Micro-CT data reveal new information on the craniomandibular and neuroanatomy of the dicynodont Gordonia (Therapsida: Anomodontia) from the late Permian of Scotland

Abstract

Dicynodontia was an abundant, globally widespread clade of Permo-Triassic synapsids on the stem lineage of mammals. Although there is an extensive body of literature on dicynodont craniomandibular anatomy, only recently has the power of computed tomographic (CT) scanning been applied to this system. CT-assisted research on dicynodonts has focused on the smallest members of the clade, while larger dicynodonts (particularly the members of the diverse, long-ranging subclade Bidentalia) have received comparatively little attention. Here, we work towards filling that gap by presenting a µCT-assisted reconstruction of ‘The Elgin Marvel’, a bidentalian specimen consisting of a complete cranium and mandible from late Permian deposits near Elgin, Scotland, which historically has been difficult to study because of its unusual preservation as void space in sandstone. This specimen can be referred to Gordonia, which is solely represented by moulds of void specimens. The µCT data reveal new information on the palate and endocranium of this taxon that could not previously be gleaned from physical moulds made from the void specimens. A phylogenetic analysis indicates that Gordonia and the Chinese Jimusaria form a clade of bidentalians characterized by narrow pterygoid medial plates, expanding our understanding of late Permian biogeography. The endocast of Gordonia is similar to that of other non-cynodont therapsids, and has a remarkably enlarged pineal body, probably related to exaggeration of the sagittal crest. Comparisons of encephalization quotients (EQ), a measure of brain size relative to body size, reveal Gordonia has a similar EQ to most other non-cynodont therapsids.

INTRODUCTION

Dicynodontia is one of the major subclades of Synapsida, the clade containing mammals and their extinct relatives. The dicynodont lineage lasted for tens of millions of years, from the middle Permian to the end of the Triassic, surviving the end-Permian mass extinction, and included an ecologically diverse set of species ranging from mole-sized burrowers to elephant-sized browsers (Cox 1972, Barry 1974, Hotton 1986, King 1988, 1990, Angielczyk and Kammerer 2018, Sulej and Niedźwiedzki 2018). Dicynodonts have been found on every continent and are often the most abundant fossil tetrapods in the assemblages where they occur (King 1990, Fröbisch 2009, Smith et al. 2012). Generally, if there is a terrestrial vertebrate-bearing assemblage of middle Permian to Early Triassic age, it probably includes dicynodonts.

A case in point is the Permian tetrapod record from the north of Scotland. late Permian continental sandstones are exposed in the area surrounding Elgin, Moray, which have long been quarried for building stone (Benton and Walker 1985) (Fig. 1A). Body fossils in these deposits are exceedingly rare and difficult to find (although ichnofossils are locally abundant; Sarjeant 1974, Benton and Walker 1985). Nonetheless, dicynodont skeletal remains have been known there since the 19th century (Newton 1893). The paucity of body fossil discoveries can be attributed in part to the unusual local preservation, which consists of three-dimensional void spaces (moulds) within aeolian sandstone, formed by the dissolution of the bone (Benton and Walker 1985, Benton and Spencer 1995). Such fossils can be hard to recognize in situ (appearing only as empty cavities in the rock) and even once collected, their study has historically been a complicated process. Sandstone blocks containing the moulds were generally broken apart in the process of creating physical casts made of rubbery materials [like gutta-percha (natural latex) and silicone] for study (Benton and Walker 1985). This process is detrimental to the original fossils, hindering future research from being carried out on them, and has contributed to the limited information available on the morphology of the Permian and Triassic ‘Elgin Reptiles’ (see: Rowe 1980, Cruickshank et al. 2005, Benton and Walker 2011, Foffa et al. 2020, 2022).

Newton (1893) originally described six species of Permian ‘Elgin Reptiles’ from the Cutties Hillock quarry west of Elgin: the pareiasaur Elginia mirabilis, the cryptodont dicynodont Geikia elginensis, and four species of the dicynodontoid dicynodont Gordonia: the type species G. traquairi plus G. huxleyana, G. duffiana, and G. juddiana. These species were represented by skulls and some postcranial material that were recovered from the Cutties Hillock Formation, which is believed to be roughly of Changhsingian age based on faunal similarities with the latest Permian Daptocephalus Assemblage Zone of the South African Karoo Basin (King 1988, Groenewald and Kitching 1995, Rubidge 1995, Viglietti et al. 2016, Lucas 2018).

Since Newton’s (1893) original description, there have been various attempts to revise the taxonomy of Gordonia and its four component species. Von Huene (1940) synonymized the genus with Dicynodon, but retained all four species as valid. Although some subsequent workers (Janensch 1952, Anderson and Cruickshank 1978) listed Gordonia as a distinct genus, King (1988) also considered it synonymous with Dicynodon, and furthermore synonymized all of its species, leaving D. traquairi as the only valid taxon. The most recent revision came as part of the comprehensive reassessment of the taxonomy of Dicynodon by Kammerer et al. (2011), who found that D. traquairi and D. lacerticepsOwen, 1845 (the type species of Dicynodon) did not form a monophyletic group. As a result, this revision resurrected Gordonia, but maintained synonymy of the four Cutties Hillock species. Kammerer et al. (2011) considered G. traquairi to be a valid species based on the combination of the following characteristics: an anterodorsally angled, rod-like lateral dentary shelf (autapomorphy); long and narrow intertemporal bar with narrow exposure of parietals; vertical orientation of the postorbitals in the intertemporal bar forming sagittal crest; short snout; and short and steep mandibular symphysis.

Amalitzky (1922) described two additional species of Gordonia from Russia: G. annae and G. rossica. Referral to Gordonia was based mainly on the large temporal fenestrae of these specimens, similar to those of the Elgin Gordonia skulls, but this morphology is widely present in bidentalian dicynodonts. Both species are currently considered synonyms of the Russian endemic dicynodontoid Vivaxosaurus trautscholdiAmalitzky, 1922, as they share an autapomorphic narrow, anteroventrally directed caniniform process with a lobe anterior to the caniniform (Kammerer et al. 2011).

The phylogenetic analysis of Kammerer et al. (2011) recovered G. traquairi as a basal dicynodontoid. Relationships within Dicynodontoidea are unstable [see discussion in Angielczyk and Kammerer (2017)], and the exact position of G. traquairi has been mutable in updated versions of this phylogenetic analysis run by other authors (e.g. Angielczyk et al. 2021, Liu 2021, Macungo et al. 2022, Shi and Liu 2023). Although a variety of factors, including rampant homoplasy, underlie instability in this part of the tree, for G. traquairi, part of the problem is the large number of unknown character states for this taxon [119/213 characters, representing 55.87% of characters coded in the recently published phylogenetic analysis by Macungo et al. (2022)]. Notably, currently missing character information for G. traquairi includes many features of the palate and endocranium, which have been shown to be to be phylogenetically important (Surkov and Benton 2004).

An additional specimen currently attributed to G. traquairi was discovered by quarrymen in 1997 in the Hopeman Sandstone Formation, at the Clashach quarry north of Elgin (Clark 1999, Hopkins and Clark 2000) (Fig. 1B). Initially identified as an unusual mould, local expert Carol Hopkins recognized its importance as the remains of an extinct animal. After extraction, the block was delivered to the Hunterian Museum of the University of Glasgow to be studied with non-destructive imaging techniques (Clark 1999). Clark et al. (2004) described these methods and nicknamed this fossil ‘The Elgin Marvel’ (Fig. 1C). CT scanning and magnetic resonance imaging (MRI) were used to generate digital models of the fossil using a Philips Easivision workstation, revealing that the fossil was of a skull of a dicynodont. A stereolithograph was then created through the rapid prototyping process outlined by Birch (1993) to create a 3D replica (catalogued as GLAHM 114914) of the fossil skull (Fig. 1D). Later, Cruickshank et al. (2005) described this specimen and attributed it to the taxon Dicynodon traquairi, supporting equivalency between the Hopeman Sandstone Formation and the Cutties Hillock Formation based on tetrapod fauna. Cruickshank et al. (2005) used the new digital data to describe aspects of the anatomy previously unavailable for the taxon, such as the palatal exposure of the palatines and the anterior border of the interpterygoid vacuity joining the vomerine crest [see table 2 in Cruickshank et al. (2005) for more]. While the new data gathered from this research have been important for building our knowledge of the anatomy and relationships of G. traquairi (e.g. Kammerer et al. 2011), computerized imaging of fossils was at the time somewhat rudimentary. The subsequent two decades have witnessed a huge increase in the capacity and power of computed tomography, with associated major advances in understanding the morphology of fossils.

Dicynodont research has greatly benefitted from the recent advancements in computed tomography and related techniques. Multiple studies have used these technologies to improve our understanding of dicynodont cranial anatomy and internal neuroanatomy (Castanhinha et al. 2013, Laaß 2014, 2015, Laaß and Kaestner 2017, Araújo et al. 2018, 2022, Benoit et al. 2018, Angielczyk et al. 2019, Simão-Oliveira et al. 2019, Macungo et al. 2022, 2023). Despite this growing interest in studying dicynodonts with CT data, there are still unexplored avenues of research. Most relevant to this study is the paucity of CT data available to study the endocranial anatomy of bidentalian dicynodonts (Clark et al. 2004, Benoit et al. 2018). This work aims to address this bias using Gordonia, a relatively small bidentalian (basal skull length <20cm). The µCT data presented herein reveals never-before-seen aspects of the external and internal cranial anatomy of this taxon and provides the first digitally reconstructed brain endocast of a bidentalian.

Additionally, this work contributes to the growing number of modern studies on the ‘Elgin Reptiles’. Research on the Triassic reptiles Erpetosuchus grantiNewton, 1894 and Scleromochlus tayloriWoodward, 1907 using µCT data has substantially changed our understanding of the anatomy and phylogenetic affinities of both these taxa, highlighting the importance of revisiting historic taxa using modern techniques (Benton and Walker 2011, Foffa et al. 2020, 2022). Moreover, Flett et al. 2024 have recently investigated the taphonomy of tetrapod tracks in the Hopeman Sandstone Formation and the Cutties Hillock Formation. Herein, we provide the first study of one of the Permian representatives of the Elgin tetrapod assemblage using µCT data.

MATERIALS AND METHODS

Specimens examined

The focus of this study is ELGNM 1999.5.1 (‘The Elgin Marvel’), a block of reddish aeolian sandstone containing a natural mould of a cranium and mandible from the Hopeman Sandstone Formation (Fig. 1C). The specimen was collected from the west face of the Clashach quarry at Hopeman near Elgin, Moray, Scotland (Cruickshank et al. 2005) (Fig. 1A, B). The Hopeman Sandstone Formation has been interpreted to have been deposited under aeolian conditions as part of either a large transverse dune system or star and crescent dunes (Glennie and Buller 1983, Clemmensen 1987, Glennie 2002). A typical outcrop of the Hopeman Sandstone in the Clashach quarry also contains several pebbly layers and rippled surfaces, indicative of flash-flood deposits (Williams 1973). ELGNM 1999.5.1 was found at the extreme top of the quarry, within a larger block of sandstone that had no internal bedding structures. The interface between the sandstone matrix and the mould within is heavily stained with dark brown material, which is probably a mixture of numerous metals, including iron and cobalt (Newton 1893, Cruickshank et al. 2005). ELGNM 1999.5.1 is associated with another block, ELGNM 1999.5.2, which contains some of the right squamosal, the right quadrate, the right quadratojugal, and a minor part of the postorbital bar of this dicynodont specimen (Cruickshank et al. 2005). This second block was not scanned for this study, which focuses only on the ELGNM 1999.5.1 block containing the bulk of the cranium and a complete mandible. ELGNM 1999.5.1 is also associated with a fossil mould of a humerus (ELGNM 1999.22) (Cruickshank et al. 2005).

The digital model of ELGNM 1999.5.1 was compared with other specimens of G. traquairi and other dicynodontoids examined by the authors (see Supporting Information, Table S1; Figs S1–S28).

Institutional abbreviations

AM, Albany Museum, Makhanda, South Africa; AMNH, American Museum of Natural History, New York, USA; BGS, British Geological Survey, Keyworth, England, UK; BP, Evolutionary Studies Institute, University of the Witwatersrand, Johannesburg, South Africa; ELGNM, Elgin Museum, Elgin, Scotland, UK; GLAHM, Hunterian Museum, University of Glasgow, Glasgow, Scotland, UK; GPIT, Paläontologische Sammlung, Eberhard Karls Universität Tübingen, Germany; IVPP, Institute for Vertebrate Paleontology and Paleoanthropology, Beijing, China; MB.R, Museum für Naturkunde Berlin, Germany; MVP, Museu Vicente Palloti, Santa Maria, Brazil; NHMUK, The Natural History Museum, London, England, UK; NMQR, National Museum, Bloemfontein, South Africa; PIN, Paleontological Institute of the Russian Academy of Sciences, Moscow, Russia; PPN, Projecto PalNiassa Collection, Museum Nacional de Geologia, Maputo, Mozambique; SAM, Iziko South African Museum, Cape Town, South Africa; TMM, Texas Science & Natural History Museum, Austin, USA; UCMP, University of California Museum of Paleontology, Berkeley, USA; UFRGS, Universidade Federal do Rio Grande do Sul, Porto Alegre, Brazil; UMZC, University Museum of Zoology, Cambridge, England, UK; UNIPAMPA, Universidade Federal do Pampa, São Gabriel, Brazil; ZPAL, Institute of Paleobiology of the Polish Academy of Sciences, Warsaw, Poland.

Micro-computed tomography scanning and processing

ELGNM 1999.5.1 was µCT scanned using a Nikon XT H 225 µCT scanner by Dr Elizabeth Martin-Silverstone with the aid of DF at the University of Bristol Palaeobiology Laboratories in August of 2021. The scanning generated 1999 CT slices with an isometric resolution (voxel size) of 0.126 mm. The µCT dataset was processed using the MATERIALISE MIMICS INNOVATION SUITE 24 (https://www.materialise.com/en/healthcare/mimics-innovation-suite/mimics). The surface data (STL files containing the digital 3D models) are available in the Supporting Information (Files S2-S6). The µCT dataset and associated surface data are freely available at https://www.morphosource.org/projects/000586798?locale=en. A stereolithograph of a humerus associated with ELGNM 1999.5.1 [GLAHM 114108 (see Supporting Information, Fig. S29); cast of the mouldic fossil ELGNM 1999.22) was also used in this study for body mass calculations (Cruickshank et al. 2005).

There are important limitations to the data that must be considered: (i) the mouldic nature of the fossil generally does not permit sutures to be discerned between bones, which prevents precise identification of the morphology of many skeletal elements; (ii) the surface texture of the original bone, which is also phylogenetically important and provides insight into palaeobiology, cannot be visualized in this mouldic fossil (though it may be visible in other mouldic specimens from the Elgin area, such as for Elginia mirabilis); and (iii) the large size of the block, which approaches the maximum size and weight capacity of the µCT scanner, affected the resolution and clarity of the µCT scan by attenuating and scattering the X-ray beam [an issue also encountered by Clark et al. (2004) to a substantially greater extent]. Due to this issue, several anatomical details observed (or not observed) in the model should be cautiously interpreted. Out of caution, any uncertain feature was left as a ‘?’ in the phylogenetic dataset.

Phylogenetic analysis

Previously, data used for Gordonia in phylogenetic analyses were gathered from physical casts created through destructive techniques and the low-resolution digital models made by Clark et al. (2004). The objectives of our phylogenetic analysis were to test the referral of ELGNM 1999.5.1 to Gordonia traquairi, and to test the impact of our new data on the relationships between G. traquairi and other species. We ran a phylogenetic analysis with two operational taxonomic units (OTUs): ‘Gordonia traquairi’ [scores based only on the holotype (coded using the cast NHMUK PV R 2106)] and ‘The Elgin Marvel’ (scores based only on ELGNM 1999.5.1). The data matrix and character list used are those provided by Macungo et al. (2022), with data for Kunpania scopulusaSun, 1978 based on Angielczyk et al. (2021) and data for Jimusaria sinkianensisYuan and Young, 1934 and Jimusaria monanensisShi and Liu, 2023 based on Shi and Liu (2023). The dataset includes 23 continuous characters, 190 discrete characters, and 122 operational taxonomic units (OTUs). The character list and differences between the scores of each of the two Gordonia OTUs are available in the Supporting Information (File 7). The TNT script and data matrix for the phylogenetic analysis are included in the Supporting Information (File).

The updated data matrix was used to conduct a maximum parsimony analysis in TNT v.1.5 (Goloboff et al. 2008). We first analysed it with a new technology search using the sectorial search, ratchet, drift, and tree fusing tools with default settings. Then, we subjected the shortest trees to an additional round of tree‐bisection‐and‐reconnection (TBR) branch swapping (traditional search), to more broadly sample tree space. Jackknife resampling (with removal probability of 36) and Bremer supports were used to determine clade support. The ‘list’ and ‘map common synapomorphies’ functions were used to identify synapomorphies of relevant clades.

Calculation of the body mass and encephalization quotient of Gordonia

Encephalization quotient (EQ), a measure of brain size relative to body mass, is commonly used as a rough estimate of intelligence of extinct taxa (Jerison 1973, Evans et al. 2009, Bertrand et al. 2022). Castanhinha et al. (2013), Laaß (2015), Laaß and Kaestner (2017), and Simão-Oliveira et al. (2019) previously listed EQ values for the dicynodonts Niassodon mfumukasiCastanhinha et al., 2013, Pristerodon mackayiHuxley, 1868, Kawingasaurus fossilisCox, 1972, and Rastodon procurvidensBoos et al., 2016 gathered using digital data, and for Lystrosaurus Cope 1870 gathered using traditional methods. Additionally, Simão-Oliveira et al. (2019) listed EQ values for many other synapsids. Here, we calculate the EQ of Gordonia to make comparisons with all these synapsids.

For simplicity, the EQ and body mass of ELGNM 1999.5.1 will also be referred to as the EQ and body mass of G. traquairi, as these are the only values available for this taxon. To calculate the EQ, the endocranial volume was digitally measured using Mimics from the mask of the endocast. Most of the endocast was reconstructed, but the olfactory bulbs and most of the olfactory tract could not be reconstructed. The measured endocranial volume probably represents a slight underestimate, because of incompleteness (see above), taphonomic distortion, and the absence of some ossified boundaries of the braincase (e.g. ventral floor and anterior boundary).

A variety of equations have been used to calculate dicynodont body mass. We performed calculations using a total of six equations listed by Simão-Oliveira et al. (2019) and Laaß and Kaestner (2017), which rely on skull length and humerus measurements. We opted to use multiple equations, as it is currently unclear which of the equations most accurately predicts dicynodont body mass values in the absence of complete skeletons [see Romano and Manucci (2019) for discussion of the pitfalls in applying standard regression formulae to calculations of dicynodont body mass]. Skull measurements were taken using Mimics tools on our new digital model and manually checked on a 1 : 1 scaled 3D print of the data from Clark et al. (2004). The humerus measurements were taken from a 1 : 1 scaled 3D print of the humerus mould [GLAHM 114108, preserved in two parts (anterior and posterior sides) that are a perfect fit] associated with ELGNM 1999.5.1, which is not distorted in any way that would negatively impact the measurements. A photograph of GLAHM 114108 and a full list of the six equations and the reasoning behind why each is used are available in the Supporting Information (Fig. S29 in File 1 for the photograph, and File 9 for the equations). The average of the estimates (with one exclusion) was used as a body mass value in the EQ calculations. We excluded the value calculated through Equation 4, because it is far greater than any other calculated value (about 3.8 times larger than the next greatest estimate) and likely represents an overestimation.

Jerison (1973) performed a regression analysis using data from a wide range of extant mammals (the closest living relatives of dicynodonts) to find a correlation between body mass and brain size. Later, Manger (2006) built upon this work by performing a regression analysis with an updated version of the dataset of extant mammals that includes more taxa, but also excludes cetaceans and primates since they have remarkably large brains for their body size. We calculated the EQ of Gordonia using the following equation expressing the regression found by Manger (2006):

The EQ values of Gordonia with and without the pineal body were calculated due to the majority of the pineal body observable in the endocast probably being glandular tissue instead of neural tissue; the pineal body occupies 9.87% of the endocast volume (see Results). These values were compared with those of other taxa, which were retrieved from Simão-Oliveira et al. (2019). A table including the Manger values and error margins for each taxon used in this study can be found in the Supporting Information (File 10).

RESULTS

Systematic palaeontology
Synapsida Osborn, 1903,
Therapsida Broom, 1905,
Anomodontia Owen, 1860,
Dicynodontia Owen, 1859,
Dicynodontoidea Olson, 1944,
GordoniaNewton, 1893

Diagnosis:

As for the type species.

Type species:

Gordonia traquairiNewton, 1893

Gordonia traquairiNewton 1893:436.

Gordonia huxleyanaNewton 1893:445.

Gordonia duffianaNewton 1893:450.

Gordonia juddianaNewton 1893:462.

Dicynodon traquairivon Huene 1940:280.

Dicynodon duffianusvon Huene 1940:280.

Dicynodon huxleyanusvon Huene 1940:280.

Dicynodon juddianusvon Huene 1940:280.

Type locality:

Cutties Hillock Quarry, Elgin, Scotland.

Horizon:

Cutties Hillock Formation [probable equivalent of the Changhsingian Daptocephalus Assemblage Zone of South Africa (Rubidge 1995, Viglietti et al. 2016)].

Holotype:

BGS GSE 4805 [incorrectly listed as BGS GSE 11703 by King (1988) and subsequent sources: Benton and Spencer 1995, Cruickshank et al. 2005, Szczygielski and Sulej 2023; Kammerer et al. 2011].

Referred material:

From the type locality: BGS GSE 11703, BGS GSE 11704 (holotype of G. huxleyana), ELGNM 1890.3 (holotype of G. juddiana), ELGNM 1978.550 [specimen labelled ‘Gordonia Traquairi?’ by Newton (1893); previously incorrectly reported as a specimen of the pareisaur Elginia mirabilis by Benton and Spencer (1995) and Cruickshank et al. (2005)], ELGNM 1978.559.1a,b (holotype of G. duffiana); from the Hopeman Sandstone Formation deposits of the Clashach quarry, Hopeman, Elgin, Morayshire (National Grid Reference NJ 163702): ELGNM 1999.5.1, ELGNM 1999.5.2, ELGNM 1999.22.

Emended diagnosis:

A small (basal skull length approximately 16 cm) dicynodontoid characterized by the following combination of character states (+ indicates an autapomorphy, ^* indicates a newly recognized diagnostic character state based on µCT data described herein): + an anterodorsally angled lateral dentary shelf with a rod-like morphology (dorsal and lateral extensions of the shelf are roughly equal in size) that does not expand into a rounded anterior boss nor a diffuse muscle scar, does not have a transverse ridge on its dorsal surface, and does not have a fossa present near its posterior end; long, narrow intertemporal bar with narrow exposure of parietals; vertical orientation of the postorbitals in the intertemporal bar forming sagittal crest; short snout (with anteroposteriorly short premaxillary region); proportionally large orbits^*; dorsoventrally thin suborbital zygoma^*; hooked snout tip^*; narrow snout in palatal view, probably resulting from mediolaterally narrow maxilla and premaxilla^*; median pterygoid plate anteroposteriorly elongate and mediolaterally narrow in palatal view^*; anterior pterygoid close to sagittal plane^*, and short, steep mandibular symphysis.

Description

The new 3D reconstructions of the cranium and mandible of ELGNM 1999.5.1 (Fig. 2) generally accord with those made by Clark et al. (2004). However, we were able to more completely reconstruct the palate of this specimen, as well as for the first time provide information on its endocranial anatomy. Throughout the following description, comparisons are made to casts of the other G. traquairi specimens (e.g. NHMUK PV R 2106), listed in the Supporting Information, Table S1 with accompanying photographs.

Cranium

As in all dicynodonts, the premaxilla of ELGNM 1999.5.1 forms the beak at the anterior end of the cranium, which would probably have been covered in a rhamphotheca in life (Jasinoski and Chinsamy-Turan 2012, Benoit et al. 2018) (Fig. 3). The premaxillary region (the bone anterior to the external nares) is anteroposteriorly short in lateral view. The snout of ELGNM 1999.5.1 is neither deflected nor dorsoventrally elongate, unlike that of Lystrosaurus. In lateral view (Fig. 3A), the premaxilla has a prominent ‘hooked’ tip, as can also be observed in NHMUK PV R 2106 (G. traquairi holotype). This condition in these Gordonia specimens approaches that of Dinanomodon gilliBroom, 1932, a taxon in which the hooked premaxillary trip is particularly exaggerated (Kammerer et al. 2011). This morphology cannot be observed in other G. traquairi specimens, and it is unclear whether this is reflective of biological or taphonomic variation. The external nares of ELGNM 1999.5.1 are anteroposteriorly longer than they are dorsoventrally tall (Fig. 3A, C), which is also the case in ELGNM 1893.6 ("G. juddiana"). NHMUK PV R 2106 (G. traquairi holotype) and NHMUK PV R 2109 ("G. huxleyana") instead have external nares that are dorsoventrally taller than they are anteroposteriorly long.

Most of the palate of ELGNM 1999.5.1 can be visualized, providing substantially more information on this part of the skull than in any other known Gordonia specimen (Fig. 4). The anterior tip of the snout is rounded in ventral view. The anterior palatal ridges of the secondary palate can tentatively be identified as proportionally small compared to most other dicynodontoids, but appear to be present, unlike in Kunpania scopulusa (Angielczyk et al. 2021). It is unclear whether the proportionally small size of the anterior palatal ridges is a true biological character or a result of taphonomic factors. A median palatal ridge can be identified along the posterior end of the palatal surface of ELGNM 1999.5.1 (Fig. 4) and is confluent with the vomer. There are no longitudinal depressions along the secondary palate (Fig. 4).

The lateral surfaces of the caniniform processes of ELGNM 1999.5.1 are convex laterally and do not have any clear depressions or protrusions (Fig. 3). They are interpreted as being formed by the maxillae, as is the case in all other dicynodonts. The caniniform processes of NHMUK PV R 2106 (G. traquairi holotype), NHMUK PV R 2109 ("G. huxleyana"), and ELGNM 1893.6 ("G. juddiana") also have this morphology. Unlike the maxillae of closely related taxa, this region in ELGNM 1999.5.1 seems to be proportionally mediolaterally narrow (Kammerer et al. 2011). A similar morphology is observed in Jimusaria sinkianensis, but in most other dicynodontoids this region is widened, with a greater degree of ‘splay’ of the caniniforms (Kammerer et al. 2011). As in most dicynodontoids, the only dentition present in G. traquairi is the caniniforms, which erupt from the maxillae (Kammerer et al. 2011). The caniniforms of ELGNM 1999.5.1 are directed ventrally, like those of the other G. traquairi specimens in which they are preserved [NHMUK PV R 2106 (G. traquairi holotype), ELGNM 1893.6 ("G. juddiana"), NHMUK PV R 2109 ("G. huxleyana"), and NHMUK PV R 2107 (referred to G. traquairi)].

A nasal boss is present in ELGNM 1999.5.1, and is best seen in lateral view (Fig. 3A, C). It is a single, slightly raised expansion of the nasals, which is continuous between both bones, like that of other Permian dicynodontoids. A boss is also present in NHMUK PV R 2106 (G. traquairi holotype). A nasal boss cannot be clearly identified in NHMUK PV R 2109 ("G. huxleyana"), ELGNM 1893.6 ("G. juddiana"), or NHMUK PV R 2107 (referred to G. traquairi), but this more likely represents taphonomic artefact than biological variation.

There is a slight circumorbital rim extending across the orbital margins of what is probably the prefrontal through the postorbital in ELGNM 1999.5.1 (Kammerer et al. 2011) (Fig. 3D, E). A similar circumorbital rim is present in NHMUK PV R 2106 (G. traquairi holotype) but is not as well developed in other specimens. A weakly-developed rim is visible at the anterodorsal corner of the orbit (probably corresponding to the prefrontal margin) of NHMUK PV R 2109 ("G. huxleyana") and ELGNM 1893.6 ("G. juddiana"), but it does not appear to continue posteriorly. Kammerer et al. (2011) noted that G. traquairi has relatively large orbits compared to other dicynodonts, and this is supported by the new data presented herein (Fig. 3).

In dorsal view, the postorbitals converge posterior to the pineal foramen, which is located near the anterior end of the intertemporal bar (Fig. 3D–F). An interorbital depression can be identified anterior to the pineal foramen, which probably corresponds to the region occupied by the preparietal. The pineal foramen is anteroposteriorly longer than it is mediolaterally wide and it is similar in shape to that of ELGNM 1893.6 ("G. juddiana"). NHMUK PV R 2109 ("G. huxleyana") also has a preserved pineal foramen, but it is proportionally larger than that of ELGNM 1999.5.1. The pineal foramen of "G. duffiana" (ELGNM 1978.559.1) is less elongated than that of ELGNM 1999.5.1 and shows greater exposure in the intertemporal bar, not being overlapped by the postorbitals (Newton 1893). The adductor fossa of the right postorbital of ELGNM 1999.5.1 is medial to the posterior margin of the right orbit. Much of the postorbitals are dorsal to the rest of the skull roof as they form the lateral surfaces of the sagittal crest (Fig. 5). As noted by Kammerer et al. (2011), the sagittal crest of NHMUK PV R 2106 (G. traquairi holotype) is proportionally large (unusually so for a dicynodont of its size), mediolaterally narrow (providing only slight dorsal exposure of the parietals), and formed by vertically oriented postorbitals, and this mostly matches what is seen in the new data presented herein. The intertemporal portion of the postorbitals shows more horizontal orientation in ELGNM 1999.5.1 than in the holotype, particularly anteriorly, although the postorbital surface is near-vertical at the apex of the crest. In dorsal view, the crest can be seen to be thinnest at its mediodorsal edge and expands in length lateroventrally (Fig. 5). The crest is also laterally thicker anteriorly than posteriorly. In lateral view, the sagittal crest increases in height posteriorly from the pineal foramen, before decreasing in height at the posterior end of the skull. Similar to ELGNM 1999.5.1, the crest of NHMUK PV R 2106 (G. traquairi holotype) gradually increases in dorsoventral height posteriorly from the anterior origin of the crest and maintains a similar height for the majority of the crest until it decreases in height at its posterior end. The crest of NHMUK PV R 2109 ("G. huxleyana") is not as tall as that of ELGNM 1999.5.1 and NHMUK PV R 2106 (G. traquairi holotype) but has similar changes in dorsoventral height throughout the crest to those of the other two specimens. The crests of NHMUK PV R 2107 (referred to G. traquairi) and ELGNM 1893.6 ("G. juddiana") are considerably less tall than those of the previously mentioned G. traquairi specimens and are instead the same height as the rest of the dorsal surface of the skull roof. Such variation in crest shape between specimens, where smaller, skeletally immature specimens have smaller crests, lines up with known trends in therapsid ontogeny (Jasinoski et al 2015, Kammerer et al. 2015, Jasinoski and Abdala 2016). This suggests a biological reason behind the variation between specimens that is related to muscle area. The postorbitals of ELGNM 1999.5.1 articulate with the parietals medially, which extend ventrally beyond the crest. The mediolaterally thin crest indicates that the dorsal exposure of the parietals was narrow.

Unlike for most of the specimen, there is a clearly identifiable suture between the left jugal and the zygomatic arch of the left squamosal (Fig. 6A). The jugal–squamosal suture forms a diagonal line immediately posterior to the left orbit in left lateral view, such as is seen in many other Permian dicynodontoids, e.g. Dicynodon lacerticeps (Kammerer et al. 2011). The other cranial specimens of G. traquairi also preserve the zygomatic arch, but precise jugal morphology is difficult to determine. The suborbital portion of the zygoma is remarkably dorsoventrally thin in ELGNM 1999.5.1 (Fig. 6), unlike the thicker structures in many other dicynodontoids [e.g. NHMUK PV OR 47047 (Daptocephalus leoniceps)], but similar to that of Delectosaurus arefjeviKurkin, 2001 (PIN 4644/1) (Kammerer, pers. observ.). Comparable zygomatic proportions are present in NHMUK PV R 2106 (G. traquairi holotype), NHMUK PV R 2109 ("G. huxleyana"), and ELGNM 1893.6 ("G. juddiana"), suggesting that this is a consistent feature in Gordonia.

Only the left squamosal is completely preserved in ELGNM 1999.5.1 (Fig. 6), while the right is split between this specimen and ELGNM 1999.5.2 (Clark et al. 2004). In dorsal view, the width of the left zygomatic arch expands towards its posterior end. In dorsal view, the posterior end of the squamosal is concave, and the medial portion of the left squamosal articulates with the bones of the sagittal crest. Near the anterior end of the zygomatic arch, there is an abrupt shift in morphology of the zygomatic arch as it is dorsoventrally taller than it is mediolaterally wide anteriorly and then becomes mediolaterally wide posteriorly for much of its anteroposterior length. This morphology is not seen in NHMUK PV R 2106 (G. traquairi holotype), NHMUK PV R 2109 ("G. huxleyana"), NHMUK PV R 2107 (referred to G. traquairi), and ELGNM 1893.6 ("G. juddiana"), indicating that this shift in morphology in ELGNM 1999.5.1 represents taphonomic artefact rather than biological variation. All the G. traquairi specimens with well-preserved left squamosals have squamosal ventral processes that are exposed posteriorly. In left lateral view, the anteroposterior length of the quadrate ramus of the squamosal of ELGNM 1999.5.1 is greatest dorsally, and gradually decreases ventrally (Fig. 6A). The adductor fossa of the left squamosal can be identified in left lateral and posterior views as the fossa between the quadrate and zygomatic rami (Fig. 6B).

The left quadrate of ELGNM 1999.5.1 is preserved with little distortion. In palatal view, the roughly equal-sized medial and lateral condyles of the quadrate can be identified with a trochlea between them, which would have articulated with the mandible (Fig. 4). The right quadrate contacts the right quadrate ramus of the pterygoid, but the left quadrate only nearly contacts the left quadrate ramus of the pterygoid. This asymmetrical difference is probably a result of inconsistent preservation and not biological variation, and in life, there would have been contact on both sides. NHMUK PV R 2106 (G. traquairi holotype) and ELGNM 1893.6 ("G. juddiana") show comparable quadrate morphology. Only the left quadratojugal of ELGNM 1999.5.1 can be confidently identified. The bone is a thin, broad plate that is flush with the squamosal. In posterior view, the ventral process of the squamosal occludes the quadratojugal. Similarly, NHMUK PV R 2106 (G. traquairi holotype) and NHMUK PV R 2107 (referred to G. traquairi) also possess left quadratojugals that are visible in left lateral view, which are continuous with the squamosals. In NHMUK PV R 2109 ("G. huxleyana") and ELGNM 1893.6 ("G. juddiana"), the quadratojugals appear to be missing, exposing the shallow articular fossa where they would have overlapped the anteroventral face of the squamosal. The left quadratojugal foramen of ELGNM 1999.5.1 is dorsoventrally taller than it is mediolaterally wide (Fig. 6A).

Part of the vomer of ELGNM 1999.5.1 can be identified as a thin ridge between the internal nares, confluent anteriorly with the median palatal ridge of the premaxilla, and has a ventral margin offset from the rest of the palate (Fig. 4). The ridge has a consistent mediolateral width throughout (narrow and blade-like), which is unlike that of other dicynodontoids such as D. lacerticeps where a large portion of the anterior part of the vomer ridge is mediolaterally wider than the rest of the ridge (Kammerer et al. 2011). Posteriorly, the vomer splits at an acute angle, bounding an isosceles triangular fossa anterior to the interpterygoid vacuity. Lateral and dorsal to this split, horizontally oriented laminar portions of the ventral surface of the vomer form part of the primary palate. The dorsal portion of the vomer was incompletely visible in the CT data; what could be reconstructed is a tall, blade-like median element partially dividing the nasal capsule, similar to that known in other dicynodontoids (e.g. Lystrosaurus; Cluver 1971).

The palatine bones of ELGNM 1999.5.1 border the internal nares laterally (Fig. 4). The exact morphology of the palatines is unclear, due to poor resolution at the edges of the internal narial cavities. The right internal narial opening in particular appears damaged, with an opening extending much further anterior (beyond the caniniform process; Fig. 4A) than in other dicynodonts. The posterior halves of the internal nares are better preserved and are almost parallel to each other, unlike those of D. lacerticeps, which show greater transverse expansion posteriorly (Kammerer et al. 2011). The right palatine of ELGNM 1999.5.1 has a dorsal projection (Fig. 7A, B), which is angled anterolaterally and has a concave medial face. This portion of the palatine appears to bound part of the labial fossa. It is similar in morphology to that of other dicynodontoids, but is mediolaterally wider and more strongly dorsally angled than that of earlier diverging taxa, such as Niassodon mfumukasi (Castanhinha et al. 2013).

Much of the following anatomy of the pterygoid of ELGNM 1999.5.1 described herein helps distinguish G. traquairi from other taxa and cannot be seen in the reconstructions made by Clark et al. (2004).

In palatal view, the pterygoid of ELGNM 1999.5.1 is X-shaped with two anterior rami diverging slightly anterolaterally and two quadrate rami diverging more widely posterolaterally from the medial plate (Fig. 4). The anterior rami of ELGNM 1999.5.1 diverge much less laterally than those of other taxa, such as Dicynodon spp. (Kammerer et al. 2011, Kammerer 2019). This morphology contributes to the relatively narrow palate of ELGNM 1999.5.1 compared to other Permian dicynodontoids (see above) (Kammerer et al. 2011, Kammerer 2019, Liu 2021), and is instead more similar to some non-dicynodontoid dicynodont taxa such as Compsodon helmoedivan Hoepen, 1934 (Angielczyk and Kammerer 2017). There are keels on the anterior rami of ELGNM 1999.5.1, which are best visible in left lateral view (Fig. 7A). Similar keels are present throughout Dicynodontoidea, other than in kannemeyeriiforms (Kammerer et al. 2011). The keels of ELGNM 1999.5.1 gradually increase in dorsoventral height towards their anterior ends. In palatal view, there is a distinct indentation anterior to the left anterior ramus of the pterygoid, which is best interpreted as the lateral palatal fenestra based on its position and its somewhat teardrop-shaped outline (Fig. 4B). The palatal fenestrae of other dicynodontoids, such as D. lacerticeps, have a similar shape (Kammerer et al. 2011). The interpterygoid vacuity can be identified anterior to the median pterygoid plate and is also roughly teardrop-shaped, but anteroposteriorly elongate (Fig. 4A). A labial fossa appears to be present lateral to the left anterior pterygoid ramus (Fig. 7B). The elements bounding this fossa cannot be determined; by comparison to other dicynodontoids they probably consist of the maxilla, jugal, and palatine (Kammerer et al. 2011). It is unlikely that the pterygoid ramus directly contacts the fossa.

The medial pterygoid plate of ELGNM 1999.5.1 is substantially anteroposteriorly longer than it is mediolaterally wide (Fig. 4). The median plate is proportionally anteroposteriorly longer and mediolaterally narrower than in most other Permian dicynodontoids, such as NHMUK PV OR 47047 (Daptocephalus leoniceps), NHMUK PV R 4039 (V. trautscholdi), and Dicynodon spp. and Turfanodon spp. (Kammerer et al. 2011, Kammerer 2019, Liu 2021) (Fig. 4), but is comparable to that of Jimusaria spp. (see IVPP V341407 in: Kammerer et al. 2011, Shi and Liu 2023). There are keels on the lateral sides of the medial plate that are continuous with the keels of the anterior rami (Fig. 7A). The crista oesophagea can be identified as a weak ridge posterior to the interpterygoid vacuity. The crista oesophagea is present in most dicynodontoids except for Daptocephalus spp. (Kammerer 2019). It should be noted that juvenile specimens of Dicynodon lacerticeps also lack a crista oesophagea, indicating potential ontogenetic control on this feature in other taxa (Kammerer 2019).

The quadrate rami of the pterygoid of ELGNM 1999.5.1 diverge posterolaterally to articulate with the quadrates (Fig. 4). Both quadrate rami extend slightly ventrally from the medial plate (Fig. 7A) like those of NHMUK PV R 2106 (G. traquairi holotype), but unlike ELGNM 1893.6 ("G. juddiana") and NHMUK PV R 2109 ("G. huxleyana") in which the quadrate rami extend ventrally at a steeper angle, possibly due to distortion. The ventral sides of the quadrate rami are continuous with the keels on the lateral sides of the medial plate. In lateral views, the height of the quadrate rami is comparable to that of the lateral sides of the medial plate.

The parasphenoid of ELGNM 1999.5.1 is a thin, elongate element flush with other internal cranial bones (Fig. 8A, B) and probably fused with the basisphenoid, as in other dicynodonts (Macungo et al. 2022). A small, median ridge on the dorsal surface of the parasphenoid can be tentatively identified as part of the base of the cultriform process. A parasphenoidal sulcus is present on the dorsal surface as a shallow depression anterior to the endocranial cavity. In palatal view, a single foramen for the internal carotid artery is visible, presumably borne on the basisphenoid (Fig. 4). It is slightly offset from the sagittal plane, indicative of it being one of a pair of foramina (the other not preserving). It is in a similar position as the same structure in Lystrosaurus, based on UMZC T758 and UMZC T788 and what is described by Cluver (1971). For the basisphenoid, two additional notable features can be identified. The first is a dorsally projecting tab-like structure that is angled slightly posteriorly, which is best interpreted as the clinoid process (Fig. 8A). The second is that the basal tubera (also formed by the basioccipital posteriorly) are nearly vertical, with no clear ridges along their long axes (Fig. 4). The tubera form the posterior portion of the stapedial facet, which is open distally. Most details of the sphenoid elements (especially those that cannot be identified in palatal view) could not be observed in the reconstructions by Clark et al. (2004).

The prootic dorsal processes compose the dorsally extending portion of the internal cranial bones that articulate with the occiput (Fig. 9B, C). The pila antotica is visible as a weakly-developed, anteriorly-directed process at the anteroventral margin of the prootic (identity of this element as prootic based on comparison with other synapsids; it is likely that the prootic is fused with other elements to form a periotic as in other dicynodonts; see below). The pila antotica is unusually low, which probably represents incomplete preservation or visualization. The only other specimens of G. traquairi with identifiable prootic regions are NHMUK PV R 2106 (G. traquairi holotype), NHMUK PV R 2109 ("G. huxleyana"), and NHMUK PV R 2107 (referred to G. traquairi). However, these specimens do not reveal much detail concerning these elements, and do not substantially aid in determining pila antotica morphology in Gordonia. The opisthotics of ELGNM 1999.5.1 are inferred to contribute to the borders of the vestibules, and their paroccipital processes are identifiable in posterior view as being flush with the rest of the occiput (Fig. 9A). A similar paroccipital process can also be identified in posterior view of the occiput of ELGNM 1893.6 ("G. juddiana").

The prootics and opisthotics also contain the vestibule (described in the description of the endocast and vestibule below). The opening for the vestibule to meet the endocranial cavity can be seen in left medial view of the basicranium, and based on the morphology known for other dicynodont braincases (Macungo et al. 2022), would have been enclosed by the prootic, opisthotic, and basioccipital (Fig. 8D). The jugular foramen and opening for the vestibule can also be seen in left medial view of the basicranium and have a very similar position to what is seen throughout dicynodonts (e.g. Lystrosaurus; UMZC T758 and UMZC T788) (Surkov and Benton 2004). The basioccipital median ridge is dorsally elevated (Fig. 8A–C). Additionally, the foramen for the cranial nerve VII (facial nerve), the fenestra basicranialis, and the foramen ovale can also be identified (Fig. 8D). These neurocranial elements of ELGNM 1999.5.1 could not be identified in the models made by Clark et al. (2004), emphasizing the value of novel tomographic study of this specimen.

The occiput is presumed to be composed of a periotic, which is a fusion of several elements (minimally the prootic and opisthotic, but often incorporating several additional elements, e.g. supraoccipital, exoccipitals, and basioccipital), as in most other dicynodonts (Kammerer et al. 2011, Kammerer 2019). An incompletely preserved occipital condyle (which would be formed by paired, lateral exoccipital and median, ventral basioccipital elements) is present ventral to the foramen magnum of ELGNM 1999.5.1 (Fig. 9A). The foramen magnum is oval-shaped, with its dorsoventral height greater than its mediolateral width, as in ELGNM 1893.6 (‘G. juddiana’), and as is common in dicynodonts (Laaß and Kaestner 2017). There is no clear floccular fossa present on the medial side of the supraoccipitals, but this may be a result of taphonomic factors. There is evidence for an ‘unossified zone’, a feature also seen throughout non-dicynodontoid dicynodonts (Laaß 2015, Laaß et al. 2017, Simão-Oliveira et al. 2019). The ‘unossified zone’ is most visible in the endocast described below.

The laminar portion of the fused mesethmoid and orbitosphenoid (anterior plate) of ELGNM 1999.5.1 is flush with the skull roof (Fig. 7C–F). Additionally, part of the olfactory cavity can be identified. Most notably, the mesethmoid posterior wall can be observed. In anterior and posterior views, this wall is simple and thin, but in lateral views, the wall is a ventrally directed and dorsoventrally taller than the anteroposteriorly long, triangular plate pointing anteroventrally. Considering that much of the ethmoid region and olfactory cavity is not well preserved, this pointed portion of the anterior plate may also not be complete, and it is possible that it formed a more extensive median structure separating the orbits, as in many other dicynodonts (King 1988).

Mandible

The mandible of ELGNM 1999.5.1 is completely preserved and is not taphonomically distorted in any way that obscures much of its anatomical details from being discerned.

The dentaries of ELGNM 1999.5.1 have a typical dicynodontoid morphology [as is described by Kammerer (2019)]. Anteriorly, the dentaries fuse and largely contribute to the mandibular symphysis, with a pointed beak at its anterodorsal tip (Fig. 10). The symphysis is rectangular (dorsoventrally taller than mediolaterally wide), and dorsoventrally shorter than that of other other Permian dicynodontoids, including Dicynodon angielczyki and Jimusaria monanensis (Kammerer 2019, Shi and Liu 2023). The anterior surface of the symphysis of ELGNM 1999.5.1 is also more steeply angled than that of D. angielczyki and J. monanensis (Kammerer 2019, Shi and Liu 2023). Additionally, there are two ridges diverging ventrolaterally from each other at about the midpoint of the anterior surface of the symphysis. They represent the ventral parts of the dentaries, and the indented part of the anterior surface of the symphysis ventral to these ridges represents the splenial. The dorsal portion of the symphysis is bifurcated into two dorsally projecting protrusions with a notch between them, which is a preservational artefact associated with incomplete preservation or scan resolution (particularly evident in posterior view; Fig. 10D), in which the left side of the dentary tip is missing. A posterior dentary sulcus can be identified on the dorsal surface of each mandibular ramus immediately posterior to the flat dentary tables, situated posterolateral to the posterior dorsal surface of the mandibular symphysis. These sulci are narrow and deep, run along the dorsal surface of the mandible, and extend approximately across the anterior third of the rami. The lateral edges of the symphysis are swollen. Unlike ELGNM 1999.5.1, NHMUK PV R 2106 (G. traquairi holotype) appears to have a rounded anterior face of the mandibular symphysis in lateral view. However, this is probably artefactual; the dorsal surface of this region in casts is strongly indented, unlike any other dicynodonts and indicative of incomplete preservation in this region.

In lateral view of ELGNM 1999.5.1, the anterior ends of the dentaries are dorsoventrally taller than any other part of the bones (Fig. 10). The lateral dentary shelf of each mandibular ramus is well exposed in lateral views. The lateral dentary shelves are angled anterodorsally, have a rod-like morphology (with the dorsal and lateral extensions of the shelf being roughly equal in size), and do not expand into a rounded anterior boss nor a diffuse muscle scar (Fig. 10). Instead, the anterior portion of each dentary shelf is slightly swollen. A similar lateral dentary shelf can be seen in NHMUK PV R 2106 (G. traquairi holotype). The combination of these characteristics is unique to G. traquairi among dicynodonts [as is noted by Kammerer et al. (2011)]. The lateral dentary shelf of Jimusaria monanensis is similar, but not as strongly angled anterodorsally (Shi and Liu 2023). The non-dicynodontoid dicynodont Compsodon helmoedi also has an anterodorsally angled dentary shelf but, unlike in G. traquairi, it is much more prominent, with a transverse ridge on the dorsal surface of the dentary shelf and a fossa present near the posterior end of the dentary shelf in that taxon (Angielczyk et al. 2023). The lateral dentary shelves largely contribute to the dorsal margin of the mandibular fenestrae, which are considerably anteroposteriorly longer than they are dorsoventrally tall (Fig. 10). This mandibular fenestra shape is also present in NHMUK PV R 2106 (G. traquairi holotype) and NHMUK PV R 2019 ("G. huxleyana"). The suture between the dentary and left angular of ELGNM 1999.5.1 cannot be identified, but is probably represented in NHMUK PV R 2106 (G. traquairi holotype) by a sinusoidal line running dorsoventrally at the posterior end of the mandibular fenestra. The sinusoidal line then connects with the posterior end of the mandibular fenestra.

The reflected laminae of the angulars are ventrally directed projections posterior to the mandibular fenestrae, and each is bisected by a central groove. A similar reflected lamina morphology can be seen in many other Permian dicynodontoids (Kammerer et al. 2011, Kammerer 2019, Olroyd and Sidor 2022). The posterior ends of the surangulars are slightly laterally flared, and this characteristic can also be seen in NHMUK PV R 2106 (G. traquairi holotype) and NHMUK PV R 2109 ("G. huxleyana"). There is a wide separation between the articulars and the reflected lamina, a trait seen throughout Permian dicynodontoids (Olroyd and Sidor 2022). The medial and lateral condyles of the articulars can be identified on either side of the median ridge, which would have articulated with the quadrate. Retroarticular processes protrude ventrally from the posterior ends of the articulars.

Endocast and vestibule

Most of the endocast and the left vestibule of ELGNM 1999.5.1 were reconstructed (Fig. 11), except for the olfactory bulbs, most of the olfactory tract, the entirety of the right vestibule, all semicircular canals, and associated bony canals enclosing blood vessels and nerves. Generally, the endocast is not drastically distorted in any way that obscures identification of much of its anatomy. In the description herein, the endocast of ELGNM 1999.5.1 is referred to as the endocast of G. traquairi, being the only endocast of this taxon known in any detail. As illustrated by Newton (1893: pl. 32, fig. 1), part of the dorsal surface of the brain endocast appears exposed in ‘G. duffiana’ (ELGNM 1978.559.1), but this comprises only a small portion directly underlying the skull roof between the orbits and pineal foramen.

In the lateral view of the endocast (Fig. 11E), the forebrain can be identified as a beam-shaped structure showing a ‘keel’ on its ventral side. This keel is a result of mediolateral compression of the parietals in this specimen, and is unlikely to have been a feature of the anterior braincase in the living animal (in which the floor was probably unossified). The forebrain is anteroposteriorly longer than it is dorsoventrally tall, and mediolaterally thin in dorsal view. This morphology can be attributed to a combination of the legitimately narrow intertemporal region in the skull as well as taphonomic distortion due to compression. The preserved portion of the olfactory tract is within the olfactory cavity in the orbitosphenoid–mesethmoid element. Overall, the elongated shape of the forebrain of G. traquairi resembles that of various earlier-diverging dicynodonts, such as Niassodon mfumukasi, Pristerodon mackayi, Rastodon procurvidens, and Diictodon felicepsOwen, 1876, as well as the dicynodontoid Lystrosaurus (Edinger 1955, Hopson 1979), rather than the bulbous and proportionally large forebrains of the cistecephalid emydopoids Kawingasaurus fossilis and Kembawacela spp. (Castanhinha et al. 2013, Laaß 2015, Laaß and Kaestner 2017, Laaß et al. 2017, Angielczyk et al. 2019, Simão-Oliveira et al. 2019, Araújo et al. 2022). There is no bisected swelling in the forebrain of G. traquairi, again similar to N. mfumukasi, P. mackayi, R. procurvidens, D. feliceps, and Lystrosaurus, and unlike K. fossilis and Kembawacela spp. (Edinger 1955, Castanhinha et al. 2013, Laaß 2015, Laaß and Kaestner 2017, Laaß et al. 2017, Angielczyk et al. 2019, Simão-Oliveira et al. 2019, Araújo et al. 2022). The most distinguishing characteristic of the endocast of G. traquairi is the well-developed pineal body (Fig. 11E). In lateral view, the pineal body protrudes anterodorsally within the mediolaterally thin sagittal crest, and has an anteriorly directed projection at its dorsal end, giving it a triangular shape. The anterior side of the triangular dorsal end is exposed to the outside of the skull through the pineal foramen immediately anterior to the sagittal crest. This morphology is unlikely to be a preservational artefact, but instead largely reflective of the true endocast morphology, as the element is entirely enclosed by preserved skeletal elements that are not drastically taphonomically altered. The anterior direction of the pineal body can be attributed to the presence of the tall, narrow sagittal crest; the same structure in Lystrosaurus (where no sagittal crest is present) is a simple, vertical tube (Edinger 1955).

The midbrain forms the ventrally directed portion of the endocast leading towards the hindbrain at the posterior end of the endocast (Fig. 11E). The midbrain is strongly dorsoventrally angled, creating an almost 45^º angle between the forebrain and hindbrain. Such an angle between the forebrain and hindbrain can be observed in N. mfumukasi, P. mackayi, D. feliceps, K. fossilis, and Lystrosaurus, but not in R. procurvidens (although this may be related to taphonomic dorsoventral flattening of the specimen used to generate the endocast) (Edinger 1955, Castanhinha et al. 2013, Laaß 2015, Laaß and Kaestner 2017, Laaß et al. 2017, Simão-Oliveira et al. 2019). Dorsal to the ventrally directed portion of the endocast, a small posteriorly directed projection can be identified as an ‘unossified zone’, a feature usually seen in dicynodont endocasts and suggested to house vascular sinuses (Laaß 2015, Laaß et al. 2017, Simão-Oliveira et al. 2019). The midbrain of G. traquairi is not as dorsoventrally elongate as that of D. feliceps (Laaß et al. 2017). The hypophysis of G. traquairi is identifiable as a ventral extension of the posterior half of the endocast, similar to that of N.mfumukasi, P. mackayi, R. procurvidens, and D. feliceps (Castanhinha et al. 2013, Laaß 2015, Laaß et al. 2017, Simão-Oliveira et al. 2019).

The hindbrain of G. traquairi is mediolaterally wider than the forebrain and the descending portion of the midbrain (Fig. 11E). This is unlike the endocasts of N.mfumukasi, P. mackayi, R. procurvidens, D. feliceps, and K. fossilis (Castanhinha et al. 2013, Laaß 2015, Laaß and Kaestner 2017, Laaß et al. 2017, Simão-Oliveira et al. 2019, Araújo et al. 2022). In lateral view, no clear parafloccular lobe can be observed, and there is no evidence of a parafloccular fossa (which encloses the lobe) in the supraoccipital. It is possible that Gordonia had no parafloccular fossae, which are greatly reduced in other bidentalian dicynodonts (Angielczyk and Kurkin 2003). However, a proportionally small parafloccular lobe might not be detectable with the available data for this specimen. All other published dicynodont endocasts have parafloccular lobes protruding laterally from the hindbrain (Castanhinha et al. 2013, Laaß 2015, Laaß and Kaestner 2017, Laaß et al. 2017, Simão-Oliveira et al. 2019). No medulla oblongata, such as that of R. procurvidens (Simão-Oliveira et al. 2019), can be confidently identified in the endocast of G. traquairi. Additionally, no longitudinal medial sulcus can be identified separating the ventral surface of the hindbrain (where the pons is inferred to be), unlike that of R. procurvidens (Simão-Oliveira et al. 2019).

The left vestibule is anteroposteriorly short in left lateral view (Fig. 11E), unlike that of N. mfumukasi, P. mackayi, R. procurvidens, and D. feliceps, which have anteroposteriorly longer vestibules relative to total endocast size (Castanhinha et al. 2013, Laaß 2015, Laaß et al. 2017, Simão-Oliveira et al. 2019). The left vestibule of G. traquairi is not inflated, unlike that of K. fossilis and Kembawacela spp. (Laaß and Kaestner 2017, Angielczyk et al. 2019, Araújo et al. 2022). Furthermore, the vestibule of G. traquairi does not taper in anteroposterior length down its dorsoventral length, as in N. mfumukasi, P. mackayi, and R. procurvidens (Castanhinha et al. 2013, Laaß 2015, Simão-Oliveira et al. 2019). In posterior view (Fig. 11C), the vestibule of G. traquairi is almost as mediolaterally wide as the hindbrain, which is unlike N. mufumukasi, P. mackayi, R. procurvidens, and D. feliceps, which have hindbrains that are considerably mediolaterally wider than their vestibules (Castanhinha et al. 2013, Laaß 2015, Laaß et al. 2017, Simão-Oliveira et al. 2019). The skeletal elements of G. traquairi surrounding the vestibule are not taphonomically altered to an extent that would cause such extreme mediolateral widening of the vestibule (e.g. dorsoventral flattening of the skull), and this wide morphology likely approximates the true vestibular morphology as it fills most of the prootic region.

Body mass estimates and encephalization quotient values of Gordonia

Body mass estimates ranged from 7726.81 g to 86807.11 g (Table 1). Since the greatest value is about 3.8 times larger than the next greatest value (22999.57 g), and there are substantially smaller differences between the other estimates, the greatest estimate was excluded from calculation of the average body mass value (11882.01 g). The endocast volume without the pineal body (9850.9 mm³) is about 90% of the volume with the pineal body (10930.04 mm³), and consequently, the Manger EQ value calculated without the pineal body (0.2) is about 91% of the value calculated with the pineal body (0.22).

Phylogenetic analysis results

The only notable differences between the scorings for the ‘Gordonia traquairi’ OTU and ‘The Elgin Marvel’ OTU that are not related to missing data for either OTU are continuous characters 12 (angle between ascending and zygomatic processes of the squamosal is 1.396 in the former OTU and 2.379 in the latter OTU) and 13 (angulation of the occiput relative to the palate, expressed as the ratio of dorsal and basal lengths of the skull is 1.067 in the former OTU and 1.49 in the latter OTU). The phylogenetic analysis resulted in 1 most parsimonious tree (MPT) with a length of 1370.86 steps (consistency index = 0.229, retention index = 0.708) (Fig. 12). ‘The Elgin Marvel’ OTU is robustly recovered as the sister-taxon to the ‘Gordonia traquairi’ OTU (recovered in 97 replicates, Bremer support value = 3.317), supporting the referral of ELGNM 1999.5.1 to Gordonia traquairi.

Furthermore, Gordonia is recovered within a clade also containing Jimusaria (recovered in 60 replicates, Bremer support value = 2). Synapomorphies uniting this clade are continuous Ch. 8: 0.117–0.118 → 0.094–0.113 (narrow median pterygoid plate relative to basal skull length), continuous Ch. 11: 9.599–9.663 → 9.319 (area of internal nares small relative to basal skull length), continuous Ch. 12: 9.9–10.25 → 8.9 (small angle between ascending and zygomatic process of the squamosal), continuous Ch. 13: 0.920–0.926 → 0.944-0.922 (strong angulation of the occiput relative to the palate), discrete Ch. 38: 1 → 0 (absence of prefrontal bosses), discrete Ch. 45: 2 → 1 (preparietal present and flush with skull roof), discrete Ch. 67: 1 → 0 (squamosal separated by tabular bone from supraoccipital), discrete Ch. 107: 1 → 0 (tabulars contact opisthotics), and discrete Ch. 172: 1 → 0 (unornamented anterior face of dentary symphysis). Within this clade, Jimusaria sinkianensis is recovered immediately outside of a subclade containing Jimusaria monanensis and Gordonia (recovered in 56 replicates, Bremer support value = 1.889). This result mirrors Kammerer (2019), Angielczyk et al. (2021), and Liu (2021) in recovering Gordonia as the sister-taxon to J. sinkianensis, but not Kammerer and Ordoñez (2021), Macungo et al. (2022), and Shi and Liu (2023). Instead, these studies recover J. sinkianensis (Jimusaria spp. in the case of Shi and Liu (2023)) immediately outside of the clade containing Gordonia and ‘higher’ dicynodontoids (e.g. Kannemeyeriiformes).

Moreover, our phylogenetic analysis recovers a relatively inclusive Lystrosauridae containing Peramodon, Daptocephalus, Dinanomodon, Turfanodon, the Jimusaria + Gordonia clade, Syops, Basilodon, Sintocephalus, Euptychognathus, and Lystrosaurus [all taxa more closely related to Lystrosaurus than to Kannemeyeria or Dicynodon; see Kammerer and Angielczyk (2009) for clade definitions], but this large Lystrosauridae clade is not strongly supported (recovered in <50 replicates, Bremer support value = 0.806). Synapomorphies uniting this version of Lystrosauridae in our analysis are continuous Ch. 3: 0.253 → 0.23–0.244 (small width of interorbital skull roof relative to basal length of skull), continuous Ch. 7: 1.468–1.485 → 1.380–1.443 (dorsoventrally short anterior pterygoid keel in lateral view relative to height of non-keel ramus), discrete Ch. 98: 2 → 1 (basisphenoid contribution to the basisphenoid–basioccipital tuber slopes anterodorsally at a steeper angle such that the parabasisphenoid contribution is still somewhat ridge-like but the portion of the ridge on the anterior surface of the tuber is more vertically-oriented), discrete Ch. 142: 2 → 3 (number of sacral vertebrae is six or more), discrete Ch. 165: 1 → 0 (proximal articular surface of the femur present as a weak swelling that is mostly limited to the proximal surface of the bone).

Recovery of Gordonia and Jimusaria within Lystrosauridae is a result not found in other recent phylogenetic analyses (Kammerer 2019, Angielczyk et al. 2021, Kammerer and Ordoñez 2021, Liu 2021, Macungo et al. 2022, Shi and Liu 2023), although the recovery of Jimusaria in this clade has previously been obtained by Kammerer et al. (2011). Additionally, Lystrosauridae and Kannemeyeriiformes represent sister-taxa in the recovered topology, a result previously obtained by Macungo et al. (2022), but unlike other recent topologies that found Kannemeyeriiformes to be the sister-taxon of a Gordonia + J. sinkianensis clade (Kammerer 2019, Kammerer and Ordoñez 2021, Liu 2021), or Kannemeyeriiformes to be the sister-taxon of a clade containing Basilodon, Sintocephalus, Peramodon, Daptocephalus, Dinanomodon, and Turfanodon (Shi and Liu 2023).

The stratigraphically problematic Chinese taxon Kunpania scopulusa is here recovered as the sister-taxon to all other bidentalians, rather than in the basal dicynodontoid position found by Angielczyk et al. (2021). Angielczyk et al. (2021) commented on the extremely weak support for Kunpania as a dicynodontoid, so this minor lability in its position on the tree is unsurprising. The more basal position of Kunpania would be consistent with an older age for the unit from which it is known (Upper Quanzijie Formation; Angielczyk et al. 2021).

Jackknife resampling and/or Bremer supports reveal that many of the recovered clades in the analysis herein are unstable, including Bidentalia (recovered in <50 replicates, Bremer support value = 0.35), Cryptodontia (recovered in <50 replicates, Bremer support value = 1.864), Dicynodontoidea (recovered in <50 replicates, Bremer support value = 0.350), and Kannemeyeriiformes (recovered in 61 replicates, Bremer support value = 0.350).

DISCUSSION

Evolutionary relationships of Gordonia and late Permian biogeography

A notable find of the phylogenetic analysis herein that is mirrored in some other studies (Kammerer 2019, Angielczyk et al. 2021, Kammerer and Ordoñez 2021, Liu 2021, Macungo et al. 2022) is that Gordonia forms a clade with Jimusaria (Fig. 12). Of the synapomorphies uniting this clade, the narrow median pterygoid plate relative to basal skull length (continuous Ch. 8: 0.117–0.118 → 0.094–0.113) is of particular interest. In both Gordonia and Jimusaria, the median pterygoid plate is very narrow compared to other Permian dicynodontoids (e.g. Dicynodon, Turfanodon, and Daptocephalus), indicating that these two taxa comprise a subclade of dicynodontoids characterized by this anatomical trait. Gordonia and Jimusaria also have anterior pterygoid rami that do not flare laterally to the same degree as other dicynodontoids [although IVPP V31929 (referred to Jimusaria monanensis) has anterior rami that flare more laterally than other Jimusaria specimens], and somewhat similar lateral dentary shelves (that of Gordonia is more strongly anterodorsally angled) (Shi and Liu 2023). Additionally, Kammerer et al. (2011) note that J. sinkianensis has a narrow snout and Shi and Liu (2023) describe J. sinkianensis as having a narrow sagittal crest with only slight dorsal exposure of the parietal, both characteristics identifiable in Gordonia. Further investigation of the similarities and differences between these two taxa is needed to test the validity of this newly proposed subclade. The other synapomorphies uniting the Gordonia and Jimusaria clade are not unique to this clade, but their distribution in Dicynodontoidea is worth exploring in greater detail. There is also taxonomic importance to such investigation since our phylogenetic analysis recovers J. sinkianensis immediately outside of a clade containing Gordonia and J. monanensis, raising doubt about the monophyletic nature of Jimusaria (in which case it could be synonymized with Gordonia, which has priority). It is possible that this subclade of dicynodontoids might have also been one of several to cross the Permo-Triassic boundary (alongside Lystrosaurus spp. and Kannemeyeriiformes), as the holotype of J. sinkianensis was collected from the Guodikeng Formation (Angielczyk et al. 2022), which has recently been confirmed to span the Permo-Triassic boundary (Yang et al. 2021). However, compounding geological factors and imprecise historical records have made reconstructing statigraphic relationships difficult in the Guodikeng Formation, so a Permian age for J. sinkianensis cannot be ruled out (Yang et al. 2021, Angielczyk et al. 2022). Similarly, the Laotian dicynodontoids Counillonia superoculisOlivier et al., 2019 and Repelinosaurus robustusOlivier et al., 2019 are from strata (the Purple Claystone Formation; Olivier et al. 2019) that have recently been shown to span the Permo-Triassic boundary (if not be entirely Early Triassic) (Rossignol et al. 2016), but given the range of error involved in radiometric dates it has been suggested that they are actually Permian (Liu 2020b).

A close relationship between Gordonia (from Elgin, Scotland, United Kingdom), J. sinkianensis [from Dalongkou, Xinjiang, China (Li et al. 2008, Angielczyk et al. 2022)], and J. monanensis [Tumed Right Banner, Nei Mongol, China (Shi and Liu 2023)] could be evidence of a Laurasian dicynodontoid clade. In general, dicynodonts were widespread animals and phylogenetic analyses for this group indicate a number of trans-hemispheric sister-taxon relationships. For example, lineages branching off from Dicynodontoidea phylogenetically proximal to Gordonia according to our phylogenetic analysis contain species so far only known from Russia (Peramodon amalitzkiiSushkin, 1926) and southern Africa (Dinanomodon gilli and Daptocephalus spp.). This adds to a growing body of evidence for Permian therapsids having close relatives on far sides of the world, including other evidence from the Scottish ‘Elgin Reptiles’. Namely, the cryptodont dicynodont Geikia is known from species in Scotland and Tanzania: G. elginensis and G. locusticeps Huene, 1942, respectively (Newton 1893, Huene 1942). The pylaecephalid dicyndont Diictodon feliceps is also known from both South Africa and China (Angielczyk and Sullivan 2008). The South African cryptodont TropidostomaOwen, 1876, has consistently been found to be the sister-taxon to the Russian AustralobarbarusKurkin, 2000 (e.g. Kammerer et al. 2011, 2015, Kammerer and Smith 2017). Other Permian therapsid clades show similar distributions. The akidnognathid therocephalians AnnatherapsidusKuhn, 1963 and ShiguaignathusLiu and Abdala, 2017 are closely related and the former is from Western Russia, while the latter is from Nei Mongol, China (Ivakhnenko 2011, Liu and Abdala 2017). Another akidnognathid taxon, Euchambersia, is known from two species: E. mirabilisBroom, 1931 from South Africa and E. liuyudongiLiu and Abdala, 2022 from Nei Mongol. The rare therocephalians IchibengopsHuttenlocker et al., 2015, and MupashiHuttenlocker and Sidor, 2016, which are endemic to the Luangwa Basin of Zambia, have been recovered as related to Russian taxa (Huttenlocker and Sidor 2016). Even the gorgonopsian InostranceviaAmalitzky, 1922, long considered a Russian endemic taxon, has been found in South Africa (Kammerer et al. 2023). Similar patterns are observed in non-synapsid tetrapods as well. The Scottish parareptile Elginia (a contemporary of Gordonia from the Cutties Hillock Formation) has also been discovered in Nei Mongol, China (Liu and Bever 2018), and closely related chroniosuchians are known from both European Russia, the Henan Province of China, and Xinjiang (including the same stratigraphic section that has yielded J. sinkianensis) (Buchwitz et al. 2012, Liu 2020a). This body of evidence strongly indicates that the biogeography of late Permian tetrapods is complex, with taxa diversifying and quickly spreading to distant regions across Pangea. Future research that takes a closer look at the phylogenetics and biogeography of late Permian tetrapods, climate, and precipitation and other environmental factors across Pangea could shed light into these Permian distribution patterns.

The relatively inclusive Lystrosauridae recovered in our analysis is not well supported (recovered in <50 replicates, Bremer support = 0.81) and the synapomorphies uniting the clade are not unambiguously unique to its members. Instability in this group is related to larger issues involving high lability in Dicynodontoidea and potential paraphyly of Cryptodontia; for recent discussion see: Angielczyk and Kammerer (2017), Angielczyk et al. (2021), and Shi and Liu (2023).

Development and evolution of the non-mammalian therapsid pineal body

The pineal body of Gordonia is the most remarkable feature of its endocast, with its enlarged anterodorsal projection and triangular shape (Fig. 11E) being the most prominent differences from other therapsids with reconstructed endocasts (e.g. Edinger 1955, Rowe et al. 2011, Benoit et al. 2017a). This morphology is here hypothesized to be a developmental artefact related to exaggeration of the sagittal crest. ELGNM 1893.6 ("G. juddiana") lacks a prominent sagittal crest, and instead shows a temporal bar that is not dorsally offset from the rest of the skull roof. Considering that ELGNM 1893.6 ("G. juddiana") is a smaller specimen than ELGNM 1999.5.1, and as such probably represents an earlier ontogenetic stage, sagittal crest development is probably an allometric feature in Gordonia [as in other non-mammalian synapsids with known growth series (Jasinoski et al. 2015, Jasinoski and Abdala 2016)]. ELGNM 1893.6 ("G. juddiana") also has a pineal foramen angled perpendicular to the skull roof, unlike ELGNM 1999.5.1 and other larger G. traquairi specimens in which the pineal foramen is angled anterodorsally. This is also related to exaggeration of the sagittal crest, with increased height and curvature of the intertemporal bar resulting in changes to the position and orientation of the foramen. In association with these changes, the pineal body of Gordonia would have enlarged, elongated, and shifted from being a tubular and vertical structure (as in, e.g. Lystrosaurus; Edinger 1955) to an anterodorsally expanded structure in order to connect the pineal foramen to the brain. The anterodorsal angulation of the foramen is likely also related to the anterior expansion of the pineal body that grants it the triangular shape in lateral view, which is unlikely to be a preservational artefact as the skull roof is well-preserved. If this hypothesis is correct, it implies a substantial degree of neuroanatomical plasticity in early therapsids.

Many other non-mammalian therapsids also had distinctive pineal bodies (Fig. 13), most also in association with elaboration of the skull roof. In the dinocephalian MoschopsBroom, 1911, a taxon with intense cranial pachyostosis resulting in the formation of a frontoparietal ‘dome’, the pineal body is greatly elongated, occupying a ‘pineal tube’ that extends through the ‘dome’ (Benoit et al. 2017b). Rhachiocephalid dicynodonts (KitchinganomodonMaisch, 2002 and RhachiocephalusSeeley, 1898) probably also had an elongated pineal body, as they possess narrow temporal bars with pineal foramina strongly angled anteriorly. As can be seen in the Rachiocephalus magnusOwen, 1876 specimen BP/1/1512, which provides a cross-sectional lateral view of the midline of the skull, the lumen of the pineal body is elongate (Supporting Information, Fig. S30). In the endocasts of the Triassic bidentalians Lystrosaurus and PlaceriasLucas, 1904 (based on natural casts and cranial sections, not CT data), the pineal bodies also have an elongate, dorsally projecting morphology (Simão-Oliveira et al. 2019). The tall sagittal crest of Placerias and other kannemeyeriiforms is related to the enlarged jaw musculature in this group (Angielczyk et al. 2018). In the case of Gordonia, the sagittal crest would have supported large masticatory muscles, which implies that the pineal body may have been reshaped as a consequence of feeding or trophic adaptations in this taxon. As such, we consider the remarkable pineal body of G. traquairi to most likely be a ‘spandrel’ (sensuGould and Lewontin 1979) that is a by-product of other factors rather than an adaptive feature.

Ontogenetic variation in the pineal body in early therapsids is currently poorly understood; we suspect that the elaborate morphologies of the pineal bodies in the aforementioned taxa developed late in ontogeny, but this needs to be tested by CT-scanning growth series. Excellent growth series are known for Lystrosaurus (Ray 2005), AulacephalodonSeeley, 1898 (Tollman et al. 1980), and Diictodon (Ray and Chinsamy 2004). Substantial variation in skull size and degree of pachyostosis is also present in the known sample of Moschops (Broom 1911), so the record is amenable to such a test. Many non-mammalian therapsids with less baroque skull roof morphologies had smaller, less elongate pineal bodies, and exaggerated versions of this structure probably evolved multiple times within Therapsida. Variation in this structure between closely related therapsid taxa and potentially during the growth history of single individuals has confounding implications for analyses of relative endocast size, which, combined with the probably minimal contribution of actual neural tissue to this region (which may have been mostly glandular tissue), means that therapsid EQs should be calculated with the pineal body volume excluded.

Variation in encephalization quotient throughout Therapsida

The EQ of Gordonia (0.22 with pineal body, 0.2 without pineal body) is within the range for most non-mammalian therapsids (approximately 0.05–0.2) but greater than all other sampled dicynodonts (0.06–0.17) with the exception of Kawingasaurus (0.43), which is an extreme outlier for the clade that is more similar to that of derived cynodonts closely related to crown mammals, such as HadrocodiumLuo et al., 2001 (0.51) and MorganucodonKühne, 1949 (0.31) (Fig. 14). Although Laaß and Kaestner (2017) argue that the increase in brain size in Kawingasaurus was an adaptation for a specialized fossorial lifestyle, a similar argument cannot be made for Gordonia. There is no direct evidence indicating that G. traquairi was a quasi- or obligately fossorial taxon, as has been suggested for cistecephalids (such as Kawingasaurus) (Cox 1972, Angielczyk et al. 2019, Kammerer 2021, Macungo et al. 2022).

The considerable difference between the EQ of Gordonia and the most phylogenetically proximate of the sampled taxa, Lystrosaurus (0.06), is noteworthy (Fig. 14). However, it should also be noted that measurement of the endocast volume of Lystrosaurus was carried out with historical techniques rather than through modern CT scanning and processing, which might contribute to the drastic difference in EQ (Quiroga 1980). Additionally, the Gordonia endocast is incomplete (in particular the olfactory tract and bulbs), which might actually mean that EV is being underestimated, and thus EQ as well. The large difference in EQ between these two closely related taxa and between dicynodontoids and cistecephalids is a reminder that more neuroanatomical data, from a broader range of species over time, is needed for dicynodonts. With this said, substantial uncertainty related to EQ patterns in synapsids is probably inescapable, given the incomplete ossification of their braincases making exact determination of endocranial volume impossible for most taxa.

There is no clear gradual increase in proportional brain size leading up to cynodonts from basal therapsids (Fig. 14). This is drastically unlike what is seen within Cynodontia, where brain size, and other neuroanatomical characteristics, increase leading up to Mammalia (although whether this increase occurred over evolutionary pulses or gradually is unclear) (Rowe et al. 2011, Rowe and Shepherd 2016, Rowe 2017, Hoffmann and Rowe 2018, Wallace et al. 2019, Benoit et al. 2023, Kerber et al. 2023). Instead, non-cynodont therapsids had a general range of EQs (0.05–0.2) that varied within and between clades. A notable exception to this is the clade Dinocephalia, which is characterized by very low EQs (0.02–0.03). Additionally, there are two species with EQ values greater than 0.3 in non-mammaliaform therapsids: the Late Triassic cynodont TherioherpetonBonaparte and Barberena, 1975 (0.32) and the aforementioned dicynodont Kawingasaurus (0.43). These taxa might have required these proportionally enlarged brains to support dedicated fossorial and/or social lifestyles (Laaß and Kaestner 2017).

It is also worth noting that conclusions regarding EQs based solely on endocast data are not equivalent to those based on brain tissue. Since the endocast shapes of many non-mammalian therapsids, especially most dicynodonts, resemble those of non-avian reptiles, it is plausible to infer endocast fill in non-mammalian therapsids might have been more similar to that of reptiles, which unlike mammals, do not fully or nearly fully fill their endocranial cavity with brain tissue (Castanhinha et al. 2013, Caspar et al. 2024). This is corroborated by the fact that clear fissures are usually not discernible in the endocasts of non-mammalian therapsids, similar to non-avian reptiles, but not mammals. These fissures indicate sulci, and the presence of them in endocasts suggests that almost all the endocast is filled by brain tissue (Laaß and Kaestner 2017). Also, there might have been a gradual increase in endocast fill as therapsids became more mammal-like across their evolutionary history. This would mean distantly related taxa, such as the biarmosuchian LemurosaurusBroom, 1949 and the cynodont DiademodonSeeley, 1894 might have similar EQ values according to the data presented herein, but also could have had drastically different degrees of endocast fill. Furthermore, endocast fill can change drastically across ontogenetic stages of modern reptiles as well, e.g. in crocodilians where the brain can fill anywhere from 29% to 95% of the endocast depending on ontogeny (Barrios et al. 2023, Ferreira et al. 2023). Whether the brain fill of non-mammalian therapsids changed to such a large extent throughout development is unknown. Such dramatic possibilities and more conservative alternatives should not be ruled out when studying non-mammalian therapsid neuroanatomy.

As a final caveat, reliable body mass estimates are required for accurate and precise calculations of EQ. Recent research has shown that estimating body mass with volumetric methods yields values that are within much smaller margins of error than those estimated with regression formulae (Sellers et al. 2012, Bates et al. 2015, Brassey et al. 2015, Brassey 2016, Romano and Manucci 2019, Romano and Rubidge 2019, Romano et al. 2019, Van den Brandt et al. 2023). Additionally, mass estimates made with volumetric methods are often substantially lower than those made with regression formulae in both non-therapsid tetrapod (Bates et al. 2015, Brassey et al. 2015, Van den Brandt et al. 2023) and therapsid taxa (Romano and Manucci 2019, Romano and Rubidge 2019), indicating that values derived through regression formulae will often be overestimates. Nevertheless, given that it is much more feasible and time-efficient to derive body mass estimates from regression formulae, and that very few taxa preserve the complete skeletons required for rigorous volumetric analysis, it is unlikely that proxy-based mass estimates using regressions will and should be wholly abandoned among palaeobiologists

CONCLUSION

The use of µCT scanning has allowed for a new look at the external and internal skull anatomy of one of the only known dicynodonts from Western Europe, Gordonia (Fig. 15). The incorporation of the new anatomical data into a phylogenetic analysis provides an updated test of the evolutionary relationships of Gordonia, and suggests it forms a clade with the Chinese Jimusaria, expanding our understanding of late Permian therapsid biogeography. Furthermore, the digital endocast of the brain and vestibule of G. traquairi are the first generated for this taxon, and the first to be published for a bidentalian dicynodont. The pineal body of Gordonia has an unusual morphology (that can also be seen in other non-mammalian therapsids), which can be explained as an accommodation to fit a skull with a proportionally large sagittal crest. Moreover, the EQ of Gordonia is within the range of most non-mammalian therapsids, and substantially less than that of mammals and other derived cynodonts. However, due to the many gaps in our knowledge of non-mammalian therapsid encephalization, there is much uncertainty surrounding how much EQ data represents neurological variation between taxa.

ACKNOWLEDGEMENTS

This publication is based on the MRes thesis (part of the Palaeontology and Geobiology MScR programme) of H.G. at the University of Edinburgh. We are extremely grateful to Alison Wright, David Longstaff, and Janet Trythall of the Elgin Museum Geology Group for providing access to, and allowing transport of, the ‘The Elgin Marvel’, and to Elizabeth Martin-Silverstone for scanning the specimen at the University of Bristol Palaeobiology Laboratories. We extend our gratitude to David Longstaff for guiding H.G. through Clashach quarry to better understand the geological setting, and to Alison Wright, Mathew Lowe (University of Cambridge), and Michael Day (Natural History Museum, London) who provided access to specimens for H.G. to examine. Additionally, we thank Paige dePolo (University of Edinburgh) for her assistance in exporting the digital models used in this study, and Candice Stefanic (Stony Brook University) for her advice regarding preparing digital models for publication. We also thank Daniel de Simão-Oliveira (Universidade Federal de Santa Maria) for permission to use the template in Figure 13, and Scott Reid for creating the brilliant artwork used in Figure 15. We also thank Julien Benoit for pointing out a taxonomic error in Figure 13. We are grateful to Kenneth Angielcyzk (Field Museum) and an editor of ZJLS for reviewing this work, which has substantially improved its quality. H.G. is supported by a University of Bristol Scholarship. D.F.’s work on using CT to study the Elgin Reptiles was supported by the Royal Commission for the Exhibition of 1851 Science Fellowship. S.L.B.’s work on brain evolution in tetrapods is supported by the Swedish Research Council (‘The Evolution of Minds’ project) and his work on early mammal evolution is supported by a European Research Council Starting Grant (PalM, number 756226). Finally, Clare Clark is thanked for intervening in 1997 to prevent the traditional, more destructive analysis of the ‘The Elgin Marvel’ and suggesting medical scanning instead.

CREDIT STATEMENT

Hady George (conceptualization, data curation, formal analysis, investigation, methodology, writing original draft, writing review, and editing), Christian F. Kammerer (conceptualization, data curation, formal analysis, investigation, methodology, writing original draft, writing review, and editing), Davide Foffa (conceptualization, data curation, formal analysis, investigation, methodology, writing original draft, writing review, and editing), Neil D.L. Clark (writing original draft, writing review, and editing), and Stephen L. Brusatte (conceptualization, data curation, formal analysis, investigation, methodology, writing original draft, writing review, and editing).

CONFLICT OF INTEREST

No conflicts of interest to be declared.

FUNDING

University of Bristol Scholarship—PhD funding to H.G..

The Royal Commission for the Exhibition of 1851—Fellowship to D.F. (CT scans).

Swedish Research Council—Grant to S.L.B..

European Research Council—Grant to S.L.B..

DATA AVAILABILITY

The µCT data used in this study and 3D reconstructions are available at https://www.morphosource.org/projects/000586798?locale=en

REFERENCES