhamphotheca (e.g., Kobayashi & Lü, 2003). In modern birds, foramina can be found on the surface of the anterior premaxilla and mandible, where the rhamphothecae may be expected to be thickest (Fig. 1) (see Morhardt, 2009). In extant palaeognaths, the beak provides a close sheath over the bones of the mandible and skull (Davies, 2003), whereas in neognaths the rhamphotheca extends well beyond the oral margins. In many species, the beak also extends well beyond the anterior margins of the bone; in extreme examples such as hornbills and toucans, the rhamphotheca may be two to three times longer than the amount of bone it covers (Seki, Bodde & Meyers, 2010).

In non-avian theropods, the picture is more complicated. Ornithomimosaurs, oviraptorids, therizinosaurs, and Limusaurus (a ceratosaur) underwent tooth loss leading to partial edentulism and inferences of rhamphothecae (Zanno et al., 2009; Zanno & Makovicky, 2011). These taxa bear regular foramina across the lateral surface of edentulous regions of the premaxilla and dentary. There are also grooves on the mandible of Erlikosaurus (a therizinosaur) that appear to demarcate a keratinous rhamphotheca/beak (Lautenschlager, 2013; Lautenschlager et al., 2013). However, neurovascular foramina are also present in large theropods (e.g., tyrannosaurs: Fig. 1; spinosaurs: Dal Sasso et al., 2005; Morhardt, 2009) where teeth are present and keratinous beaks are not inferred.

As the presence of foramina is not a reliable characteristic for modelling rhamphothecae, we must rely on other lines of evidence. Because ornithomimosaurs (and other edentulous theropods) had downturned dentaries, the jaws do not occlude across the entire oral margin (Zanno et al., 2009; Zanno & Makovicky, 2011). As this would limit the functionality of the jaws, it is reasonable to expect the rhamphotheca to fill the gap to form an occlusal surface. Preserved rhamphothecae also exist on two ornithomimid specimens. In Ornithomimus RTMP 1995.110.0001 (the specimen used in this analysis) the rhamphotheca is around 4.30 mm in dorsoventral depth on both the upper and lower jaws. This is similar to a remnant of rhamphotheca approximately 3.0 mm depth on the Gallimimus specimen (GIN 100/1133) (measured from Norell, Makovicky & Currie, 2001). Assuming that the jaws occluded along their oral margins, the rhamphotheca was modelled here in all taxa to fill the oral margins, deeper at the anterior (using the preserved specimens as indicating a minimum dorsoventral thickness) and tapering posteriorly (as in modern birds). Two reconstructions accommodated uncertainty about the extent of the rhamphotheca beyond the oral margins, and two morphologies were made for the skull. These include: a conservative, ‘small’ beak model that is modelled on an ostrich beak, with limited extension of the rhamphotheca around the nares; and a more extensive ‘big’ beak model where the beak margins border the antorbital fossa. In neornithines, a naricorn rhamphothecal plate covers variable extents of the nares depending on the species (Hieronymus & Witmer, 2010), and we have taken a conservative approach by not covering any of the nares. In addition, we have not covered any of the antorbital fossa similar to the practice of Lautenschlager et al. (2013) (Fig. 7), who did however partially cover the larger nares of Erlikosaurus (Lautenschlager et al., 2013). As the lower jaw was not used in any functional studies, beaks were not reconstructed for the mandibles.

Myology

The reconstructions do not find any major differences between insertions and originations of the ornithomimosaurian myology and other dinosaurs (Fig. 6 and Table 2), except that we could not reliably restore the M. pseudotemporalis profundus. This muscle usually attaches on the epipterygoid in extant sauropsids and has been identified in other dinosaurs (Holliday, 2009). Because none of the specimens had an identifiable epipterygoid attachment visible on the quadrate (as in birds: Holliday & Witmer, 2007) the muscle was not reconstructed. It is possible the muscle occupies some of the space used here in the reconstruction of the M. adductor mandibulae posterior.

The amounts by which muscle moment arm lengths and mechanical advantages are affected by retrodeformation are variable between taxa and between different muscle groups (Tables 3–5 and Fig. 8). Muscle moment arms and mechanical advantages are modified most in Garudimimus and least in Ornithomimus. The M.AMEm, M.AMEs and the M.AMP are least affected by retrodeformation. The M.AMEp and the pterygoideus complex are most affected by retrodeformation. Comparison between species shows that for all three ornithomimosaurs the mechanical advantage for the pterygoideus complex is always very low pre- and post-retrodeformation, because the muscle centroids are close to the jaw joint (Table 4). The rest of the muscles possess broadly similar mechanical advantages (Table 4). Using the muscle moment arms and PCSA estimates, muscle forces were calculated (Table 6). There are some notable differences in comparable adductor muscle forces. For example, Ornithomimus has typically less forceful muscle contraction, with the exception of the M. pterygoideus dorsalis. Struthiomimus and Garudimimus have broadly comparable adductor muscle force production with the exception of lower force production in the M. pterygoideus complex of Garudimimus. Struthiomimus produces the highest total adductor force. Given that the skulls are all similar lengths and therefore the ‘out’ lever arms (jaw lengths) are similar in length, Struthiomimus produces the highest bite forces at any of the positions along the jaw, whilst Ornithomimus produces the lowest. The presence of a rhamphotheca marginally reduces estimated bite forces.