CRANIAL ANATOMY OF THE LATE PERMlAN DICYNODONT DIICTODON,

AND ITS BEARING ON ASPECTS OF THE TAXONOMY, PALAEOBIOLOGY AND

PHYLOGENETIC WATIONSHIPS OF THE GENUS

by

Corwin Sullivan

A thesis submitted in conformity with the requirements for the degree of Master of

Science, Graduate Department of Zoology,

in the University of Toronto

O Copyright by Corwin Sullivan 2000 National Library Biblioth&que nationale I*I of Canada du Canada Acquisitions and Acquisitions et Bibliographie SeMces services bibliographiques 395 Wellington Street 395, rue Wellington Ottawa ON K1A ON4 Ottawa ON K1A ON4 Canada Canada

The author has granted a non- L'auteur a accordé une licence non exclusive licence allowing the exclusive permettant à la National Library of Canada to Bibliothèque nationale du Canada de reproduce, loan, distribute or sell reproduire, prêter, distribuer ou copies of this thesis in microform, vendre des copies de cette thèse sous paper or electronic formats. la forme de microfiche/film, de reproduction sur papier ou sur format électronique.

The author retains ownership of the L'auteur conserve la propriété du copyright in this thesis. Neither the droit d'auteur qui protège cette thèse. thesis nor substantial extracts fiom it Ni la thèse ni des extraits substantiels may be printed or otherwise de celle-ci ne doivent être imprimés reproduced without the author's ou autrement reproduits sans son permission. autorisation.

Canada Cranial anatomy of the Late Permian dicynodont Diictodon, and its bearing on aspects of

the taxonomy, palaeobiology and phylogenetic relationships of the genus

Corwin Sullivan

Master of Science 2000

Department of Zoology, University of Toronto

ABSTRACT

A study of the Late Permian dicynodont Diictodon confirrns suggestions that only one species, D. feliceps, can be presently recognized. The genus is charactenzed by features such as a palatal notch, a large dentary table with a media1 cutting blade. and the absence of postcanine teeth.

Numerous anatomical variations exist within Diictodon, but many appear biologically insignificant. However, there is a clear distinction between specimens having canine tusks and those lacking hem, as tusked specimens are generdly larger and more likely to develop a pineal boss. This probably reflects sexual dimorphism, with the tusked sex almost certainly being the male.

A phylogenetic analysis of dicynodonts indicates that previous analyses are correct in identifying Robenia as the sister-group of Diicfodon and Diqnodon and Aulacephalodon as close relatives of Lystrosaum .However, the results suggest that Endothiodon, rather than Eodicynodon, may be the most basai known dicynodont. TABLE OF CONTENTS

Page Introduction 1 Materials 9 S ystematic Palaeontology 13 Osteological Description 19 Intrageneric Variation 9 1 Dicynodont Interrelationships 132 Acknow ledgements 160 Literature Cited 161 Appendix 1 - Information on Diictodon specimens examined in this study 177 Appendix 2 - Definitions of characters used in the phylogenetic analysis 182 Appendix 3 - Data matrix for the phylogenetic anaiysis 187

iii LIST OF UUSTRATIONS Page Figure 1. Biostratigraphy of the Beaufort Group Il Figure 2. Diictodon feliceps. Reconstruction of skull in dorsal view 23 Figure 3. Diictodon feliceps. Reconstruction of skull in lateral view 25 Figure 4. Diictodon feliceps. Reconstruction of skull in ventral view 27 Figure 5. Diictodon feliceps. Reconstruction of skull in occipital view 39 Figure 6. Diictoàonfeliceps. Reconstruction ol mandibic in donal view 3 1 Figure 7. Diictodon feliceps. Reconstruction of mandible in lateral view 33 Figure 8. Diictodon feliceps. Reconstruction of mandi ble in ventral view 35 Figure 9. Diictodon feliceps (R97.1). SkulI in donal view 37 Figure 10. Diictodon feliceps (R 97.1). Skull (with mandible) in lateral view 39 Figure 1 1. Diictodon feliceps (UT Von Huene 1922 r. 1-4). Skull in ventral view 4 1 Figure 13. Diictodon feliceps (UT Von Huene 1922 r. 1-4). Skull in occipital 43 view Figure 13. Diictodon feliceps (R 97.2 and SAM-PKX7730). Suborbital region in posterolaterai view 93 Figure 14. Diictodon feliceps (R 97.2 and SAM-PKX7730). Intertemporal region in donal view 95 Figure 15. Diictodon feliceps (R97.1 and SAM-PK-K7795). Snout in donal view 97 Figure 16. Reconstmcted skull of D. feliceps showing morphomeuic measuremen ts 106 Figure 17. Results of a principal components analysis of morphornetric cranial proportions in Diictodon feliceps 112 Figure 18. Cladogram showing dicynodont interrelationships (King, 1988) 134 Figure L9. CIadogram showing relationships among major groups of dicynodonts (Cox, 1998) 136 Figure 20. Cladogram showing dicynodont interrelationships as determined in this study 142 LIST OF TABLES Page Table 1. List of quditative and quantitative variations in the cranial anatomy of Diictodon feliceps 103 Table 2. Co-occurring matornical features in the skull of Diictodon feliceps 110 Table 3. Morphometnc principal components results for the skull of Diictodon fer iceps 114 Table 4. Statistical support for various clades obtained in a phylogenetic analysis of dicynodonts 144 Table A 1. Information on individual Diictodon feliceps specimens exmined in this study 178 Table A?. Data matrix for a phylogenetic analysis of dicynodont 188 interrelationships LIST OF ABBREViATIONS

Abbreviations used in Figures

A Angular ANT PL Antenor plate of sphenethmoid ANT R Anterior ridge (on palatal surface of premaxilla) ART Articular BO Basioccipital CULT PR Cultriform process (of pmbasisphenoid) D Dentary DEN TAB Dentary table ECT Ectopterygoid EO Exoccipi ta1 EP Epipterygoid F Frontal FEN OV Fenestra ovalis FOR MAG Forarnen magnum ICC Internai carotid canal INT VAC Interpterygoid vacui ty JU Jugal rCTG FOR Jugular forarnen L Lacrimai L FOR Lacrimai forarnen L FS Labial fossa L PAL FOR Laterd paiatal foramen M Maxilla MAN FEN Mandibular fenestra N Nasal N. VII Foramen for cranial nerve VI1 NAS BOSS Nasal boss P Parieta1 PAL Palatine PAL NO PaiataI notch PBS Parabasisphenoid PLANT Pila antotica PIN BOSS Pineal boss PIN FOR Pineai foramen PMX Premaxilla PO Postorbital POF Postfrontal POP Postparietal POS R Posterior ndge (on palatal surface of premaxilla) PP Preparie t al PR Pro0 tic PRF Prefron ta1 PRS Presphenoid PT Pterygoid PTF Posttempord fenestra Q Quacirate QJ Quadratojugal REF LAM Reflected lamina (of angular) SA Surangular sa Scleml ossicle SO Supraoccipi ta1 SP Splenial SQ Squarnosal ST Stapes T Tabular v Vomer VLTUB Ventrolaterd tubcr

Institutional abbrevations

AMNH Arnencan Museum of Natural History, New York Bm Namal History Museum, London BPI Bernard Price lnstitute for Palaeontological Research, Johannesburg BSP Bayerische Staatssarnrnlung fur Palaonto logie und historisc he Geologie, Munich rVPP Institute for Vertebrate Paieontology and Paleoanthropology, Beijing R Robert R. Reisz temporary field collection, Toronto RC Rubidge Collection, Wellwood, Graaff-Reinet SAM South African Museum, Cape Town TM Transvaal Museum, Pretoria WC University Museum of Zoology, Cambridge USNM Srnithsonian Institution, Washington, D.C. UT Universitat Tübingen Museum und Institute für Geologie und Palaontologie, Tübingen INTRODUCTION

The Beaufort Group strata of South Ahica's Karoo basin have yielded the nchest and most diverse Late Perrnian terrestrial vertebrate palaeofauna in the world. These mainly fluvial deposits contain the remains of numerous species of synapsid amniotes, with less abundant diapsids, anapsids and anamniotes. As investigations of the Karoo pdaeoflora

(e-g., Rayner, 1993) and sedimentology (e.g., Smith, 1993a) proceed dong with srudies of the fossil vertebrates, the emerging picture is that of an ancient ecosystem occupying a region of low-lying floodplains cut by a senes of extensive braided river channels. The vertebnte fauna was supported at the lowest trophic levels by plants ranging from the tree-like Glossopteris to the horsetail-like PlzyZZotlieca to a variety of fems, lycopods and mosses (Rayner, 1992) and was dorninated by a diverse array of therapsids. Most abundant of these were the dicynodonts, a clade of denved anomodonts that seem to have represented the fint major assemblage of terrestrial venebnte herbivores (Hotton, 1986) and were, therefore, of considerable ecological, as well as numencal, importance. Their stratigraphic range extends well into the Triassic and their distribution was more or less worldwide, with occurrences recorded on every continent (King, 1993).

Although it is clear that dicynodonts were both highly diverse and highly abundant, exact quantification of either parameter is dificult. King (1993a) believed dicynodont taxonomy to be plagued by excessive splitting at the generic and particularly the specific levels, but nevertheless recognized no fewer than 35 valid genera. In terms of individual specimens, Watson (1948: p. 845) referred to one Beaufort stratigraphic interval as "the

Endothiodon zone, where anornodonts swarm", whiIe Smith (1993b) found that over 90% of 329 vertebrate fossils he collecteci during a taphonomic study of three Beaufort cliff exposures were dicynodont rernains (mainly isolated skulls). The sheer number of

Permian dicynodont fossils available for study is unparalleled in other Palaeozoic

terrestrial vertebrate groups, and provides an opportunity to gather meaninghil data on

intraspecific and perhaps even intrapopulational osteological variation. Such data are of

obvious palaeobiological interest, but are aiso taxonornically useful in thar they may

allow systematists to clearly cûstinguish genuine ciifferences between species From

variations that result from ontogeny, sexual dimorphism or other intraspecific factors.

The importance of intraspecific variation in dicynodont studies hm only recently

begun to be fully realized. During the early part of the twentieth century, such highly

prolific worken as Broorn (e.g., 1921, 1940) directed their efforts largely toward

recognizing and describing new species, and often did so on the basis of supposed

anatomical differences that later work has shown to be ta~onomicallyunimportant or

even entirely preservational. This approach resulted in the problern of severe oversplitting

noted above, with a total of 54 genera and 263 species listed in the exhaustive

compilation of Haughton and Brink (1954). The existence in the litenture of such an

unmanageable number of species and genera, many of which were not well diagnased,

seems to have impeded attempts to investigate the evolutionary history of the group and

define higher taxa; Watson and Romer (1956) left dicynodonts unclassified above the

generic level, noting the large number of recognized genera.

However, more recent work has done much to improve this situation, as the plethora

of dubious, poorly-known taxa is progressively giving way to a smaller number of clearly

diagnosed species and genera. An important aspect of this process is the realization that

many features in widespread former use as taxonomic charac ten, especidl y details of the sutura1 patterns on the skull roof, are in fact unreliable. Investigators such as Toerien

(1953) and Keyser (1975) demonstrated in various dicynodonts a high degree of inconsistency in the outlines of cranial bones such as the parietal, lacrimal, postfrontal and preparietal, representing a continuum of variation rather than a smail number of discrete States. Other characters, such as the slope of the occiput, the orientation of the tusks, and the shapes of the various skull openings are highly susceptible to distomon, as are linear measurements of cranial proportions. However. some features of the palate

(Toerien, 1953) and the occurrence of structures such as bosses, grooves and forarnina on the exterior of the skull (Keyser, 1975) seem to show a greater degree of consistency and taxonomic relevmce, and offer a firmer basis for classification.

Guided by an emerging consensus as to which characten are likely to be of taxonomic significance (Toerien, 1953; Keyser, 1975; King, 1993b), and by the use of statistical methods to evaluate the importance of morphometric differences (Tollman et al., 1980) modem researchers have succeeded in clarifying many aspects of dicynodont taxonomy and eliminating large numbers of invalid species. The species diversity of some well known genera, suc h as Aulacephalodon and Oudenodon, has been reduced by more than an order of magnitude, and Lystrosaums and Cistecephalus have also been extensively synonymized (King, 1993a). An important corollary of this work is that the differences among the remaining species are generally clearly understood and explicitly stated, as was often not the case in the older literature.

These taxonomic improvements have important implications for studies of dicynodont systematics and palaeobiology. Although several genera are still in need of revision and redescription before they can be fully incorporated into phylogenetic analyses, recent efforts to elucidate the interrelationships of dicynodonts (e.g., Cluver and King, 1983) have benefited greatly from the clearer generic diagnoses now available. Cluver and King

(1983) succeeded in producing an explicit phylogenetic hypothesis for Permian dicynodonts, grouping taxa according to proposed synapomorphies. and their cladograrn was subsequently extended to include Triassic forms (King, 1988). Their analysis was apparently carried out "by hand" rather than by a rigorous cornputer parsimony algorithm. with the significance of possible synapomorphies being evaluated more or less subjectiveIy, but nevertheless represents the clearest and most complete hypothesis of dicynodont relationships that has so Far been proposed. As more ngorous methods are brought to bar, and paaicularly as more data regarding the anatomy of particular dicynodonts become avaiiable. it should be possible to produce increasingly reliable cladograrns for Dicynodontia. Furthemore, discussions of the declining diversity of dicynodonts towards the end of the Permian (King, 1990a), the evolution of their unique feeding system (King et al., 1989; Cox, 1998), and similar topics are ail dependent to some extent on the existence of well defined species and genen, which are used in most studies as operational units of macroevolutionary change and biological diversity.

Increasing knowledge of the anatomy and taxonomy of individual dicynodont genera can be expected to shed light on the evolutionary and ecological history of the group as a whole.

The Late Permîan genus Diictodon. probably the most cornmon of ail dicynodonts, is particularly important ffom this viewpoint. It is an extremely widespread, abundant and temporally persistent fom, with a stratignphic range spanning five Karoo biozones and perhaps comesponding to as much as 10 million years of gological time (King, 1993b). The genus has a complicated taxonomic history, having been erected originally by Broom

(19 13) to accomodate a single specimen (which he assigned to D. galeops) in which the preparietal appeared to surround the pineal foramen, a highly unusual feature. One additional nominal species showing the same peculiarity, D. sesoma, was subsequently added to the genus (Watson, 1960). Although preparietal shape is now considered

taxonomically unreliable, as noted above, other features of D. gaieops (pünicuiariy lh<: lack of postcanine teeth, the large size of the ectopterygoid, the wide dentary tables, the

"notch" between the leading edge of the caninifonn process and the palatal nm, and the presence of a median flange of the mailla that separates the palatine from the prernaxilla) were sufficiently distinctive that a number of species originally described as

Dicynodon. Oudenodon and Emydorhynchus could be reassigned confidently to

Diictodon (Cluver and Hotton, 1977, 198 1; Maisch. 1995).

The reallocation of these nominal species to Diictodon is in nearly ail cases unquestioned, as the highly distinctive palatal notch is virtually unrnistakeable and apparently shared only with Robertia, which has postcanine teeth (Cox, 1998). and an early form from the Tapinoceplzalus Assemblage Zone of the Beaufort (S. Modesto, pers. cornm.). However, King (1993b) examined the characters used to distinguish the 20 species now assigned to Diictodon from one another, and concluded that al1 of them were unreliable, leading her to suggest that al1 Diictodon specimens could be accommodated within a single species. The primary critena used to evaluate the characters were similar to those employed by Keyser (1975): a "good" character was (a) easily separable into discrete character States, (b) relatively immune to preservational distortion, and (c) unrelated to ontogeny or sexual dirnorphism. No character unambiguously met al1 three requirernents. However. it must be noted that King (1993b) did not look for correlations arnong the characters, implicitly disrnissing the possibility that several individually doubtfbl characters might prove useful if they dl happened to CO-vary,and in many cases provided little information regarding the actual distribution of a given character arnong the 37 specimens she examined. Skull proportions were not quantitatively studied, and dthough some vanations were explicitiy reiated to ontogeny or sexuai dimorphism miiny others were simply left as unexplained "intraspecific variation", begging the question of whether their causes can be identified.

Thus, although King's (1993b)study of Diictodon represents a valuable step toward understanding this important genus, it left several questions unexplored. The idea of a monotypic Diictodon is certainly plausible in light of the rampant synonymization of species in other dicynodont genen, but it must be viewed with caution until the possibility that morphometric criteria or suites of correlated c haracten might serve to distinguish multiple species has been considered. The stratigraphic range of Diictodon is the longest of any amniote genus occumng in the Beaufort Group, and the majority of

Karoo dicynodonts are known only from one or two assemblage zones (Rubidge et al.,

1995); the suggestion that a single species persisted over so great an interval, without either extinction or cladogenesis, requires close scrutiny. Futhemore, given that wide anatomicai variations clearly exist within Diictodon, it is necessary to look for alternative explanations for them if they are deemed taxonomically irrelevant. If only a single species is present, it potentially provides an excellent oppominity to identify and examine sources of variation such as ontogeny, sexuai dimorphism, geographic differences among populations, and anagenetic change through time. As Diictodon is widespread, well- stuàied, and sufficientiy abundant to allow the use of statistical methods of analysis, it represents an ideal candidate for a study of this kind.

An additional consideration is that the cranial anatomy of Diictodon has never been comprehensively descnbed in a way that incorporates the observed range of variation in the genus. Revious descriptions (e.g.,Sollas and Sollas, 19 13, 19 16; Agnew, 1959) have without exception concentrated on a small set of specimens, which were assigned ar the time to various species of Dicynodon in any case. A redescnption of Diicfodon with due attention to intrageneric variation has the potential to aid in interpreting its highly unusual ecological and physiological charactenstics, including a peculiar feeding mechanism

(Cox, 1998), a tendency to dig distinctive spiral burrows (Smith, 1987) and a mode of bone growth that is unique arnong dicynodonts in lacking distinct growth rings sepuated by lacunae (Chinsamy and Rubidge, 1993). In the larger context of dicynodont evolution, an accurate knowledge of Diictodon will provide useful data for phylogenetic and palaeobiologicai studies of the group as a whole, while the patterns of variation seen within this extremely abundant genus may act as a standard of reference for foxms represented by fewer specimens.

A good example of the latter point is the question of sexual dimorphism, a long- standing one in dicynodonts. The variable occurrence of canine tusks in many genen, including Diictodon, has suggested to many workers (e.g., Broom, 1935; Toenen, 1953;

Olson, 1969; King, 1988, 1990b. 1993b) that tusked skulls might represent males and tuskless skulls fernales. Subtler manifestations of dimorphism, partly morphometric, have been detected even in genera such as Lystrosaums (Broorn, 1932; Thackeray et al., 1998) and Aiilucephalodon (Tollman et aL, 1980) in which nisks are always present. However, cornparisons with extant tusked animals and examination of fossils material led Barry

(1957) to argue that tusks were probably present in al1 adults of "Dicynodon" (now

Diictodon) grimbeeki, with the tuskless specimens representing juveniles, and Cluver

(197 1) doubted the evidence for dimorphism in Lystrosaunts. In general, the case for sexual dimorphism in fossil animals is nearly impossible to prove or refute conclusively, particularly in taxa such as dicynodonts that iack close lrvrng relauves, and it is of course possible that some dicynodont species were strongly dimorphic while others displayed little or no sexual dimorphism. However, an abundant and well-studied form such as

Diictodon, which inc ludes both tus ked and tus Hess indi viduals. may yield enough data to indicate whether sexual dimorphism is indeed the best explanation for the variable occurrence of canine tusks and certain other cranial features. If the evidence in Diicrudon points to dimorphism, then its existence in other dicynodont genen (particularly close relatives) will become more plausible; if another explanation is better supponed, then dirnorphism will become less plausible except in taxa for which the aitemative cm be ruled out.

The present study aims to provide a full redescnption of the cranid anatomy of

Diictodon, with particular emphasis on anatomicai variations among available specimens.

In addition to defining the range of morphologies within the genus, it is hoped to determine whether observed variations result from ontogeny, sexual dimorphism, species differences, or other identifiable factors; the question of possible species differences has an obvious bearing on King's (1993b) hypothesis that al1 Diicrodon specimens fail within the single species D. feliceps. Finaüy, data regarding the cranid anatomy of Diictodon wili be incorporated into a preliminary cornputer-assisted cladistic analysis of dicpodont phylogeny, in an effort to clarify the taxonomie position of Diictodon and major patterns of evolution within the entire group.

MATERIALS

A total of 49 Diictodon skulls, primarily from the collections of the South African

Museum, Bemarci Price Institute, and Amencan Museum of Xliturai History, werz examined in the course of this study (Appendix 1, Table Al). This materiai varied widely in completeness and quality of preservation, and in some cases postcranial elements were associated with the skull. One specimen (SAM-PK-K7028)actudly consists of three associated skeletons. Only three skulls (R 97.1, R 97.2, and R 97.3) were prepared rnainly by the author. It should be noted that a number of additional specimens, previously identified as Diictodon, were rejected as sources of data because the presence of diagnostic features such as the palatd notch and the entry of the maxilla into the choanal margin could not be established with certainty. These dubious cases included the type specimens of five traditionally recognized Diicrodon species: D.joubeni (SAM-PK-695),

D. pygrnaeus (SAM-PK-2664), D. ictidops (AMMI 55 IO), D. palustris (AMNH 55 12) and D.pseudojoiiberti (SAM-PK-775).

There is considerable overlap between the present sample of Diictodon material and that studied by King (1993b), although a few of her specimens were unavailable for study and others were excluded because of doubt regarding their generic identification. As in

King's (1993b) sample, the rnajority of the skulls considered here (at least 26 of the 49 specimens) were found in the Tropidustoma Assemblage Zone of the Beaufort Group

(see Fig. 1), only one of the five zones in which Diictodon ha been recorded. Although the CistecephalusAssemblage Zone was aiso well represented, only a few specimens from the other three zones were available. The Tropidostoma and Cistecephalus Zones account for more or less the middle part of the stratigraphie range of the genus, and the comparative lack of older and younger material unfortunately limits the potential for understanding the e volutionary his tory of Diictodon in general and detecting magenetic change in panicular.

The sample includes the holotype of D. galeops, the type species of Diictodon

(AMNH 5308). in addition to type specirnens of the traditionally recognized species D. psittacops (AMMI5534), D. testzidirostris (SAM-PK-2354) and D. sollasi (SAM-PK-

7430). Al1 of these specimens clearly fdl within Diictodon, though the validity of the individual species is uncertain.

It should be noted that the sample as a whole was examined for anaiornical data, and variations were taken into account as far as possible in desctibing the skull of Diictodon and in preparing the data matnx for the phylogenetic anaiysis. However, only a few well preserved specimens were considered in drawing the skull reconstruction, as detailed below (see Osteological Description). Apart from Diictodon specimens, the only fossil matenal used in this study was a skull of the dicynodont Pristerodon (BPI 3024), utilized as a source of phylogenetic data; al1 genera in the data matrix other ihan Diictodon and

Pristerodon were coded on the basis of descriptions and illustrations in the literature. Figure 1. Biostratigraphic zonation of the Beaufort Group according to the influentid scherne of Kitching (1977) and according to Rubidge et al. (1995). Note that the exact position of the Pemo-Triassic boundary is unclear. Period Zone (Kitching, 1977) Assemblage Zone (Rubidge et al., 1995)

Cynognathus Cynognathus 4m 1

- - - -

Lyst rosaurus Lystrosau nis

1 Daptocephalris Dicynodon

Cistecephalus Cistecephalris

Tropidostoma SYSTEMATIC PALAEONTOLOGY

Synapsida Osborn, 1903

Therapsida Broom, L905a

Anomodontia Owen, 1860

Dicynodontia Owen, 1859

Roàeniidae King. 19SÛ

Revised Diagnosis

Small dicynodonts characterized by the following apomorphies: maxillary flange separates palatine frorn prernaxilla. distinct notch lies between palatal rim and anterior edge of caniniform process, minimum intertemporal and interorbital widths approximately equal, and lateral dentary shelf weak and rounded (note that this last chancter has an ambiguous distribution). Easily recognizable on the basis of the palatal notch, as the angle between the palatal rim and the leading edge of the caniniform process appean roughly perpendicular in lateral view.

Genus Diictodon Broom 19 13

Type species: Diictodon galeops (Broom, 19 13; subjective junior synonym of D.feliceps)

Revised Diagnosis

Robertiid dicynodont with the following autapomorphies: interpterygoid vacuity more than twice as long as median vomenne septum, postorbitals (and the underlying Banges formed by the parietals) approach one another posterior to the pineal foramen in most specimens, medial edge of dentary table fomelevated blade (though this may be primitive for robertiids). Distinguished from Robertia by the presence of a media1 blade on the dentary, the close proximity of the postorbitals on the skull roof, and the absence of postcanine teeth; distinguished from Dicynodon by the presence of a palatal notch, the presence of a medial dentary blade, the weak and rounded lateral dentary shelf, the lack of an intertuberal ndge on the basioccipital, and the lack of palatine-premaxilla contact and pterygoid-maxilla contact.

Diictodon feliceps (Owen, 1876, p. 45)

Synonyrns

Rhachiocephalodon feliceps (Owen);Seeley , 1898, p. 108

?Dicynodonjouberti Broom, 1905b, p. 33 1

Dicynodon psitîucops Broom, 19 12, p. 869

?Emydorhynchus palustris Broom, 1913, p. 456

Diictodon feliceps B room. 19 13, p.

?Dicynodon ictidops Broom, 19 13, p. 466

Dicynodon ~estzidirostrisBroom and Haughton, 19 13, p. 36

?Dicynodon pygmaelis Broom and Haughton, 19 17, p. 123

Dicynodon sollasi Broom, 1921, p. 648

Dicynodon macrorhynchus Broom, 1921, p. 657

Dicynodon harcghronianw von Huene, 1931, p. 30

Dicynodon ncbidgei Broom, 1932, p. 189

?Sintocephulusjoubem' (Broom); van Hoepen, 1934, p. 93

?Pylaecephalus ictidops (Broom); van Hoepen, 1934, p. 93 Pylaecephalus teshtdirostris (Broom and Haughton); van Hoepen, 1934, p. 93

Pylaecephalus sollasi (Broom); van Hoepen, 1934, p. 93

îylaecephalus mbidgei (Broom); van Hoepen, 1934, p. 93

Diqnodon grimbeeki Broom, 1935, p. 7

Diqnodon nanirs Broom. 1936, p. 379

Dicynodon huenei Broili and Schroder, 1937, p. 130 (preoccupied)

Dicynodon broomi Broili and Schroder, 1937, p. 133

Dicynodon grossanhi Broili and Sc hroder. 1937, p. 16 1

Diqnodon anneae Broom, 1940, p. 18 1

Dicynodon broilii (Broili and Schroder); Boonstra, 1948, p. 57

?Dicynodon psertdojoubefli Boonstra, 1948, p. 60

Dicynodon vanderhorsti Toerien, 1953, p. 9 1

Dicynodon antjiesfontteinensis Toerien, 1953, p. 93

Oudenodon Ituenei (Broili and Schroder); Toerien, 1953, p. 97

Dicynodon whitsonae Toerien, 1954, p. 937

Diictodon sesoma Watson, 1960, p. 142

Dicynodon tienshanensis Sun, 1973, p. 56

Anomodon huenei (Broili and Schroder); Keyser, 1975, p. 74

Diictodon tienshanensis (Sun) Cluver and Hotton, 1977, p. 179

Diicrodon feliceps (Owen); Cluver and Hotton, 198 1, p. 125

?Diictodonjouberti (Broom);Cluver and Hotton, 1981, p. 127

Diictodon psinacops (Broom); Cluver and Hotton, 198 1, p. 129

?Diicrodon ictidops (Broom); Cluver and Hotton, 1981, p. 130 ?Diictodon palustris (Broom); Cluver and Hotton, 198 1. p. 130

Diictodon tesntdirosrris (Broom and Haughton); Cluver and Hotton, 198 1, p. 130

?Diictodon pygmaeus (Broom and Haughton); Cluver and Hotton, 198 1, p. 132

Diictodon sollasi (Broom); Cluver and Hotton, 1981, p. 132

Diictodon macrorhynchus (Broom); Cluver and Hotton, 198 1, p. 132

Diicrodon Iiauglitonianus (von Huene); Chver and Hotton, 198 1, p. 132

Diictodon ntbidgei (Broom); Cluver and Hotton, 198 1. p. 133

Diictodon naniis (Broom); Cluver and Hotton, 198 1, p. 133

Diictodon grirnbeeki (Broom);Cluver and Hotton, 198 1. p. 133

Diictodon huenei (Broili and Sc hroder); Cluver and Hotton, 198 1, p. 133

Diictodon broomi (Broili and Schroder); Cluver and Hotton, 1981, p. 134

Diictodon grossarthi (Broili and Schroder); Cluver and Hotton. 198 1. p. 134

Diictodon wliitsonne (Toerien, 1954); Cluver and Hotton, 1981. p. 134

?Diictodon psetidojouberti (Boonstra); Cluver and Hotton, 198 1, p. 134

Diicrodon vunderhorsti (Toenen); Cluver and Hotton, 198 1, p. 135

Diictodon antjiesfonteinensis (Toerien); Cluver and Hotton. 198 1, p. 135

Diictodon anneae (Broom, 1940); King, 1993b, p. 3 11

Holorype

AMMI 5308, a humerus and moderately well-presewed skull lacking the lower jaw, stapes and quacirates. bcality

"Slachter's Nek', Somerset East District, Eastern Cape Province, South Afnca.

Horizon

Upper Permian, probably Cistecephaltu Assemblage Zone.

Refe rred specimens

Numerous specimens, mainly isolated skulls, from the Karoo basin, South Afnca.

Specimens descnbed and figured in the literature include: AMNH 55 10 (holotype,

Diictodon ictidops), AMMI 55 12 (holotype, Diictodon pakzistris), AMNH 5534

(holotype, Diictodon psitîacops), B MNH 11 184. B MNH 47052 (holotype, Diictodon feliceps), BMNH 4708 1 (Diictodon nlbidgei), BPI 17 1, BPI 175 (holotype, Diicrodon vanderhorsti), BPI 179, BPI 228, BPI 294, BSP 1934 VIII 46 (holotype, Diicrodon huenei), BSP 1934 Vm 47a and b (holotype, Diictodon broomi), BSP 1934 Vm 48

(holotype, Diictodon grossarthi), IVPP V.3260 (holotype, Diictodon tienshanensis), RC

42 (holotype, Diicrodon anneae), SAM-PK-695 (holotype, Diictodon jouberti), SAM-

PK-774 (holotype, Diictudon pserrdojouberti), SAM-PK-2354 (holotype, Diictodon testzldirostris), SAM-PK-7420 (holotype, Diictodon sollasi), SAM-PK- 10086, SAM-PK-

K5 105, SAM-PK-K7730, SAM-PKX7673, SAM-PKX7730, TM 253 (holotype,

Diktodon grimbeekr?, TM 268 (holotype, Diictodon nanus), TM 307, TM 38 1, TM 390,

UMZC R3 14 (holotype, Diictodon sesoma), UT Von Huene 1931 Abb 25 (holotype,

Diictodon haughtonianus) Distribution

Karoo basin of South Africa; Turfan basin, Sinkiang Province, People's Republic of

China.

Stratigraphie Range

Upper Permian. In the Karoo, tiom the Tapinocephaius Assemblage Zone of the

Beaufort Group to the Diqnodon Assemblage Zone; in China, within the Upper Jijicao

Group.

Diagnosis

As for genus, as Diictodon is monotypic.

Discussion

The rationaie for recognizing only a single species of Diictodon is $ven below (see

Intrageneric Variation). King (1993b) referred to this species as D. galeops, the binomen introduced by Broom (19 13) to designate the type species of his new genus Diictodon.

However, Owen (1876) descnbed as Dicynodon feliceps a species that has since been reassigned to Diictodon (Cluver and Hotton, 198 l), and its specific name has priority over that of D. galeops. The type and only species of the genus Diictodon is therefore D. feliceps. Although most species previously assigned to Diictodon are clearly synonymous with this form, a few (indicated by question marks in the synonymy list) are founded on such inadequate type material that their inclusion in the pnus is questionable. 19

OSTEOLOGICAL DESCRIPTION

Reconstruction of Skull

Although 49 Diictodon specirnens were examined in the course of this study, the skull was restored primarily on the basis of a few particularly well preserved examples. The details of the palate and basicranial region corne almost entirely from SAM-PK-K7673, although aberrant depressions near the anterior edge of the palatines and on the ilanks of the caninifom processes were omitted. The remainder of the reconstruction Iqely represents a composite of, and, in many details, a compromise between, R 97.1 and

SAM-PKK7730. The general shape of the skull is taken mainly from the latter specimen, as are the lower part of the occiput, the lateral wall of the braincase (including the sphenethmoid complex and cultriform process), and most of the snout region, although R

97.1 provided information on the septomaxilla and nasal boss. R 97.1 was dso the basis for the upper part of the occiput, the stapes, the ventral rmus of the squamosal, and the majority of the skull roof, though the structure of the intertemporal bar was furnished by

SAM-PK-K7730. As the exact pattern of sutures on the central part of the skull roof is extremely variable in Diictodon, the decision to follow R 97.1 was largely arbitrary, and a few details (such as the smooth outlines of the posdrontds) were taken from SAM-PK-

K7730. USNM 21982 (a specimen examined too late for inclusion in the main sarnple) influenced the structure of the braincase, particularly the pila antotica region, and some basicranial sutures were taken from UT Von Huene 1922 r. 1-4. The quadrate is that of

SAM-pK-K7643.

Restoration of the zygomatic arch was unfominately somewhat arbitrary, as this region of the skull is exceptionally prone to distortion and breakage; only in a minonty of skulls are the left and nght arches approximately identical. A "reasonable" configuration, based on examination of several specimens, was atternpted, but SAM-PKX7730 was the single most influential skull.

The lower jaw was based mainly on R 97.1, though the dentary tables and the configuration of the elements surrounding the mandibular fenestm were drawn from

SAM-PKX7730. Details of the articular region were taken from SA\LPE;-K7643 and

USNM 22982, with some inevitable adjustment for compatibility with the reconstnicted quadrate.

The skulls incorporated in the reconstruction Vary widely in size, quality of preservation and suturai details, and include a combination of tusked and tuskless specimens. Therefore, it was decided arbitrarily to make the reconstructed skull a typical tusked individual, with the tusks being taken from USNM 72982 and the pineal boss

(which is much more common in tusked specimens - see Intragenenc Variation, below) from R 97.2. A near-modal skull length of 9.5 cm was chosen. thought to represent a typical adult. Some of the sutures depicted on the restored skull are usually indistinct in specimens of this size, but were retaîned in order to make the reconstruction as informative as possible. The degree of sphenethmoid ossification - which may also vary with size - is probably approximately comect, because it is taken from a specimen with a skull length of 9.4 cm.

A "typical" tuskless skull would be slightly smaller than the reconstruction depicts, and would lack the pineal boss as well as the tusks; other features would be unchanged. Snout and skuil roof

The snout of Diicrodon is massive. and consists of a relatively flat and prominent antenor plane flanked posterolaterally by more recessed areas. It is fomed antenorly by the premaxilla, dorsally by the nasals, laterally by the maxilla and the exposed parts of the prefrontal and lacrimal; posteriorly, it is closed off by the lacrimal and jugal, which forrn the broad floor of the orbit. The skull roof 1s broad antenorly, but narrows behind the oval pineal opening to a bar flanked by elongate temporal fenestrae. The back of this intertempord bar usually curves smoothly into the occipital surface. The skull roof is braced anteriorly by the snout and postenorly by the braincase, but is unsupported centrally and often depressed in this region in distorted specimens.

Prenzadla

The premaxillae of Diictodon are fused into a single median structure, as is the case in dl dicynodonts other than Eodicynodon (Cluver and King, 1983). However, most specimens retain a short partiai midline suture, extending up to 2 cm anteriorly from the point at which the premaxilla meets the nasals in the midline. The premaxilla rnakes an extensive anterior contribution to both the snout (Figs. 2,3,9, 10, 15) and the palatal surface (Figs. 4, 11). It forms the antenor part of the palatal rim, and extends upward on the lateral surface of the snout to fom the ventral and anterior margins of the extemd naris. Posteriorly, it meets the maxilla dong a suture extending from the posterovenual corner of the naris to a point on the palatal nm about hdfway between the anterior edge of the caniniform process and the tip of the snout. The maxilla tends to be the more prominent of the two bones, so that this contact takes the fom of a well-developed shelf or ledge.

The anterior tip of the snout is blunt, allowing the premaxilla to form a definite (but narrow) forward-facing plane that slopes posterodonally toward a contact with the paired nasals. This process foms the dorsal edge of each extemal naris, and its forked posterior end intrudes medially between the two nasals, thus contributing directly to the large nasal boss. The entire antenor plane of the premaxilla is. in fact, very prorninent, forming a single continuous surface with the boss itself and giving the front of the snout a robust, solid appearance. In Oudenodon this surface bears prominent median and lateral longitudinal ndges (Keyser, 1975), but in Diictodon the ndges are usually weakly developed or entirely absent.

The premaxilla Forms the bulk of the secondary palate (Figs. 4, 11), which is slightly concave but does not show the deep vaulting of certain other dicynodont genera, such as

Endothiodon (Cluver and King, 1983). Posteriorly the palatal rim is thin and sharp, but at the tip of the snout it becomes thicker and curves smoothly upward to meet the palatal surface. Along this curvature lie two paraIIel rounded ridges, referred to by Cox (1998) as the "antenor palatai ridges" (Figs. 4, 11: ANT R), with a shallow trough mnning between them. Posteriorly they contact a single median ridge ("posterior palatal ridge", Figs. 4,

11: POS R) that extends back to the posterior edge of the premaxilla. This arrangement of ndges is common in derived dicynodonts, although in Kingoria and Pe~anomodonthe antenor palatal ndges are lacking and in Oudenotion and Tropidostoma they appear to teminate well in front of the antenor end of the posterior palatal ridge (Cluver and King,

1983). Figure 2. Reconstnicted skull of Diictodonfeliceps in dorsai view, 1.5 X natural site for a typical tus ked speci men.

Figure 3. Recoostnicted skull of Diictodon feliceps in lateral view, 1.5 X natural size for a typical tusked specimen.

Figure 4. Reconstnicted skull of Diictodon feliceps in ventral view, 1.5 X natural size for a typical tusked specimen.

Figure 5. Reconstnicted skull of Diictodon feliceps in occipital view, 1.5 X natural size for a typicd tusked specimen. / PIN BOSS Figure 6. Reconstructed mandible of Diictodon feliceps in dorsal view. 1.5 X natural size for a typicai tusked specimen.

Figure 7. Reconstructed mandible of Diictodon feliceps in lateral view, 1.5 X natural size for a typical tusked specimen. MAN FEN Figure 8. Reconstnicted mandible of Diicmdon feliceps in ventral view, 1.5 X natural size for a typical tusked specimen.

Figure 9. Diictodon feliceps (R 97.1) Skull in dorsal view. 1.5 X naturd size.

Figure 10. Diicrodon feliceps (R 97.1) Skull (with mandible) in lateral view, 1.5 X natural size.

Figure 1 1. Diictudun feliceps (UT Von Huene 1922 r. 1-4). Skull in ventral view, 1.5 X natural size. Note ihat sutures in the braincase region are unusually clear in this specimen. ov VL' I cm TUB Figure 12. Diicrodon feliceps (UT Von Huene 1922 r. 14). Skull in occipi ta1 view , 1.5 X natural size. FOR In Diictodorz, the posterior palatal ridge becomes more prominent posteriorly, evidently reaching its maximum depth near its posterior end. On either side of the posterior ndge lies a shallow but well defined groove, terminating anteriorly against the back of the anterior palatal ndge. Further Iaterally a low, rounded eminence runs

longitudinally along the suture between the premaxilla and maxilla. Although Cluver and

Hotton (198 1) reterred to this suucture as an additionai paiatd ridge, it is muçh iess distinct than this description would suggest. The posterior margin of the premaxilla forms the antenor border of the choana, and sutures with the vomer rnedially. In contrast to

Diqnodon and Oudenodon (Cluver and Hotton. 198 1) it is excluded completely in

Diictodon from contact with the palatine by a broad flange of the maxilla that extends

inward to the lateral margin of the choana.

Toenen (1953) stated that the shape of the choana varies in Diictodon, with the anterior end either being bluntly rounded or extending forward as a nmow invagination

between the premaxilla and vomer medially and the maxilla and palatine laterally. The

latter condition is overwhelmingly more common, and it is possible that the antenor end

of the choana appears wide only when the bone around its margins has eroded away.

Although the premaxilla is edentulous, its three palatal ridges were presurnably

reflected in the topology of the keratinous beak and must have played a prominent role in

food mastication. Smdl foramina, believed to correlate at least roughly with the extent of

the beak in the living animal (Cox, 1998), cover the entire extemal exposure of the

prernaxilla. They also occur anteriorly on the mediai part of the palatal surface, thickly

concentrated on the paired anterior ndges but extending only a short distance back along the postenor ridge. More laterally and posteriorly, the palatal surface of the premaxilla is omarnented with small, obliquely aligned grooves.

S~diesusing sectioned specimens of Diictodon (Sollas and Sollas, 1913; Agnew,

1959) have show that the premaxilla foms a vertical median septum, which rises to contact the vomer, above the secondary palate.

Madla

This bone is exposed broadly on the lateral surface of the snout (Figs. 7,3,9, 10, 15). and sen& a process posteriorly to contribute to the suborbital bar, overlapping the jugal and squamosal. More antenorl y, the posterodorsai margin of the maxilla curves in parallel with the orbital margin but is excluded from it by small lateral exposures of the jugal and lacrimal. The dorsal edge of the rnaxilla meets the nasal and an anteriorly projecting flange of the lacrimal dong a roughly horizontal suture.

The maxilla foms the venual part of the posterior margin of the extemal naris. From the posteroventral corner of the naris, it slopes anteroventrally, as noted above, to form a shelflike contact with the less prominent premaxilla and ultimately descend to the palatal rim. The maxiliary part of the palatal rim is straight antenorly, forming a thin, bladelike edge that extends well below the level of the palatal surface, but posteriorly is intempted by the large and highly distinctive caninifom process. This structure appronimates a right-angled triangle in cross-section, with medial, anterolateral and posterior faces.

Whereas the posterornedial and posterolateral edges of the caniniform process are smoothly rounded, its anterior edge foms a narrow vertical blade that is aligned obliquely to the palatal rim and appears in latenl view to intersect it at approximately a right angle. However, because of its oblique orientation, the blade actually continues to the palatal surface rather than merging smoothly with the palatal nm, and a small notch

(Figs. 4, 11: PAL NO), opening anteromedidly, lies between rim and blade. The presence of this notch. and of a well-defined blade that meets the maxillary rim at a steep angle, is highly characteristic of both Diictodon and Ruberfia, and Cluver and King (1983) interpreted it as an apomorphic charxter irnpiying a ciose reiahmship beiwren these genera. Cox (1998) suggested that the notch rnight have acted in combination with the uptumed anterior edge of the dentary to cut through plant stems and roots.

In some specimens the caninifom process also supports a large tusk, which is the only tooth in the skull of Diictodon. The tusk is slightly offset medially, so that its base is overhung latenlly by the caniniform process. The tusk is typically broken near the base, but when neariy complete it extends beyond the ventral edge of the mandible. Well preserved examples generally show a weak postenor curvature, in addition to a number of minute vertical surficial gmoves.

Whether or not a tusk is present, the anterolaterai face of the caniniform process is swollen outward, perhaps accornrnodating a vestigial and unerupted tooth in apparently tuskless specimens. The swollen prominence follows the curvature of the posterior margin of the maxilla, and may extend as far dorsally as the ventral edge of the suborbital process, or even a short distance beyond it. The posterior face of the caniniform process is usually slightly concave, and extends medially to contact the jugal and epipterygoid bones, which together exclude the maxilla from the labial fossa.

Both Dicynodon and Kingoria have well developed caniniform processes, but in both these genera the caniniform process merges smoothly with the palatal rim rather than intersecting it at a sharp angle and forming a distinct notch (Cluver and Hotton, 1981);

Kingoria also has a postenor keel extending from the caniniform process, not seen in

Diictodon, while the palatal rim antenor to the caniniform process foms a rounded prominence rather than a sha.cutting blade (Cox, 1959). In Oudenodon the caniniform process is at best weakly developed, but bears a posterior keel similar to that of Kingoria.

Tusks are variably present in Kingoriu, but absent in al1 specimens of Oudenodon and present in al1 specimens of Diqnodon (Cluver and Hotton, 198 1).

The maxilla of Diictodon forms the lateral part of the palatal surface (Figs. 4, 1 l), suturing rnedially with the premaxilla. A more posteriorly positioned flange, onginating near the back of the caniniform process, extends medially to enter the lated choanal margm and separate the prernaxilla from the palatine, as noted above. It contacts both these bones dong short interdigitating sutures.

Like the premaxilla, the maxilla bears numerous small pits over much of its external surface, except for the suborbital process and other areas adjacent to the orbital or narial rim. On the palatal surface, foramina are sparse except on and around the caniniform process, where they occur denseiy. The posterior face of the caniniform process is also thickly covered. In many well preserved specirnens the maxilla forms a small boss or knob of variable shape and prominence slightly posterolateral to the base of the caniniform process. In some cases a depression in the surface of the secondary palate lies just posterior to this stnicnire. Septomaxilla

This bone is frequently rnissing or fragmented even in otherwise well preserved specimens. It is probable that, as suggested by King (1993b) it is both fragile and easily separable from the skull and thus is preserved only rarely. As in Oudenodon and

Kingorirr, but not Dicynodon (Cluver and Hotton 1981), it lies deep within the extemal naris (Figs. 2,3,9, IO), being strongiy recessed in cornparison ro the adjacent niiuriiia and nasal, and even the premaxilla. It resembles an elongated straplike sheet of bone twisted into a loose spiral, with its ventral end extending aimost medially. Postenorly and dorsally, it follows the curvature of the narial rim and eventually extends upward to abut against the ventral edge of the overhanging nasal bone. There are no clear indications of distinct foramina dong the septomaxilla-maxilla contact, as have been observed in various positions in other dicynodonts (Cox, 1959; Cluver, 1971: Keyser, 1975), but the entire ventral edge of the septomaxilla is separated from the maxilla by a narrow intervening groove.

Nasal

The paired nasals lie between the frontals and the premaxilla in the skull roof (Figs. 2,

3, 9, 10, 15) and descend laterally to contact the external nares and the prefrontal. lacrimal and maxillary bones. Medially, the two nasals clasp the posterodonaily tapering anterior plane of the premaxilla, which in combination with the nasals forms the large and prominent nasal boss (Figs. 2,3,9, 15: NAS BOSS), a common structure among dicynodonts (Cluver and King, 1983). It is an elevated, subcircular knob of bone, extending laterally to the posterodorsal corner of the external naris (and slightly overhanging the naris in most cases) and posteriorly to the contact between the nasals and the frontals. Antenorly it is continuous with the equally prominent antenor plane of the snout. The edges of the nasal boss are relatively steep, clearly delineating it from the test of the skull roof. Adjacent to the posterodorsal corner of the extemal naris its edge is slightly recessed. The entire surface of the nasal boss is pitted with scattered foramina. perhaps indicatmg a keratinous covering in Me. in a few specimens ine mediui piut of ihé boss is subdued, and the edges relatively prominent (Fig 15b), while in others there are weak but distinct median and lateral swellings. However, the boss in Diictodon never approaches the paired condition typical of dicynodont genen such as Oudenodon and

Rhachiocephalus (Keyser, 1975) and Aulacephalodon (Tollman et al., 1980).

Posterior and lateral to the nasal boss, the surface of the nasal is smooth and continuous with the rest of the skull roof. It forms a transverse but sharply interdigitating suture with the frontal, and smoother lateral contacts with the maxilla, prefrontal and laccrimal. Its sharp posterolatenl corner closely approaches the orbital margin, but is excluded from it by the frontal and prefrontal.

Prefrontal

This large bone foms the dorsal part of the antenor margin of the orbit, being exposed both on the extemal surface of the skull and intemally on the orbital wall (Figs. 2,3,9,

10, 15). Its extemai exposure is roughiy triangular, with its ventral base abutting the lacrimal and its posterodorsal apex lying in the orbital margin, slightly posterior to the level of the Frontonasal suture. The media1 face of the prefrontal thus contacts the nasal dong most of its length, but the frontal at its posterior extremity. The su- contacts with the lacrimal and frontal are carried without interruption onto the inner wall of the orbit, the frontal contact turning medially to allow the frontal to penetrate deeply into the orbital wall. The prefrontal foms part of a prominent, continuous ndge surrounding the dorsal and antenor orbital margins, but there are no distinct prefrontal bosses of the kind found in Oudenodon and Rhachiocephakis (Keyser,

1375j.

Lacrimal

The lacrimal lies ventral to the prefrontal and dorsal to the jugal in the orbital margin

(Figs. 2,3,9, 10, 13, 15). Between these two bones it forms the whole inner wdl of the

orbit, but its external exposure is primarily lirnited to a narrow strip running parailei to

the orbital margin and bounded anteriorly by the mailla. Above the dorsal edge of the

maxilla, however. the lacrimal sends a small rectangular flange forward between this

bone and the prefrontal, thus achieving a brief contact with the nasal. The lacnmal also

fomthe ventral part of the narrow elevated orbital rim, which disappears a short

distance above the ventral suture with the jugal.

The most conspicuous feature of the lacnmal is the large lacrimai forarnen (Figs. 2,3,

9, 10, 13, 15: L FOR), which lies near the dorsolateral corner of this bone's exposure in

the anterior wall of the orbit. It is dorsoventrally elongated and its lateral margin

sometimes bears a small notch that extends forward into the orbital rim and is visible in

lateral view. The foramen opens antenorly, and Sollas and Sollas (1913) reported that it

cornmunicates with the nasal cavity. Jugal

The jugal of Diictodon consists of a broad and thick anterior body, exposed within the floor of the orbit and on the postenor wall of the nasal capsule, and a long posterior process appressed to the media1 face of the squamosal (Figs. 2,3,9, 10, 13, 15). The anterior body forms the ventralmost part of the inner orbital wall, and is stmcturally continuous with the other bones of the circumorbitai series (Le. postfronrai, frontai, prefrontal and lacrimal). Anteriorly. it contacts the lacrimal along a smooth, approximately transverse suture, and laterally it is bounded by the maxilla. Although the maxilla overlaps the jugal almost completely in this region, the jugal may be laterally visible to a variable degree, typically in the anteroventral corner of the orbit but sometimes additionally along the dorsal edge of the suborbital bar.

Posteriorly, the medial part of the jugal curves sharply downward, leaving most of the orbit without a solid floor of bone. The jugal maintains its relationship with the maxilla in this region, suturing with the area around the root of the caninifom process latenlly, and is bounded medially by a dorsal flange of the palatine bone (Fig. 13). Its ventral edge meets the ectopterygoid dong a short jagged suture, and also forms the dorsolaterd corner of the labial fossa (Figs. 3, 10, 11, 13: L FS), which thus lies between the jugd, ectopterygoid and palatine.

The lateral edge of the jugal sen& a long process posteriorly, forming a single continuous suborbital and subtemporal bar in combination with the thin anterior rarnus of the squamosai. The squamosal laterally overlaps the dorsal part of the jugal in this area, and is itself overlapped anteriorly by the posterior process of the maxilla. Ventrally, however, the jugal extends below the squamosal and is therefore IatenlIy exposed Near its antenor end it is relatively broad, contributing most of the thickness of the suborbital bar, but it gradually tapers posteriorly. At the level of the postorbital bar it forms a dorsal process that lies against the media1 face of the ventral ramus of the postorbital (Fig. IO), evidently reinforcing the contact between this bone and the squamosal.

In Oudenodon (Keyser, 1975) and Kingoria (Cox. 1959) the jugal appean to display a greater lateral exposure in the region of the orbit dian is usud in Diicrodon, achieving a lateral contact with the ventral ramus of the postorbital and effectively excluding the maxilla and squamosal from the ventral orbital margin. However, the high vaiability of jugal exposure in Diictodon may be replicated to some extent in these genera.

Frontal

This oblong bone forms most of the skull roof immediately media1 to the orbit, including the dorsal orbital margin (Figs 2,3,9, 10). Anteriorly it meets the nasal dong a transverse but strongly interdigitating suture, as noted above. Just posterior to the lateral end of the frontonasal suture, the frontal contacts briefly the prefrontal, but more posteriorly the frontal itself forms the dorsal orbital margin. Like the prefrontal and lacrimal, it forrns a prominent orbital rim; more medially, the frontal seems to thicken slightiy, giving the dorsal wall of the orbit a pronounced ventrornediai slant.

Postenorly the frontal tapers to a narrow point, forming an extensive triangular flange which is evidently bounded mediaily by the preparietal and a broad medial flange of the parietal, and laterally by a much narrower laterai parietal flange and by the postfrontal.

On the inner wall of the orbit the frontal may be overlapped slightly by the postfiontal, which forms the postemdorsal corner of the orbital margin. This contact represents a continuation of the extemally exposed suture between these bones ont0 the inner orbital wall. Sectioned specirnens show that the ventral surface of the frontal contacts the sphenethmoid compIex in Diictodon (Sollas and Sollas, 19 13; Agnew, 1939) and seems to have roofed the olfactory canais.

In Oudenodon the structure and relationships of the frontal bone are strikingly similar to the condition in Dizctodon, but in Krngoria ana Diqnohn irs posterior contact wih the parietai is much less extensive (Cluver and Hotton, 198 1). Kingoria also appears to lack a postfrontal bone (Cox, 1959).

Preparietal

As noted by previous authors (Toerien, 1953; King, 1993) the structure of this unpaired mediai bone is extraordinarily variable in Diictodon and other dicynodont genera, and much of this variability probably has little taxonomie or biologicai meaning.

In al1 specimens examined in the course of rhis study the preparietal is roughly similar in its sutural outiines to its counterparts in Oiîdenodon, Khgona, and Dicynodon (Cluver and Hotton, 1981). It forms the antenor border of the pineal forarnen, and extends

forward between the parietals and fioontals (Figs. 2,9, 14). It is approximately rectangular, but the outline of the anterior tip is highly variable. Usually the antenor tip

forms a rounded point, but sometimes it tapers sharply or ends abruptly at a nearly

transverse edge. Regardless of the details, the tip is clasped between the frontals and interdigitates with hem extensively. Posteriorly, the preparietal may extend for some distance dong the lateral margins of the pineal foramen as well as forming its anterior

border. Sutures in this region are often somewhat indistinct. The genus Diictodon was erected originally by Broom (19 13) largely on the basis of what he interpreted as a highly unusual condition of the pineal foramen in the type specimen of D.feliceps, AMMI 5308: it appeared to be completely enclosed by the preparietal, rather than bounded posteriorly by the parietal as in the majority of dicynodonts. Furthemore, the preparietal showed a marked expansion at its anterior tip rather than coming to a point. No other available specimen of Diictodon shows a simlar condition of the preparietal, however, and among other descnbed species this bone is said to completely surround the pineai foramen only in D.sesoma (Watson, 1960).

Re-examination of AMNH 5308 with these observations in rnind showed that, aithough suture lines in the vicinity of the pineal foramen are not entirely clear. the preparietal has been rnisinterpreted and appears to enclose only the anterior half of the foramen. The skull roof in this area has been over-prepared, so that the exposed sutures actually represent a level somewhat beiow the original bone surface. Therefore, the anterior expansion of the preparietal may occur only at depth, as could presumably be confirmed by transverse sectioning of other Diictodon specimens; Agnew (1959) did not comment on this feature.

The pineal foramen itself (Figs. 2,9, 14: PIN FOR) is an elliptical opening in the skull roof, anteroposteriorly elongated to a variable extent. It may be sumounded by a "boss", or elevated ring of bone (Figs. 2, 14a: PIN BOSS), formed by the preparietal in conjunction with the parietals, but in rnany specimens (particularly those lacking tusks) the area surrounding the foramen is almost entirely smooth. In Oridenodon a boss is reported to be present only in larger specimens, though it is always present in the apparentiy related Rhachiocephalus (Keyser, 1975); in Kingona it is always lac king.

intertemporal bar (Figs. 2,9, 14). The structure of this part of the skull is highly variable within Diictodon, but the position of the posterior ramus itself is relatively constant. It is always appressed closely to the underlying parietd, and takes the shape of a thin slightly curved strip of bone, the dorsal edge being somewhat convex. Its orientation varies somewhat, but is always more or less donomedial. In some cases it coven the parietal part of the ridge almost completely, but in others the parietal extends metiiaiiy oeyond irs inner edge. Laterally, by contrat, the posterior ramus appears to extend further than the parietd, so that its edge overhangs slightly the temporai fenestra, as noted by Sollas and

Sollas (1916). Postenorly, the dorsal flange of the squamosal interposes itself between the postorbitai and parietal, so ihat the postorbital directly overlaps the squamosal and can be seen to contact it dong an approximately horizontal suture.

Parietal

This bone lies between the frontal and postparietal in Diictodon, forming the main part of the intertemporai bar (Figs. 2,9, 14). Anteriorly, it sends forward a broad medial flange that is clasped between the frontal and preparietal, and a thinner lateral flange bordered medially by the frontal and latenlly by the postfrontal and postorbital. Posterior to these flanges, the body of the parietal contributes extensively to the margin of the pineal forarnen, forming its posterior border and often (depending on how far posteriorly the preparietal extends dong either side of the forarnen) al1 or most of its lateral border.

Thus, when a pineai boss is present, it is paaly fonned by the parietal. The two parietals do not meet anterior to the foramen, but postenor to it they join one another dong a straight rnidline suture which extends back to the transverse interdigitating contact between the parietais and the unpaired median postparietal. The position of this contact, however, varies considerably among specimens, so that the postparietal sometimes lies primarily within the intertempord bar and sornetimes primarily on the occipital surface.

Latedly, the parietal forms a prominent longitudinal ridge in association with the postorbital, postparietal and squamosal. The parietal seerns to make the principal contribution, in the form of a heavy flange of bone that projects dorsomediaily above the surface of the skull roof (although its orientation is variable, and strongly subject to distortion). Anteriorly, as noted above, the posterior ramus of the postorbital is directly appressed to the outer surface of the parietai Range, but more posteriorly the squamosal interposes itself between the two bones. The flange cm be seen to extend posteriorly well beyond the contact between the postparietal and the body of the parietal, suggesting that the parietal body uuderlies the postparietal to a significant extent. In any case, lateral flanges of the postparietal are closely applied to the inner surface of the parietal flange in this region, and may extend fonvard a considerable distance. At its posterior extremity the parietal flange also achieves a bnef contact with the tabular in the few specimens in which the latter bone is c1earIy visible. It should be noted that the degree of development of the parietal flange is highly variable: in some cases it is small and covered almost entirely by the postorbitals and squamosals, while in others it extends medially beyond these bones and the two flanges may meet in the midline of the skull to form a thick median crest.

In Kingoria the parietals are typically appressed to one another, fomiing a median crest (Cox, 1959). In Oudenodon and Dicynodon distinct parietal flanges similar to those

of Diictodon exist; in the former genus, but not the latter, they are always separated widely to expose the intervening surface of the temporal bar. In Dicynodon and Kingok, the anterior flanges of the parietal appear somewhat reduced in cornparison to those of

Diictodon (Cluver and Hotton, 198 1).

Internally, the parietal of Diictodon forms an elongate vertical flange that extends postenorly beneath the postparietal. The posterior end of this flange forms the dorsal part of the laterai wail of the braincase, while the antenor part is in sunird contact with the expanded upper portion of the epipterygoid.

Squarnosal

This bone is triradiate, with dorsal, ventrd and anterior rami. Both the dorsal and ventral rami are short and broad, and the two together fom a continuous thin sheet of bone from which the anterior ramus projects Iaterdly. The ventral rarnus cm therefore be viewed as the part of this sheet that lies ventral to the anterior ramus, while the dorsal ramus is the part lying dorsal to it. The ventral ramus (Figs. 3,4, 5, 10. 1 1, 12) descends to contact the paroccipital process, which abuts against its medial face. Its lateral face

(which is tilted slightly mtenorly) supports the loosely attached quacirate and quadratojugal. As descnbed by King (198 1) in Dicynodon trigonocephalus, the ventral part of this surface is slightly recessed to accommodate the plate-like quadratojugal, while its anteroventral corner foms a deep pocket of unfinished bone for the quadrate.

The posterior tip of the quadrate process of the pterygoid lies between the anterior rim of this pocket and the medial face of the quadrate. Like its dorsal counterpart, the ventral ramus projects well posterior to the occipital surface, so that the occiput is bordered on either side by a prominent, narrow flange of bone (Figs. 5, 12). Above the paroccipital process, the ventral ramus foms the lateral margin of the subcircular posttemporal fenestra (Figs. 5, 12: ETF), which emerges anteriorly into the space below the temporal opening and probably transmitted the vena capitis dorsalis (Cluver, 1971; King, 1988).

Dorsal to the posttemporal fenestra the squarnosal contacts the supraoccipital. There are no grooves on the supraoccipital extending medially toward the postemporal fenestra, as is the case in Kingorïa (Cox, 19%). When well preserved, the fenesuz is usuüiiy slighiiy elongated dong a ventromedial-dorsolaterd mis, but its shape is frequently altered by erosion of bone from its edges or distortion of the skull.

As noted by Agnew (1959). the dorsal ramus of the squamosal (Figs. 2.3,5, 9, 10.

12) is overlapped medially by a brod but very thin tabular bone; however, this element is frequently lost or darnaged. Beneath the tabular the squarnosal forms a sutunl contact with the supraoccipital, and more dorsally it interposes itself between the postorbitai and parietal, thus contributing to the dorsal ndges that lie at the edges of the intertemporal bar and fonthe upper margin of the temporal fenestra.

The anterior ramus (Figs. 2, 3,4,9, 10, 11) foms the bulk of the zygomatic arch. Its alignment is strongly susceptible to distortion, but in the best preserved specimens it arises frorn the surface formed by the dorsal and ventral rami as a bony shelf projecting laterally and slightly dorsally, so that in postenor view a concave "saddle" lies between it and the dorsal ramus (Figs. 5, 12). As it extends forward, the anterior rarnus rotates into a more dorsovenual orientation, forming a deep vertical blade. However, the zygomatic arch is perhaps more prone to distortion that any other part of the skull, and its orientation inevitably diffea kom specimen to specimen and in most cases between the two sides of any given specimen. The antenor ramus defines the lateral edge of the temporal opening, and continues antenorly to overlie the long posterior process of the jugal. The descending ventral ramus of the postorbital overlaps the squamosal at the posteroventral corner of the orbit; antenor to this point, the squamosal tapea dorsoventrally and finally pinches out between the jugal and the posterior process of the maxilla partway dong the suborbital bar (Figs. 2,

9).

The structure of the squamosal is relatively constant among Permian dicynodonts

(Cluver and King, 1983), though variation in its proportions and orientation does occur.

As noted above, the squamosal of Kingoria and Oudenodon overlaps the anterior part of the jugal to a much lesser extent than is seen in Diictodon.

Quadrarojugal

As in other dicynodonts, the quadratojugal and quadrate of Diictodon form a single structural unit: they are tightly sutured to one another. but apparently are attached only

weakly to the rest of the skull. The quadratojugd forms the dorsolatenl part of this complex (Figs. 3, 10). It consists of a flat plate of bone, widest dondly and tapering

sornewhat unevenly to a nmow ventrai neck. Its ventral end is tightly sutured to the

dorsal face of the lateral quadrate condyle. The quadratojugal rests in a shallow flat

depression fomed by the ventral ramus of the squamosal, so that its lateral face is

continuous with that of the squamosal bone itself. Although the dorsal end of the

quadratojugal (which is siïghtly peaked) abuts solidly against the shelf that forms the

dorsal boundary of this depression. the connection between the quadratojugal and the

squamosal does not seem to be a particularly solid one: in several specimens the quadratojugal and quadrate have been cleanly separated from the skull, with little or no damage to the squarnosal. The sirnilar condition of Diqnodon trigonocephalus led King

(1981) to suggest that the quadratequadratojugal complex in this fonn was slightly mobile and held in place largely by soft tissue, but in Diictodon at least the mediai surface of the quadratojugal is tightly applied to the squamosal over a large area and therefore probably immobile. King's (1981: p. 262) statement that "it is obvious dhat fhe quadrate is as free to move post mortem as is an element such as the stapes" in both D. trigonocephalus and Oudenodon also seems inapplicable to Diictodon, in which loss of the stapes from the skull is much more common than clean detachment of the quadrate and quadratojugal.

Although the ventral edge of the quadratojugal contacts the lateral part of the quadrate. the two bones are partially sepanted by a narrow, vertically extended quadratojugal foramen. As in Kingorfa (Cox, 1959) and D. trigonocephalus (King, 198 l), the squamosal coven this opening posteriorly.

Postprr rie ta1

Although most dicpodont workers (e.g., Cluver, 1971 ; Keyser, 1975; Cluver and

Hotton, 1981; King, 1988) prefer to cal1 this bone the interparietal, it unquestionably represents a fused homologue of the paired postparietals typical of non-synapsid amniotes, and is therefore referred to as the postparietal in the present paper. It (Figs. 2,5,

9, 12, 14) usually forms the dorsal part of the occipital surface, though in some specimens it is definitely associated with the skull roof and lies behind the parietals at the posterior end of the intertemporal bar. In the former, more common case the transition between skull roof and occiput is marked by a smooth curvature, and the postparietal forms approximately transverse suturai contacts with the parietals dorsally and the supraoccipital ventrally. In some specimens the postparietal is entirely excluded from the intertemporal bar, and restricted to the occipital surface. In contrast to Ozidenodon, in which the sides of the postparietal are approximately parallel (Keyser, 1975), this bone widens ventrally in Diictodon. It is a relatively superficial eiement, being underiain postenorly by the supraoccipital (Agnew, 1959) and antenorly by the parietals.

When it enters the occipital surface, the postparietal is in lateral contact with the tabulan (Figs. 5, 12). More dorsally and anteriorly, it forms thin laterai flanges that are closely appressed to the medial faces of the parietal ndges but appear too insubstantial to provide much structural support. These flanges may extend forward to the antenor edge of the postparietal. At least briefl y, al1 four bones contributing to the lateral temporal ndges - postparietal, parietal, squarnosal and postorbital - directly overlap one another.

Scleral Ossicles

The scleral plates of Diictodon (Figs. 9, 10) are thin and fragile, and rarely presewed; in oniy one specimen (SAM-PK-KS189) does the sclerotic ring approach completeness.

The individuai plates were approximately square, and overlapped one another significantly. Because contacts between adjacent plates are often unclear, it is difficult to estimate the total number of sclentes present. Palate and paiatoquadrate

As in other dicpodonts, the secondary bony palate of Diictodon is formed mainly by the premaxilla, with contributions from the maxilla laterally and palatine posteriorly. The postenor edge of the secondary bony palate is slightly embayed by the choanae, which flank the narrow vomerine septum rhat acts as a skucturüi continuation af the mcdian posterior ridge of the premaxilla. The secondary bony palate is surrounded by a sharp rim, and its surface is relatively flat apart from the three palatal ridges (see Premaxilla, above) and low lateral erninences formed dong the premaxilla-mailla suture. Posterior to the secondary bony palate the deep anterior rami of the pterygoids enclose a wide depression, the interpterygoid fossa. whose vaulted dorsal roof is pierced by an opening that represents the me interpterygoid vacuity. It is continuous with the choanae, and presumabiy connected to the extemal nares via a straight channel.

Vomer

The vorners are applied closely to one another in the midline of the skull (Figs. 4, 1l), and are fused together in the vast majonty of specimens. The vomers cm therefore be ueated as a single median septum lying in the sagittal plane, and separating the two intemal nares. Antenorly, its ventral edge is continuous with the median postenor ndge of the premaxilla, and these two structures are tightly sutured to one another. The vorner extends only a short distance posterior to this contact, but data from sectioned specimens of Diictodm (Sollas and Sollas, 19 13, 1916; Agnew, 1959) and other dicynodonts

(Cluver, 197 1; Keyser, 1975) indicate that the vornerine septum extends anteriorly above the surface of the secondary bony paiate, classping the median plate of the prernaxilla and fonning dong its dorsal edge a groove to receive the cultriform process of the parabasisphenoid.

The posterior margin of the vomerine septum curves dorsally, eventually bifurcating into paired horizontai expansions that fom the anterornedial roof of the interpterygoid lossa. These honzontd plates are approximateiy trianguiar, with their üpices direçtd posteriorly. Laterally, they suture with the palatine anteriorly and the pterygoid posteriorly, probably overlapping both bones to some degree; their medial edges enclose the anterior part of the interpterygoid vacuity, an opening piercing the roof of the interpterygoid fossa. In Diictodon the interpterygoid vacuity is elongate and approximately teardrop-shaped, with its posterior end bluntly rounded and its anterior end gradually tapering to a point between the two horizontal vomerine plates. As in

Oudenodon (Keyser, 1975) and Kingoria (Cox, 1959), but in contrast to Dicynodon

(Cluver and Hotton, 198 l), the vomenne septum of Diicrodon is somewhat shorter than the interpterygoid vacuity.

Palatine

The palatine consists of a curved sheet of bone that contributes extensively to the wall of the interpterygoid fossa, and sends a small ledge of bone medially from its anteroventral corner to fonn the posterior part of the secondary palate. The mediai ledge

(Figs. 4, 11) sutures anteriorly with the maxilla, which extends to the margin of the choana and thus excludes the palatine from an anterior contact with the prernaxilla.

Posteriorly, the ledge borders the wide interpterygoid fossa and tapen rapidly to merge with the main body of the palatine. Its ventral surface is pitted with small formina and was presumably covered by keratinized tissue in life. Keyser (1975, p. 28) described its counterpart in Ozidenodon as "irregularly conugated".

Posterior and dorsal to the projecting ledge, the palatine takes the form of a thin, curved sheet applied to the medial face of the ectopterygoid and pterygoid bones, which extend well beyond it ventrally. Overlying the pterygoid (in palatal view), the palaune forms the anterolateral wall of the interpterygoid fossa, and rneets the horizontal plate of the vomer medially. Its relationship with this bone is cornplex: the palatine appears to continue dorsally, passing Lateral to the vomer while remaining media1 to the pterygoid.

Its contact with the vomer is thus short. consisting of the line along which the latenl edge of the horizontal vomenne plate intersects the media1 surface of the palatine. The palatine extends dorsally beyond both the vorner and the bar formed by the ectopterygoid and pterygoid, and becomes medially deflected so that it forms a transverse wall ai the back of the snout below the orbit (Figs. 3, 10, 13). It achieves a lateral contact with the jugal along an extensive vertical suture, and its ventrolateral corner contributes to the margin of the labial fossa.

The nature of the palatine contribution to the secondary bony palate is a useful taxonornic feature in dicynodonts generally (e.g., Toerien, 1953). In Dicynodon and

Oudenodon, the medial ledge of the palatine is larger than in Diictodon, and passes medial to the maxilla to contact the premaxilla antenorly, thus forming the entire laterai margin of the c hoana (Cluver and Hotton, 198 1). In Kingoria, the palatine approaches the premaxilla closely, but the medial ledge is narrow and blade-like (Cox, 1959), in contrast to the broad ledge of Dicynodon, Diictodon and Otidenodon. Ectopte rygoid

The ectopterygoid of Diictodon lies anterior to the ventral part of the pterygoid, so that the two bones combine to form a single continuous bar or strut that lies against the lateral face of the palatine and extends ventrally beyond it to form a narrow, prominent keel

(Figs. 3,4, 10, 11, 13). The ectopterygoid iorms the entire thicicness of this ridp anteriorly, to the exclusion of the pterygoid. It thus borders the palatine directly, and is separated frorn it in palatal view by the slit-Iike lateral palatal foramen (Figs. 1, 10: L

PAL FOR). More donally , a labial fossa (Figs. 3, 10, 11, 13: L FS) penetrates between the ectopterygoid, jugal and palatine from the posterolaterd side. but the development of this opening is highly variable. In some specimens it is deeply excavated and penetrates far into the snout capsule, while in others it is no more than a small indentation. The left and right labial fossae of a single skull may also differ noticeably in depth. Antenor to the labial fossa, the ectopterygoid is slightly expanded, and closely applied to the posterior faces of the maxilla and jugal.

In Kingda and Oiidenodon the ectopterygoid is large, as in Diictodon. but in

Dicynodon it is reduced and displaced laterally so that the pterygoid extends anteriorly dong its medial border to contact the maxilla (Cluver and Honon, 1981). The occurrence of the labial fossa is widely variable arnong dicynodonts (Cluver, 197 1). It is present in

Lystrosaurus, Daptocephalzis, Kannemeyena and Aidacephalodon, as well as Diictodon, and is normal1y absent in "endothiodontids" (i.e., small toothed dic ynodonts) and

Dicynodon, though it does occur in the aberrant D. trigonocephalus (King, 198 1). Pte rygoid

The pterygoids (Figs. 3,4. 10, 11) meet in the midline to form a united rectangular

"body", and send processes antenorly toward the palatine, ectopterygoid and vorner

(palatine ramus) and posterolaterally toward the quadrate and squamosd (quadrate rmus). The median body is an approximately square structure, and @es rise to a deep median crest or keel lying dong the iine of contact between the rwo pterygoids. Tiiz elongated footplate of the epipterygoid overlaps the donal edge of the lateral face of the pterygoid body (Figs. 3, IO), and also extends for sorne distance along both the quadrate and palatine rami, following the curvature of the pterygoid. This is in marked contrast to the condition in Oudenodon, in which the epipterygoid footplate is situated entirely on the quadrate rarnus (Keyser, 1975). Posteriorly, the pterygoid body meets the parabasisphenoid along an approximately transverse and strongly interdigitating suture.

At its posteroventnl corner, just posterior to the base of the quacirate ramus, the pterygoid body forms a small depression.

The median keel is fragile and frequently broken, but when complete is prominent and extends along the entire length of the pterygoid body. It has been suggested, in

Dicynodon trigonocephalits, as the probable origin of the postenor pterygoideus muscle

(King, 198 1). The palatine rami diverge slightly as they continue antenorly, and also

broaden dorsoventrdly so as to enclose between them a deep vaulted cavity, the

interpterygoid fossa (Figs. 4, 1 1). Dorsaiiy, they cuve inward to form a partial roof for

the interpterygoid fossa, in combination with the vomers and palatines. These bones

overlap the dorsal part of the medial surface of the palatine ramus, which continues

anteriorly to meet the ectopterygoid. However, the roof of the interpterygoid fossa is penetrated by an elongated opening that represents the true interpterygoid vacuity (Figs.

4, 11: INT VAC), homologous to the same structure in other tetrapods. The vacuity is bounded posteriorly by the pterygoids and antenorly by the vomen, the medial edges of these bones being continuous with one another.

The quadrate rarnus of the pterygoid is a mediolateraily cornpressed blade extending from the posterolateral corner of the pterygoid body toward the quacirate region. it deepens slightly toward its posterior end. which is tightly wedged between the media1 face of the quadrate and the adjacent part of the squamosal. Its contact with the quadrate was apparently not a strong one, as in at least one specimen (R 97.2) the quadrate has fallen away from the skull without damage to the quadrate mus. It was impossible to confirm the presence, reported by Sollas and Sollas (1913), of a suture between the quadrate rarnus and the rest of the pterygoid. They may have been rnisled by the tendency of the bone to crack in this fragile region.

In Diqnodon, the pterygoid extends further antenorly than in Diicrodon, passing medid to the ectopterygoid to contact the maxilla (Cluver and Hotton, 198 1). Oudenodon is intermediate between the two conditions, with the pterygoid extending medial to the ectopterygoid for a considerable distance, but terminating at the posterior edge of the lateral palatal foramen rather than passing laterd to this opening as in Dicynodon. In

Kingoria the pterygoid is tmncated by the large ectopterygoid. as in Diictodon. and the body of the pterygoid appears to be overlapped ventrally by a triangular antenor fliuige of the parabasisphenoid (Cox, 1959). Quadrute

As noted above (see Quadratojugal), this bone is intimately joined to the quadratojugal, but less fidyconnected to the remainder of the skull. Its primary structural feature is a pair of ventrally located condyles (Figs. 4, 5, 11, 12) which provide an articulating surface for the Iower jaw. The condyles are not smoothly rounded, but rather dope steeply toward the median ans of the quadrate so that the venuai surface oi the bone is deeply recessed between them. The lateral condyle is slightly wider than the medial, and extends slightly further ventnlly; its ventral surface is smoothly convex, whereas that of the medial condyle forms a shallow longitudinal groove to accommodate the medial Range of the articular. Thus, although the dicynodont quadrate is sometirnes described as "W-shaped" (cg., King, 1988, p. 132), it should be noted that its two concavities (one being the shdlow groove, the other the deep cleft between the two condyles) are very unequal in depth. The quadrate as a whole represents the most ventral part of the skull, apart from the tusk.

The quadratojugal, as descnbed above, descends to suture with the dorsal side of the lateral quadrate condyle (Figs. 3, 10). The media1 condyle bears a trianguiar dorsal flwge, which is digned obliquely to the quadratojugal (curving anteromedidly) and is separated from it by the vertical quadrate foramen. This flange lies primarily antenor to the quadratojugal, and occupies a concave pocket at the anteroventral corner of the squamosal. The blade-like posterior tamus of the pterygoid is clasped between the quadrate fiange and the antenor nm of its squamosal pocket. Posteroventrdly, the quadrate extends slightly beyond the squamosal so that the pamccipital process abuts directly against its medial face over a small area, but elsewhere the squarnosal interposes itself between the quadrate and the paroccipital process.

The structure and function of the quadrate-articularjoint in dicynodonts is relatively well understood (Crompton and Hotton, 1967; King, 198 1; Cox, 1998). The convex quadrate condyles interlocked with similar but more elongated structures on the articula., allowing considerable propaliny of the man& ble without the risk of dislocation.

Crornpton and Hotton (1967) argued that mastication involved a vertical "beak-bite", carried out with the jaw in its most anterior position. followed by a retractive stroke of the closed jaw to grind food between the dentary and the palate. Diictodon presumably conformed to this generai pattern, though the distinctive blade at the anterior edge of the caniniform process may have contributed to the process by slicing into the food as it was drawn backward (Cox, 1998).

Epipteqgo id

This bone is a well-developed strut in Diictodon, as in other dicynodonts (Cluver and

King, 1983), extending donoventrally between the pterygoid and the skull roof (Figs. 3,

10). Ventraily, it forms an elongated footplate that overlaps partially the lateral face of the pterygoid. The footplate is approximately uiangular, with the tubular shaft of the epipterygoid rising from its dorsally located apex. The footplate extends for a considerable distance both anteriorly and posteriorly, following the curvature of the underlying pterygoid. Agnew (1959) found that it contacted the basipterygoid process of the parabasisphenoid in his sectioned specimens. The shaft often rises dmost vertically from the footplate, though in many specimens it is inclined slightly fonvards. Its lower portion is narrow and rounded, but near the skull roof it expands anteropostenorly to form a wide blade. The dorsal edge of this blade contacts a descending intemal flange of the parietal, as in other dicynodonts. In

Oudenodon, the epipterygoid lies Bat against the parietal without fomüng a suture

(Keyser, 1975), but in Diictodon an interdigitating suture berween these eiemenrs is clearly present. Sutura1 contact also occurs in Adacephalodon. Pelanomodon,

Cistecephalus and Kannemeyeria, though in Pnsterodon it seems to be lacking (Keyser,

1975).

Braincase, occiput, and stapes

The braincase is heavily constmcted, and sutures between the various elements are generally indistinct, paticularly in older individuals. Structurally, the dominant features of the posterior part of the braincase are the massive paroccipital processes, firmly braced against the squamosal and quadrate on either side, the broad occipital plate, and a recessed lateral wall that extends up from the paroccipital process to meet a parietal flange descending from the skull roof. Formed by the prootic and supraoccipital, the lateral wdl is solidly ossified (particularly the prootic portion), but extends only a short distance antenorly from the occipital surface.

More anteriorly the braincase consists of the long, blade-like cultnform process and the variably ossified sphenethmoid complex. The sphenethmoid complex consists of two platelike structural components, one forming a median wall between the orbits and the other resting on the donal edge of the cultriforrn process. Although the occiput is fonned mainly by braincase elements, it is flanked by platelike postenor extensions of the squamosal, and the supraoccipital is partly overlapped by the postparietal dorsdly and the abulars Iaterally. The foramen magnum is tall, tapering gently toward its rounded dorsal end.

Parabasisphenoid

As in other dicynodonts (King. 1988) the parasphenoid and basisphenoid ossifications in Diictodon are fused indistinguishably. This braincase element, herein referred to as the parabasisphenoid, lies antenor to the basioccipital within the floor of the braincase, and sends a long, robust cultriform process fonvard toward the snout capsule (Figs. 3,4, 10,

12). Just posterior to its contact with the pterygoids, the parabasisphenoid is pierced by a pair of foramina (Fig. 1:ICC) that are thought to have served as passages for the interna1 carotid arteries (King, 1988). In many specimens these foramina remain distinct, though others show a single median opening of the kind reponed by Agnew (1959). An additional foramen lies on the suture between the parabasisphenoid and prootic on each side of the braincase, interpreted by Cluver (1971) as admitting the seventh cranial nerve.

Although the ventdateral tubera are formed primzily by the basioccipital (see below), their sloping anterior faces extend forward ont0 the parabasisphenoid (Figs. 4,

1I) and may reach the level of the sutural contact with the pterygoids before becorning indistiguishable from the general surface of the basicranium. As these prominent structures are located latenlly, the ventrai surface of the parabasisphenoid appears transveaely concave, though in some specirnens this effect is much more pronounced than in others. In one apparently aberrant specimen (R 97. 1). medial expansions of the tubera unite to fom a flat plate of bone that lies ventral to the surface of the parabasisphenoid and extends forward to meet the posterior end of the median pterygoid crest. Each internai camtid canal must have opened laterally in this individual, through a gap between this superficial plate and the underlying parabasisphenoid. Cox (1959) described what appears to be a similar condition in Kingoria, but no other available specimen of Diictodon shows this structure, the venuohteral tubera remaining separate anteriorly.

The cultriform process slopes anterodorsally, and is blade-like with a U- or V-shaped cross-section (Agnew, 1959). Anteriorly, sectioned specimens also show that the ventral edge of the cultriform process rests in a dorsal groove in the vomenne septum, while a similar dorsal groove in the cultnform process itself receives the presphenoid and orbitosphenoid elements (Sollas and Sollas, 1916; Agnew, 1959). The ventral edge of the cultriform process (well posterior to its meeting with the vomers) can be seen through the interpterygoid vacuity (Fig. 4). and its lateral face is exposed between the back of the nasal capsule and the level of the epipterygoid. It is slightly expanded dorsoventrally near its postenor end, in the region just ventral to the presphenoid and antenor to the epi p terygoid.

Basioccipitd

This median element forms the postenor part of the basicranium, lying just behind the parabasisphenoid, and also contributes the ventrd part of the occipital condyle (Figs. 4,5,

11, 12). Its ventral surface consists of a deep groove lying between two ventrally extended structures which previous authors have variously referred to as basioccipital- basisphenoid tubera (e-g., Cluver, 197 l), basisphenoid-basioccipital tubera (e.g., King,

198 l), basioccipital tuberosities (e-g., Keyser, 1975) and even hypapophyses (e-g., Sollas and Sollas, 19 13). They are associated primarily with the basioccipital, but their sloping anterior faces extend, as noted above, ont0 the lateral edges of the basisphenoid, while their posterior tips appear to be a contribution of the exoccipitals. Although perhaps irnprecise, the concise and neutral term "ventrolateral rubera" seems preferabie to

"basisphenoid-basioccipital-exoccipitaltubera", or the like, and will be used in this description.

The ventrolateral tubera (Figs. 4, 1 1: VL TUB) extend several millimetres below the surface of the basioccipital, and are angled slightly laterally so that they diverge ventraily from one another. Their primary structural roie is to enclose the fenestrae ovales (Figs. 4,

L 1: FEN OV), which open lateraily and slightly ventnlly, the medial faces of the tubera being strongly convex. In the few specimens in which the stapes remains in place. its expanded proximal end lies well within the sernicircular rirn of the fenestra ovalis. The ventral surface of each ventrolateral tuber is surfaced with unfinished bone. The anterior face dopes dorsally, eventually extending ont0 the basisphenoid before fading into the basicranial surface, while the postenor face is more abrupt and essentially in the occipital plane. Nthough sutures in the occipital region are extremely difficult to distinguish in most specimens, as is the case in dicynodonts generally (e.g., Sollas and Sollas, 1913;

Olson, 1944; Agnew, 1959; Cox, 1959; Ewer, 1961), evidence frorn apparently juvenile specimens with open sutures demonstrates that the posterolateral part of the ventrolaterai tuber is formed by the exoccipitai (Figs. 4, 11). A vertical suture between the basioccipital and exoccipital bones descends dong the posterior face of the tuber, and angles rnedially to separate the ventral basioccipital part of the occipital condyle from the more dorsal exoccipital part.

The contribution of the basioccipital to the occipital condyle consists of a rounded knob that remains partly distinct from the similar knob formed by each exoccipital. The occipital condyle thus consists of three swellings, arranged in an approximate triangle, with a slight centrai hollow between them (Figs. 5, Ilj. The exoccipitais are sepiirated by a vertical median suture, and the basioccipital-exoccipital suture seems to be aligned approximately transversely in this region. Despite the partial separation of the three contributing elements, it is clear that the condyle could have articulated with the atlas vertebra as a single structural unit. The entire surface of the condyle remains unfinished.

EKoccipital

Although the sutures separating the exoccipital from the supraoccipital and opisthotic are not distinguishable in any available specimen, it is cleûr that each exoccipital contributes a dorsolateral swelling to the occipital condyle (Figs. 5, 12). As described above, these swellings are intimately associated with one another and the more ventrai basioccipital contribution to the condyle, but remain separated by sutural contacts.

Although the postenor surface of the swelling is unfinished, its dorsal surface is smooth and forms a shelf projecting postenorly from the ventrai edge of the foramen magnum.

This opening (Figs. 5, 12: FOR MAG) is donovenually elongated in most specimens, and tapers gradually toward its rounded dorsd extremity.

The exoccipital forms the posterolateral end of the ventrolateral tuber (Figs. 4, 11); laterally it contacts the opisthotic, and dorsally it probably reaches the supraoccipital. As in other dicpodonts, the jugular forarnen of Diictodon (Figs. 5, 12: JUG FOR)is Iocated just lateral to the exoccipital part of the occipital condyle, and it may be entirely enclosed within the exoccipitai (Sollas and Sollas, 1913). However, in Oudenodon the jugular foramen lies between the exoccipital and the opisthotic (Keyser, 1975), and as this seems to be generally tme in dicynodonts (King, 1988) it may well apply to Diic~odun.As well as the jugular vein, the foramen probably pemtted passage of craniai nerves K. X ana

XI from the braincase (King, 1981).

Opisthoric

Although many authors (eg.,Agnew, 1959; Keyser, 1975) have descnbed the dicynodont opisthotic and prootic as a single periotic ossification, as is the case in mammais, rnost recent workers have regarded them as sepûrate elements (e.g., King,

1988). As a few Diictodon specimens show a clear suture between the opisthotic and the prootic, extending dong the ventral edge of the paroccipital process (Figs. 4, 1 1), the latter view will be adopted here.

The opisthotic foms the postenor part of the paroccipital process (Figs. 4,s. 11, 12), which connects the ventral part of the occipital plate to the squamosal and forrns the ventral border of the posttemporal fenestra (Figs. 5, 12, PIF). It appears to send a small

Range of bone forward across the floor of this opening. MediaIly, the opisthotic probably contributes to the rnargin of the jugular foramen and is indistinguishably fused to the exoccipitai, while its lateral end abuts solidy against the posterornedial face of the squamosal. There seems to be no sepûraùon of the paroccipital process into distinct

"mastoid" and "quacirate" components, as suggested by Agnew (1959); although the end of the paroccipital process is broad enough dorsoventrally to contact the medial face of the quadrate ventral to the squamosal, there is no separate descending process.

The posterior face of the opisthotic is venically convex, and at its lateral end it forms a small but distinct protuberance that can presumably be identified with the "tympanic process" described by Cox (1959) in Kingoria. However, the tympanic process of

Kingoria was much more prominent, with a length of some 7 mm in a skuil thar was approximately 8 cm long. Many Diictodon specimens are in the same generd size range. or slightly larger, and none appears to have a tympanic process extending more than 2 or

3 mm posterior to the surrounding bones.

Prootic

The prootic forms the antenor part of the paroccipital process, and also contributes significantly to the lateral wall of the braincase (Figs. 3,4, 10. IL). Unlike the opisthotic, which descends beyond the squarnosd to achieve a direct contact with the quadrate, the more anterior prootic seerns to be entirdy separated from the quadrate by the intervening squamosal. From this contact, the prootic portion of the paroccipital process extends rnedially and slightly anteriorly, passing below the posttemporal fenestra. The bone then bifurcates, sending a thick triangular flange dorsally and a less robust extension antenorly.

The triangular flange is high and nmow, and forms the anteroventral part of the ossified laterai braincase wall. It is bounded postemdorsaily by the much thinner and more recessed supraoccipital. The shelf-like contact between the supraoccipital and the prootic is evidently equivalent to the "deep cleft" described by Agnew (1959) as lying between these bones in D. grimbeeki, and the similar "venous groove" in Kingoria (Cox,

1959). In Oudenodon, however, the supraoccipital and prootic are indistinguishably fused, and no cleft exists (Keyser, 1975). Cluver (1971) suggested that an equivalent structure in Lystrosarms housed a blood vesse1 that terminated dorsally in a shallow notch lying between the supraoccipital and the dorsal tip of the prootic flange. A slight indentauon at ths point is common in Diicrodon, but he feature does not stem to b< universal and may simply reflect incomplete ossification.

Ventrally, the prootic extends to the border of the fenestra ovalis. It contacts the parabasisphenoid along a complex suture that extends dorsally frorn the fenestra ovalis but eventually tums anteriorly, allowing the anterior extension of the prootic to partially overlie the parabasisphenoid. The foramen for cranial nerve W (Figs. 3.4: N. W) lies partway along the vertical part of this suture, overlain by an extended shelf that is formed by the prootic posteriorly and the parabasisphenoid antenorly. In Lystrosaun

(1971) identified the underside of this shelf as the probable location of the vena capihs lateralis as it ran postet-iorly toward the gap between the stapes and the ventral edge of the paroccipital process.

At its antenor end the prootic abuts against the posterior edge of the elevated cultrïform process, though this contact is usually concealed behind the epipterygoid in intact specimens. Near its antenor tip the prootic foms a small triangular dorsal process, the pila antotica (Figs. 3, 10: PLANT). SupraoccipitaZ

This bone forms the centrai part of the occipital plate in Diictodon (Figs. 5, 12), and also sends forward interna1 flanges that contribute to the walls of the braincase (Figs. 3,

10). Its ventral contact with the exoccipitals is often unclear, but it appears to sunound at least the dorsal half of the foramen magnum; it also contributes to the dorsal and media1 rnargins of the posttemporai fenesua. In some specimens it forms a narrow prominent ndge slightly above the dorsal edge of this opening. The supraoccipital has a shon contact with the squamosal above the posttemporal fenestra, but more dorsally its lateral edges are overlapped by the tabulars. Its dorsal edge meets the postparietal along an approximately transverse suture; in some individuals this contact occurs at the top of the occiput, as the postparietal lies within the intertemporal bar, but in others it is more ventrally located. In either case the supraoccipital tends to be slightly indented just ventral to its suture with the postparietai.

The intemal flange of the supraoccipital can be seen in lateral view on either side of the skull. It forms the greater part of the ossified laterd wall of the braincase, lying between the prootic and the intemal flange of the parietal, but is thin and strongly recessed in cornparison to the adjoining dorsal fIange of the prootic (see above). It also appears less well ossified.

Sphenethmoid

The sphenethmoid complex of Diictodon and other dicynodonts consists of two distinct portions: a median plate lying between the orbits (Fig. 3) and a more posterior ossification situated on the dorsal edge of the cultriform process (Figs. 3, 10). The identity of the homologues of these elements in other groups (particularly mammals), and

hence their correct nomenclature, is controversial, and could only be conclusively

determined if their ontogenetic history were known; however, King (1988) followed

Cluver (197 1) in refemng to the postenor ossification as the presphenoid and the anterior

plate as the fused mesethmoid and orbitosphenoid, and this usage will be adopted here for

the sake of consistency.

The presphenoid is approximately triangular, with its apex directed dorsally and its

base lying almg the dorsal edge of the cultriform process. Sectioned specimens show that

the presphenoid actudly rests in a groove in the cultriform process, and that the groove

continues antenorly beyond the ossified presphenoid, suggesting that it was continued in

cartilage (Agnew , 1959). In lateral view the contact between presphenoid and

parasphenoid is indistinct, although the surface of the latter bone is generally much

smoother.

The anterior plate is highly variable in size, presumably because it ossified

progressively over an individual's growth (though more fully exposed exarnples would be

needed to establish that the degree of ossification is clearly correlated with skull size).

The dorsal edge of the median plate lies against the underside of the skull roof, and its

smoothly convex ventral edge contacts the anterior part of the culaiform process in some

cases but fails to reach it in others. In well ossified specimens the premaxilla, vomer,

cultnform process and anterior sphenethmoid plate are thus stacked vertically on top of

one another in a continuous senes, strongly bracing the snout.

As in Lystrosaums (Cluver, 1971), the anterior plate seems to consist of two fused

ossifications. The more posterior bone, the orbitosphenoid, bifurcates near its dorsal edge to form a pair of elongated dorsolaterally sloping "wings" that are fdyconnected to the underside of the frontal bone and may have enclosed the forebrain (Cluver, 197 1). Its posterior edge is deeply invaginated. The anterior bone, the mesethmoid, remains an undivided flat septum throughout its height. Although there is no clear suture between the two components of the anterior plate, a notch in the dorsal margin (Fig. 3) appears to separate the dorsally bi furcated orbitosphenoià component from the undi v idzd mesethmoid component. In at least one specimen (SAM-PK-K7730) a large forarnen lies slightly posteroventral to the ventral extremity of the notch.

Tabrilar

This thin, flat bone is either indistinct or missing in most available specimens. It lies in the dorsolateral part of the occipital surface (Figs. 5, 13, partially overlapping the squamosal and contacting the postpiuietal and supraoccipital. Its lateral edge is congruent with that of the squamosal, and it is widest at the level of the supraoccipital-postparietal contact, tapering both dorsally and ventrally from this centrai point. Dorsally it reaches the top of the occipital surface, achieving a brief contact with the parietal; ventrally it approaches the posttemporai fenestn, but does not normdly enter the margin of this opening (UT Von Huene 1922 r. 1-4 is a probable exception; see Fig. 12).

The tabular of Diictodon, although thin and often lost €rom the skull, is much larger than its counterpart in Oudenodon, which is small and restricted to the uppermost part of the occiput (Keyser, 1975). The tabular of Khgoria appears generally similar to that of

Diictodon in size, shape and position (Cox, 1959). Stop es

The stapes of Diictodon, like that of other dicynodonts, is easily detachable and often missing even from otherwise complete skulls. It is a slender dumbbell-shaped bone, extending from the fenestra ovalis to the media1 surface of the quadrate (Figs. 4,5). As in

Kingoria (Cox, 1959), and apparently the majority of other dicynodonts (King, 1988), there 1s no stapediai foramen. Tne meciiai end of the stapes is expündrd donaiiy, ancl to il lesser extent antenorly and posteriorly, to form a large footplate fitting over the fenestra ovalis, while the distal end is extended posteriorly so that its contact with the quaârate is extensive. The ventral surface of the stapes is thus quite flat, while the donal surface is slightly concave. The ventrai edge of the occipital plate is deeply emarginated just above this concavity, forming a wide channel that is thought to have transrnitted the vena capitis lateralis (Cluver, 197 1; King, 1988).

No extrastapes or extrastapedid facet is visible in Diictodon. This is pnerally the case in dicynodonts (King, 1988), although Cox (1959) found an extnstapedial facet in

Kingoria and Ewer (1961) described in Daptocephalus a small bony element that she interpreted as an ossified extrastapes.

Lower Jaw

The mandible of Diictodon is edentulous and heavily built. As in other dicynodonts

(King, 1988) the dentary is the largest bone in the lower jaw, and the only one to participate duectly in oral processing of food The angular forms a large reflected lamina, and the prominent condyles of the articular interlock with those of the quadrate to fom a joint that would have permitted considerable propaiiny while keeping transverse movement of the lower jaw to a minimum (King et al., 1989). While this general structure of the mandible is very conservative within Anomodonùa, a number of more specific characters such as the structure of the symphyseal region, the size and shape of the mandibular fenestra, and the presence or absence of a lateral shelf on the dentary are variable and thus diagnostic of individual genera.

Dentary

This bone (Figs. 6,7, 8, 10) is the most massive in the mandible, and its dorsal surface is the only part of the lower jaw to occlude with the palate. Although lacking teeth, its pitted surface indicates that it was covered in life by a keratinous beak, and the dorsal surface (Fig. 6) can be seen to form blade-like ridges that must have been effective in mastication. The dentary contributes approximately the anterior half of the lateral surface of the mandible.

Although a few specimens show a clear midline suture between the right and Ieft dentaries, they are fused together in the vast majonty of cases, as in Kingoria (Cox, 1959) but in contrast to Oudenodon (Keyser, 1975). The dorsal surface of the symphyseal region is upturned slightly at its anterior tip, and more posteriorly foms a deep median groove, matching the median ridge on the palatal surface, that is flanked by two broad

"dentary tables" (Fig. 6: DEN TAB). These approximately horizontal surfaces are msversely concave, each being bounded laterally and medially by sharp ndges. The

medial ridge is the higher and more distinct of the two. The dorsal surface posterior to the dentary tables forms a sharp ridge, and lacks the deep groove or "dentary sulcus" that occurs in Dicynodon and Oudenodon (Cluver and Hotton, 1981). As in the majority of dicynodonts, there is no coronoid process. It is noteworthy that this combination of features dso occurs in Robertia (King and Rubidg. 1993), despite the presence of small dentary teeth in this genus. Kingorïa, by contrast, lacks dentary tables, and the margins of the dentary are rounded rather than forming distinct cutting edges (Cox, 1959). In

Oudenodon, the dentq tables are narrower than in Diictodon (Cluver and Hotton, 1977).

The dentary overlaps the angular, surangular and splenial (Figs. 7, 10). Lts postenor margin is invaginated deeply by the anterior end of the low and elongated mandibular fenestra (Figs. 7,8, 10: MAN FEN), an opening lying between the dentary, angular and surangular. Dorsal to the fenestra the dentary is closely applied to the lateral surface of the surangular; ventral to the fenestra, it broadly overlaps the angular, although a thin sliver of the angular is exposed ventral to the dentary in laterd view.

A depression on the lateral surface of the dentary extends anteriorly from the mandibular fenestra. This depression is bounded dorsally by a prominent and smoothly rounded ledge of bone that continues posteriorly for a short distance along the upper edge of the fenestra. It appean to represent a functional equivalent of the sharper and more prominent lateral dentary shelf that occurs in certain other dicynodonts such as Kingoria,

Oudenodon and Dicynodon (Cluver and Hotton, 198 l), and is believed to have provided a point of insertion for the Iateral extemal adductor muscle.

Antenorly the ventral margin of the dentary can be seen to curve sornewhat dorsally as it approaches the symphysis, slightly revealing the lateral face of the splenial in many specimens. The ventral margin sweeps abruptly upward at the anterior tip of the jaw, resulting in an extensive, approximately biangular exposure of the splenials in anterior view. Splen ial

This thin, flat bone foms the anteroventnl part of the medial surface of the jaw ramus

(Figs. 6,7,8), entenng the symphysis and extending postenorly to approximately the

level of the antenor end of the mandibular fenestra. The two splenials contact one another

in the jaw symphysis, and are fused together in many specimens.

The outer surface of the splenial is visible only near the antenor end of the mandible,

where the ventral margin of the dentq curves upward (see above) to expose it. In media1

view, the splenial overlaps approximately the ventral half of the dentary. Near the

symphysis this contact is direct and the splenial is closely appressed to the medial face of

the dentary, although at the symphysis a small unossified gap occupies the suture

between the two bones. A short distance below it, just above the ventrd edge of the

splenial, lies a deep elliptical depression, with the longer axis onented donoventnlly. It

is vertically divided by a weak median ridge in most specimens.

More posteriorly, the thin anterior blade of the angular intervenes between the dentary

and the splenial, which also foms a complex interlocking contact with the prearticular:

the splenial bifurcates, with one flange lying lateral to the prearticular and the other

medial to it. Generally, in fact, there is considerable overlap in this area of the mandible,

with the dentary, angular, splenid and prearticular dl forming successive overlying

layers over a bnef distance. Angular

The angular (Figs. 6,7,8, 10) consists of a donoventrally expanded posterior body, from which arises a thin "reflected lamina" of bone, and a narrow anterior blade. The anterodorsal corner of the body forms part of the margin of the mandibular fenestra, while its long posterodorsal edge contacts the surangular. Anterioriy, this contact is an overlapping one, the angular overlying the sunngular, but more postenorly the surangular becomes more laterdly prominent and forms an overhanging shelf above the angular.

Anterodorsally, the lateral face of the angular body gives rise to a thin sheet of bone, the reflected lamina (Figs. 6,7.8, IO: REF LAM), which extends postenorly and ventnlly to lie well extemal to the surface of the angular proper. The lamina extends back nearly to the postenor edge of the anguhr body, and projects ventrally well below the remainder of the jaw ramus. Its ventral portion is sernicircular, thin and fragile, and medially deflected. The surface of the lamina bears three rounded prominences, separated by shailow grooves and radiating posterioriy, ventral1y and posteroventrall y from a point just behind the antenor border of the lamina. The reflected lamina is apparently univenal not only in dicynodonts but in non-mammalian therapsids generdly (Kemp, 1982)' and probably held a pharyngeal diverticulurn associated with the hearing system (Allin,

1975).

The anterior blade of the angular is clasped between the dentary laterally and the spleniai medially, though the dentary covers its surface only partialIy. Posterior to the splenial, the prearticular overlaps the angular's medial face apart fiom a thin strip of ventral exposure. Su rangular

The surangular is a postenorly expanded rod of bone forming the posterodorsal part of the mandible (Figs. 6,7,8, 10). It slopes slightly anterodonally, thus meeting the upper part of the dentary at a weakiy obtuse angle. Although this contact represents the lower jaw's highest point, and the dentary overlaps the sunngular stightiy, thete is no distinct coronoid process.

Posterior to its contact with the dentary, the surangular may enter the dorsal margin of the mandibular fenestra (Fig. IO), though in rnany cases the dentary extends back slightly beyond the posterior edge of the fenestra (Fig. 7). Posterior to the mandibular fenestra, the surangular is briefly overlapped by the dorsal tip of the angular body, but its dorsal surface soon broadens latenlly to form a wide shelf overhanging the postenor part of the angular. Posteriorly the surangular also extends ventrally, forming a wide plate that faces posteriorly and is appressed to the antenor face of the articular. These two bones combine to form a weak ventral projection that King (1988) referred to as a retroarticular process

(contra Watson, 1948). The mutually continuous dorsal surfaces of the surangular and articular, separated by an anteromedially angled suture, form a shallow concave depression lying just anterior to the articular condyles. With the jaw in its most retracted position, this fossa ("articular recess") would have received the ventral condyles of the quadrate (Cox, 1998). On the ventral surface of the jaw the sutures between the preariicular, surangular and articular are indistinct, and these bones may have been partially hsed. Prearticular

This is the only bone in the mandible of Diictodon that is not exposed at al1 in lateral view. It is a broad, strap-like bone lying pnmarily against the media1 face of the anguiar, but extending anteriorly to interlock with the splenial (as described above) and postenorly to contact and possibly fuse with the articular (Figs. 6, 8). In medial view it

Fomthe entire lower border of the rnandibular ienestra.

Articrilar

As descnbed above, the articular (Figs. 6,7,8,10) combines with the surangular to form a retroarticular process and shallow dorsal articular recess. Its main articular surface is donally concave, with elevated lateral and medial margins, and fits over the laterd condyle of the quadrate when the lower jaw is in situ. The media1 edge, fitting into the space between the quadrate condyles, is the more elevated of the two. Posteriorly the articulating surface descends steeply, forrning the postenor face of the retroarticuIar process.

Further stability in the jaw articulation was provided by a medial flange extending from the body of the articular, and angled slightly donomedially (Figs. 6,8). Its medial edge fits the longitudinal groove on the mediai quadrate condyle, allowing anteroposterior sliding but little lateral displacement. The homologue of this median artïcular condyle in less derived forms was not involved in the jaw articulation. INTRAGENERIC VAFUATION

As discussed above (see Introduction), King (1993b)studied a large sarnple of

Diictodon skulls and concluded that al1 important variations among them could be explained as results of ontogeny, sexual dimorphisrn, and sirnilar intraspecific factors, rather than as differences between Diictodon species. Though Diicrodon müy weil bc: monotypic, further documentation of variations within the genus is necessary in order to be sure that they are taxonornically insignificant. Furthemore, even intraspecific variations may be palaeobiologically important if they can be shown to be characteristic of groups such as sexes, age classes or distinct populations.

It therefore seems worthwhile to descnbe significant rnorphometric and qualitative variations in the cranial anatomy of Diictodon, and to atternpt to detemine whether these variations relate to species differences, ontogenetic or anagenetic patterns, sexual dimorphism, or other Factors. The problem is viewed as one of assigning observed differences among specimens to these various causes, in an effort to address both the taxonomie question of whether multiple species of Diictodon exist and the paiaeobiological question of what sex and age differences rnight have existed within

Diictodon populations.

Methods of analysis - qualitative skuii characters

Most points of potentially important variation in the skull of Diictodon were discussed by King (1993b), who considered none to be diable as a basis for the recognition of multiple species within the genus. She demonstrated clearly that nearly al1 characters used by previous authors were vaguely defined, prone to distortion, or redundant with one another, and therefore unhelpful. A few cranial features, however, seem to exhibit variation among well defined character states, and were either not discussed by King

(1993b) or ciismissed by her only because there was evidence (usually a suspected correlation with age or sex) that they varied intraspecifically rather than interspecifically.

Ten of these t'eatures ment tùrther consideration, and are aiscussed beiow:

1. Presence of canine tusks (Fig. 13)

The fact that the caniniform processes of some Diictodon specimens give rise to prominently developed canine tusks, while other specirnens are entirely edentulous. is perhaps the most obvious point of variation in this genus, and in most cases the canine tusk is either fully formed or entirely absent. However, one specimen in the present sarnple (SAM-PK-10078) may have been undergoing tusk replacement at the time of death; as noted by Cluver (1970). both caniniform processes bear deep grooves on their medial faces, with the tip of a small canine visible at the base of each groove. It seems possible that these small teeth were in the process of replacing a shed pair of tusks that formerly occupied the grooves. Two other specimens (SAM-PK-K7730and SAM-PK-

K6921) also have media1 grooves on their caniniform processes, but these are more weakly developed and lack any indication that they ever held tusks. Both caniniform processes of a third specimen (SAM-PK-10377) are somewhat damaged, and on the right side (though not the left) this breakage has partially exposed a tiny tooth that would have been entirely concealed had the caniniform process remained intact. Figure 13. Anatornical variations in the suborbital region of Diicfodonfeliceps. (A), posterolaterai view of R 97.2, with a canine tusk and deep labial fossa; (B), posterolateral view of SAM-PKX7730, lacking the tusk and having only a shallow labial fossa. Both specimens 2 X naturd size.

Figure 14. Anatornical variations in the postenor skull roof of Diictodon feliceps. (A), dorsal view of R 97.2. showing the prominent pineal boss and closely juxtaposed parietal flanges; (B), donal view of SAM-PKX7730, in which the flanges are widely separated and the pineai boss is absent. Both specimens 2 X natural size. POP Figure 15. Variation in the fom of the nasal boss in Diictodon feliceps. (A), donal view of R 97.1. in which the surface of the boss is relatively even; (B), dorsal view of SAM-

PK-K7795,in which the boss is medially subdued but forms lateral prominences that resemble the small paired bosses of genera such as Oudenodon. Both specimens 3 X natural size.

2. Pineal boss (Fig. 14)

In many specimens the rim of the pineal foramen, normally formed by the preparietal and parietal bones, is swollen into a bony boss. This boss, when present, varies in degree of development from a thin elevated collar around the foramen to a large mound of bone that rnay laterally contact the Ranges formed by the parietals. Despite this variability, the basic distinction between presence and absence of the ~ossis reiativeiy çiear.

3. Stnictiîre of the nasal boss (Fig. 15)

Al1 Diictodon specimens have an elevated boss situated posterornediai to the extemal nares; it is fomed by the nasals and the posterior part of the premaxiila, and is antenorly continuous with a media1 elevation of the snout that extends between the nares and down to the maxillary rim. The boss is smooth and undifferentiated in many specimens, but may also be separated into two laterai swellings by a more or less distinct medial depression. However, a medial ridge or crest may also extend along the boss (and often further anteriorly along the snout), whether or not latenl elevations can be distinguished, and the various combinations of these features gnde continuously into one another. A rough distinction cm be drawn between specimens in which strong lateral swellings are present and those in which they are not, but these conditions admittedly represent the endpoints of a continuum rather than fully discrete character states.

4. Presence ojbosses posterior to the caniniform process

Tubercles have been recorded in Diictodon both at the anterior edge of the palatine bone and at the posterornedial corner of the caniniform process (King, 1993b). However, no specimen in the present sample has more than one clearly evident tubercle in this region, suggesting the presence of a single structure (whose position is widely variable) rather than two independent ones. This posicanine boss, as it shall be termed here, seerns to be present in some specimens and absent in others, though in some cases its apparent absence may result from poor preservation or minor over-preparation. When present it usually occurs on or near the posterornedial t'ank oi the caniniform process, 'out ir may iie more postenorly, near the anterior margin of the palatine.

5. Degree tu wlzicli postorbitals overlap parietal flanges (Fi g. 14)

An upraised flange of bone extends longitudinally dong the lateral edge of each parietal, sloping dorsomedially throughout its length. A long postenor process of the postorbital overlies the Iatenl face of this parietal flange. However, the extent of overlap may be very incomplete, leaving the media1 part of the flange exposed. Although diagenetic effects might displace the postorbitals with respect to the underlying parietals

(King, 1993b), the fact that the postorbitals are symrnetrically positioned in most cases suggests that this is a rare phenornenon, as it would be unusual for both postorbitals to be equally displaced Thus, the fact that the postorbitals extend nearly to the media1 edges of the parietai flanges in sorne specimens, but leave a large portion of the Banges uncovered in othen, is probably not a preservational effect.

6. Proximity of parietaiflanges (Fig. 14)

The pmetai flanges mentioned above are highly variable in the extent to which they approach one another behind the pineal foramen: in some specimens the flanges are 101 widely separated, but in others they actually meet in the midline. The proxirnity of the parietal Ranges to one another represents a potentially significant point of variation.

7. Presence of the labial fossa (Fig. 13)

King (1993b: p. 3 19) noted the presence of "an ovai depression at the junction of the ectopterygoid, jugal and palatine" in many oi her specimens. Tnis structure is çirÿriy identical to the labial fossa of many other dicynodonts, including Lystrosatrnu,

Kannerneyeria, Aulacephalodon and Dicynodon leoniceps (Ewer, 196 1). In the presen t siunple it is evident in dl skulls that are adequately prepared in the relevant area of the skull, and it seems to be a consistent feature of Diictodon. Its size, however, is highly variable, ranging from a tiny, blind depression to a deeply excavated cavity.

8. Position of the postparietal

In some cases the postparietal is clearly associated with the upper part of the occipital surface, despite extending far enough antenorly to appress itself to the inner face of the parietai Range on either side. In other specimens, however, the postparietal is positioned more anteriorly and forms the posterior part of the intertemporal bar, so that its orientation is largely horizontal nther than vertical. In many cases i@ position is clearly intermediate, centred on the smooth cuve that links the occiput to the intertemporal bar.

Although this feanire may be somewhat susceptible to distortional effects (which might change the structural relationship of the intertemporal bar to the occipital surface) it seems to Vary even among relatively pristine specimens. 9. Fusion of the splenials

A vertical suture can often be seen between the two splenials at the mandibular symphysis, but in other specimens the spleniais appear to have indistinguishably fused into a single element. When present, the suture is usually quite fine, and is rnost clearly visible on the outer face of the symphysis where ventral emargination of the dentaries exposes the underiying spleniais rnediaiiy. On fie internai face of the symphysis the suture tends to be obscured by a deep depression in the splenial.

1O. Posttemporal ridge

In many specimens its margins of the posttempord fenestra are smooth and subdued, but in othen its dorsal border is thickened to form a narrow but prominent ridge. The distinction between presence and absence of this posttemporal ndge is quite clear.

These ten characters seem sufficiently free of distortion, and separable into distinct character states (Table la), to be suitable for statisticai analysis. Characters 1,2,9 and 10

(respectively presence of tusks, presence of pineal boss, fusion of spleniais and presence cf posttemporal ridge) are considered to meet these cntena fully, though in many cases the symphyseal area was too wom to determine whether the splenials were fused or separate. Character 8 (position of postparietal) may be ss~ewhatsusceptible to distortion, while characters 3 (structure of nasal boss), 4 (presence of postcanine boss), 5

(degree to which postorbitals overlap parietals), 6 (separation of parietal flanges) and 7

(depth of labial fossa) proved difficult to code precisely; a few intermediate specimens, for which assignment to one character state or the other becarne alrnost arbitrary, were Table 1. List of qualitative variations in the cranial osteology of Diictodon (A) and skull measurements used in the morphometric analysis (B). Figure 16 shows the morphometric measurements graphically.

A. Qualitative Characters 1 Canine tusks: present (P) vs. absent (A) 2 Boss surrounding pineal foramen ("pineal boss"): present (P)vs. absent (Aj 3 Nasal boss: laterally subdued (S) vs. laterally elevated (E) 4 Boss behind caniniform process ("postcanine boss"): present (P) vs. absent (A) 5 Coverage of parietal flanges by postorbitals: partial (P) vs. nearly complete (NC) 6 Parietal flanges: widely separated (S) vs. close together (C) 7 Labial fossa: shallow (S) vs. deeply excavated @) 8 Postparietd: associated with intertemporal bar (IT) vs. associated with occipital surface (O) vs. intermediate (1) 9 Splenids: fused at symphysis (F) vs. separated by a vertical suture line (S) 10 Ridge above posttemporal fenestra ("posttemporal ridge"): present (P) vs. absent

B. iMorphometric measurements I Skull length (tip of snout to posterior edge of basioccipital condyle) 2 Snout length (antenor edge of orbit to tip of snout: slightly oblique) 3 Minimum distance between extema1 nares 4 Minimum distance between orbits 5 Minimum width of intertemporal bar 6 Diarneter of extemal naris 7 Distance between palatal rim and ventral edge of extemd naris ("narial devation") observed in every case. The problem was particuiarly acute for character 4, which was left uncoded in a large number of specimens.

Once character States had been assigned for as many specimens as possible, the possibility of correlations among the various characters was explored by applying a two- dimensional chi-square test of inde pendence (Pearson, 1922) to every possible pair of characters, using a = 0.05 as the ievei oi significance. Wniie it might have ken more mathematically satisfactory to consider al1 ten characters simultaneously, by constnicting a ten-dimensional contingency table and evaluating it using a log-linear modelling approach (Agresti, 1990). this procedure would have the undesirable effect of spreading a relatively minuscule number of data (49 x 10 = 490 entries. assuming cornplete coding) over a large number of cells (z9x 3 = 1536). Considering only two characten at once is equivalent to "collapsing" the ten-dimensional contingency table over the eight unused dimensions, thus maximizing the number of data in the remaining cells. Unfortunately,

this prevents analysis of higher-order interactions among the variables, and also leads to

the possibility of spurious positive results being obtained through chance alone. as (at a =

0.05) five percent of the 45 pairwise tests would be expected to randomly show non-

independence of the variables involved. For this reason, ail apparent correlations should

be conservatively viewed only as potential indicators of biologically or taxonomically

relevant patterns.

Methods of analysis - morphometrics

In addition to analysis of the qualitative characters discussed above, an effort was

made to identih meaninel intragenenc variation in the quantitative proportions of the skull of Diictodon. Although any linear memurement may be affected by distortion, it was hoped that statistical techniques would ailow the recognition of significant patterns despite the bbnoise"introduced by distortion into the data set, given a suficient sample size. In contrast to previous morphometric studies of dicynodonts (Toilman et al., 1980;

Thackeray et al.. 1998), which relied on bivariate methods, the multivariate technique of principal components andysis (Momson. 1976; Shea. 19Sj was appiieci to the present sarnple. Intuitively, this involves combining a number of variables. in this case linear morphometric measurements such as skull length, to produce a new set of variables (the

"principal cornponents") that are inherently uncorrelated with one another.

Following Jolicoeur (1963), many worken applying the principal cornponents technique to morphometncs have focused on allometric change during the growth of an individual (e.g., Dodson, 1975; Shea. 1985; Stnuss, 1987). Men the original variables from w hich the principal cornponents are calculated are linear morphometric measurements, the first principal component almost always correlates positively with dl the starting variables and can therefore be interpreted as a multivariate measure of body size (Jolicoeur, 1963). Allomeq in the starting variables can then be detected on the bais of their loadings on the first pnncipd component, which are divided by the reciprocal of the number of starting variables to yield an "aIlomeûic coefficient" whose value indicates the degree of positive (for values exceeding unity) or negative (for vdues less than unity) allometry. The second and subsequent principal components may express non-allometnc differences in shape Oodson, 1975).

For purposes of the present study, the seven morphometric measurements selected for analysis (Table Ib; Fig. 16) were topological variables such as the total length of the skuil Figure 16. Schematicdly reconstructed skull of Diictodon feliceps. show ing measurements used in the morphometnc malysis. (A) Dorsal view of skull; (B) Lateral view of snout. Scale arbitrary; for key to numbers see Table 1b.

(snout to occipital condyle) and the minimum interorbital width. Some possible variables were rejected because they could be accurately measured in too few specirnens; principal components analysis requires that there be no rnissing data.

Al1 measurernents were recorded to the nearest 1 mm. For measurements that could be taken on either side of the skull (such as diarneter of the externd naris), values obtained from the left and right sides were averaged. Principal components analysis was appiieà to the variance-covariance rnatrix of the logarithmicaily transformed data, and the resulting eigenvectors were normalized so that the squared coefficients sumrned to unity for each principal component. The allometnc coefficient of each variable was calculated based on its loading on the fint principal component, as outlined above. The significance of the resulting principal components was evaluated both by jackknife analysis (MostelIer and

Tukey, 1977; Marcus, 1990) and according to the amount of variation in the data set explained by each principal component (Zelditch et al.. L 989). Jackkni fe anal ysis generates mean and standard error values for each eigenvector coefficient, by repeatedly recaiculating the coefficient with a single point omitted f'rom the data set; the ratio of the mean to the standard error (T) measures the robustness of each original result. A high value of T indicates statisticai significance, but alternative threshold levels of 3.00

(Gibson et al., 1984) and 5.00 (Marcus, 1990) have been suggested, and it is not clear which is preferable. Both are considered in the present study. The criterion of Zelditch et al. (1989) accepts a principal component as significant if the percentage of variation it explains is greater than the reciprocd of the number of starting variables.

Al1 principal component calculations were performed using version 1.60 of the computer program NTSYS-pc (Rohlf, 1990). Statistical Results - Qualitative Characters

Pair-wise independence tests of the ten qualitative skull characten resulted in only four significant correlations (Table 2), despite the potential for spurious positive results engendered by the large number of individual pair-wise tests. The presence of tusks appears to correlate posinvely with the presence of a pineai boss i~'=8.03, df = i, p c

0.005),and in tusked individuals the postparietal is also much more likely to be associated with the occipital surface rather than the intertemporal bar (f= 6.83, df = 2, p

< 0.05). In specimens with a deep labial fossa the fianges formed by the parietal in the intertemporal region almost always approach one another closely, whereas they are equally likely to be close together or widely separated when the labial fossa is weakly excavated (f = 5.04, df = 1, p < 0.025). Finally, the postorbitds are much Iess likely to completely overlap the parietal flanges in specimens having posttemporal ridges (f =

6.17, df = 1, p c 0.025).

Statistical Results - Morphornetrics

Eigenvectors and eigenvalues for the f'irst three principal components obtained are shown in Table 3. Al1 starting variables showed positive loadings on the firjt principal component, suggesting that this component represents a measure of overall size. It accounts for 83.5% of the total variance in the data set, greatly exceeding the threshold value implied by the number of starting variables (Il7 = 14.3%; see Zelditch et al., 1989).

Jackknife analysis shows that dl eigenvector coefficients are significant if the critical Table 2. Four pairs of qualitative characten showing statistical non-independence. Each two-way table shows the number of specimens displaying each combination of character

States, the average skull length of specimens displaying each combination (L, in cm), and the X' value, number of degrees of freedom and p-value for the two-wq table as a whole.

The total number of specimens varies among the two-way tables because of incomplete character coding in most specimens.

A. Tusks and pineal boss B. Labial fossa and parietal flanges

Canine tus ks Parietal flanges Pineal Boss Absent Presen t Labial fossa Close Widely toaether separated ...... ------Absent 15 9 Deep 11 1 (L = 7.7) (L = 9.1) (L=9.L) (LdL.5) Present 4 16 Shallow 6 6 (L = 9.2) (L = 9.9) (L = 9.6) (L = 8.8) f= 8.03, df = 1, p < 0.005 x'= 5.04, df = 1, p c 0.025

C. Post-temporal ridge and postorbiîal coverage of parietals D. Tusks and postparietal position

Post-temporal ridge Canine tusks Postorbital Absent Presen t Postparietal Absent Present Coveraze position Nea riy 10 1 Within inter- 5 2 complete (L= 9.1) (L = 10.3) temporal bar (L = 7.1) (L = 7.6) 6 8 8 7 Partial (L=9.1) (L4.8) In temediate (L= 8.3) (L = 9.1) xL= 6.17, df = 1, p < 0.025 3 13 WithinOcciput (L4.6) (L=I0.2) x2= 6.83, df = 2, p < 0.05 value for T is taken to be 3.00, and al1 but two are significant if it is taken to be 5.00. This high level of statisticai significance suggests that dometric coefficients calcuiated on the basis of the eigenvectors of this principal component (Table 3) are meaningful. The

strongest deviation from isometry involves skull length, which is negatively allornetric

(allometric coefficient = 0.83), and only for this variable does the 95% confidence

interval of the allornetric coefficient, calculated using jacklcnife results, exciucie uniry.

Thus, deviations from isometry involving the other variables mûy be regarded as

statistically insignificant. Tusked specimens score significantly higher on the first

principal component (t-test: n=28, df=26, p = 0.044; see Fig. 17), indicating that they are

larger on average than tuskless specimens. This difference cannot be directly translated

into an intuitively comprehensible average size discrepancy, but the fact that the average

intertempord width (to choose an approximately isometric quantity) was 1.8 for the

tusked specimens included in the principal components analysis and 1.5 cm for tuskless

ones gives a rough impression. The size difference cmbest be described as statistically

detectable but not large.

Principal components beyond the first are apparently insignificant. The second and

third principal components explain only 6.8% and 5.1 % of the total variance,

respectively, and thus fa11 below the 14.3% threshold. Jackknife analysis shows no

significant eigenvector coefficients for the second principal component even at the 3.0

significance level, and only two for the third principal component (though their T values

fa11 below 5.0). Subsequent principal components explain even Iess variation and,

although they were not subjected to jackknife anaiysis, can presumably be regarded as

insignificant . Figure 17. Positions of 28 specimens of Diictodon feliceps on the fint two principal components obtained during a morphometric analysis of cranial proportions. Squares represent tusked specimens and circles tuskless specimens. Tusked specimens tend to score higher on the fint principal component, but (if the single tuskless outlier is disregarded) there is little or no separation into distinct "clusten" of proportionally sirnilar individuals.

Table 3. Eigenvectors and eigenvalues (expressed in terms of percentage of variance

explained) associated with the first three principal components obtained during a

morphometric analysis of the skull of Diictodon. Allornetnc coefficients for the starting

variables have been calculated on the basis of their loadings on the fiat principal

component, which appears to represent physical size. Each eigenvector is normalized so

that the sum of the squares of its coefficients is equal to unity. Jackknife analysis was

used to compute a significance value (T) and a 95% confidence interval (expressed as a

standard enor value in parentheses, and also applied to the allometric coefficient) for

each eigenvector coefficient; eigenvector coefficients with T values exceeding 5.0 are

considered statistically signi ficant.

* I Variable Eigenvector Coefficient T Allometric Coefficient l - Skull length 0.3 14 (+ 0.03) 20.97 0.83 1 (20.08) 2 - Internarial width 0.393 (f 0.57) 13.94 1.041 (+ 1.51) 3 - Interorbital width 0.344 (+ 0.13) 4.95 0.9 10 (+ 0.34) 4 - Intertemporal width 0.399 (f 0.07) 12.88 1.055(f0.19) 5 - Snout length 0.369 (+ 0.10) 7.89 0.975 (I0.26) 6 - Nariai diameter 0.368 (f 0.17) 4.78 0.973 (20.45) 7 - Narial elevation 0.445 (f 0.12) 7.46 1.178 (+ 0.33) Variance explained - 83.53 %

Second Principal Component Variable Eigenvector Coefficient T 1 - Skull length - O. 155 (+ 0.63) i -62 2 - Internarial width 0.089 (+ 0.52) 1.69 3 - Interorbital width - 0.528 (+ 1.67) I .39 4 - Intertemporal width 0.006 (t 0.67) 1.12 5 - Snout Iength - 0.198 (+ 1.37) 1.50 6 - Narial diameter - 0.2 17 (+ 0.82) 1.52 7 - Narial elevation 0.776 (f 3.16) 1.67 Variance explaineci - 6.82 % Third Principal Component Variable Eigenvector Coefficient T 1- Skull length - 0.067 (1 0.29) 0.5 1 2 - intemarial width 0.103 (& 0.3 1) 1.35 3 - Interorbitd width 0.659 (+ 0.80) 2.22 4 - Intertemporal width - 0.376 (f 0.38) 3.03 5 - Snout length - 0.578 (i 0.43) 3.53 6 - Narial diameter - 0.0 14 (f. 0.64) 0.35 7 - Narial elevation 0.375 (k 1.18) 0.77 Variance explained - 5.06 % Field data gathered by Smith (1989) confirm that tusked Diictodon individuals have greater total skull length (t-test: n=246, d-44, p = 8.94 x 10'"). Although skull length is only an approximate indicator of overall size, the fact that it is negatively ailometric implies that skull length cornparisons should underestimate rather than overestimate size discrepancies; therefore, the implication that tusked skuIIs tend to be Iarger is valid.

These sarne data indicate that localities where Diictodor~1s abundant invariabiy yieid botii tusked and tuskless specimens; though the relative numbers of the two morphotypes were occasionally uneven, they were generally comparable, if not exactly equd. A total of 203 tusked skulls and 173 tuskless skulls were identified in the course of Smith's (1989) field

work, a proportion (46% tuskless) that agrees remarkably well with that recorded in the present sample (21 out of 49 tuskless, or 43%).

Patterns of variation

The fact that only four pain of cnnial chiiracters (out of 45) showed statistical non-

independence suggests that much of the observed anatomicd varation in Diictodon is essentially "random". in the sense that it probably cannot be linked to clear differences between species, sexes, age classes, or even populations. Any of these biologically meaningful sources of variation would be likely to act on several chancters at once,

resulting in a suite of intercorrelated features. This impression of overall randomness is reinforced by the fact that inspection of the data mahix (Table AL) does not suggest that any of the recorded variations are associated with particular localities or stratigraphie zones. Although the power of the analysis is somewhat compromised by large numbers of

missing data and occasional dificulties in separaùng out discrete character States, the conclusion that few cranial variations in Diictodon are taxonornically or palaeobiologically relevant is unavoidable. At least with regard to taxonomy, King

(1993b) reached a similar conclusion, arguing that no characters could be found to distinguish multiple Diictodon species. However, the four statistically supported correlations may reveal genuine patterns, and are worthy of further consideration.

The positive association berween tne presrnçe of tusks and thc presenîe of thc pincal boss is quite clear (Table 2): the boss occurs in 16 of 28 tusked individuals and only 4 of

21 tuskless ones. As noted by King (1993b), skull size also seems to be a factor in determining the presence or absence of a boss, but the relationship is not a simple one

(Table 2). Among tuskless skulls, three of the Four specimens bearing pineal bosses occupy the extreme upper end of the size range, while among tusked specimens bosses are more evenly distributed and occur in small as well as Iarge individuals. The most straightfonvard conclusion is that the pineal boss developed frequently in tusked individuals without much dependence on size, whereas in tuskless individuals it appeared rarely and was genenlly limited to the largest specimens.

The presence of tusks also correlates positively with an occipitally positioned postparietal bone (seen in 13 of the 28 tusked individuals, in contrast to only 3 of the 21 tuskiess individuals). In this case size dependence seems somewhat more significant, because for both tusked and tuskiess specimens the larger individuals tended to have an occipitally placed postparktai; small size was associated with the postparietal's lying at the back of the intertemporal bar, while specimens in which the position of the postparietd was intermediate were also intermediate in size. However, the fact that each position of the postparietd is associated with markedly different skull lengths in tusked as opposed to tuskless individuals (Table 2) implies that the association between the presence of tusks and the position of the postparietal does not result merely from the fact that both characters are affected by size.

The rather surprising association between presence of the posttemporal ridge and incomplete coverage of the parietal flanges by the postorbitals is stnking (only one specimen has àoth the posttempotai ridge ÿnd a high Jegre2 of cwerage) but ma): k compromised by the fact that the presence or absence of the posttemporal ridge could not be determined in many specimens owing to poor preservation or incomplete preparation of the occiput. Similarly, the association between development of the labial fossa and proximity of the parietai flanges rests largely on a single rare combination (only one specimen has both a swng labial fossa and widely separated flanges) and is slightly compromised by uncertainty in the coding of both characters.

With regard to the morphometnc results, tusked specimens are slightly larger on average than tuskiess ones. Negative allometry of skull leneth is aiso well supported statistically.

Causes of variation

As noted above, many variations in the cranial anatomy of Diictodon show so little consistency or mutual CO-occurrencethat they must be placed under the convenient but uninformative heading of "random intraspecific variation" - even if multiple species of

Diictodon exist, present evidence suggests that they must have shared much of this apparently meaningless variability. It presumably results from a combination of environment. intluences and underlying genetic variation in Diictodon populations. The patterns described above, however, are potentially meaningful, and may refiect difference between sexes, species, geographically remote conspecific populations, ontogenetic stages in the development of individuals, anagenetic stages in the evolution of the genus as a whole. or some combination of these factors; altematively, of course, they may involve characters that happen to covary simply because both are affected by some genetic or environmentai influence that shows no consistent parrern of occurrence and is therefore not of interest in the present context. As with most palaeobiological problems, the available data do not provide a basis for distinguishing among these alternatives with absolute certainty. However, it rnay be possible to identify the most

"parsimonious" explanation for each set of associated variations: that is, the explanation that accounts for the facts with a minimum of unwarranted assumptions and inferences.

It seems unlikely that any of the observed variations can be ascribed to anagenetic change. As noted above, the fact that the majority (at least 26 out of 49) of the specimens were collected in Tropidostoma Assemblage Zone stnta imposes limits on the potential of the present sample to reveal long-term evolutionary changes in Diictodon, particularly as most of the remainder corne from the immediately overlying Cistecephalus Zone. The stratigraphic range of Diictodon incorporates both the Dicynodon Zone, which lies above the Cistecephnius Zone (Fig. 1). and the Pristerognathus Zone, which lies immediately below the Tropidostoma Zone (Rubidge et al., 1995). The genus may also occur in the

TapinocephaIus Zone, which foms the base of the Beaufort Group, but as all putative

Diictodon specimens hom these strata are relatively pooriy preserved it has been suggested that they may in fact represent the closely related Robertia (King, 1993b). Only one specimen in the present sample, SAM-PK-11563, is from the Tapinocephalus Zone; it is severely laterally compressed, and rather badly worn, but lacks any sign of the postcanine teeth characteristic of Robertia. It shows no obvious anatomical differences from Tropidostomu Zone and Cistecephalus Zone skulls, and the same applies to the few available Pristerognathus and Dicynodon Zone specimens. Anagenetic trends in

Diictodon may exist, but could only be detected with a more stratigraphically diverse collection of matend.

The association between the presence or absence of a posttemporal ndge and the degree of overlap of the piinetal flanges by the postorbitals is unexpected and rather difficult to interpret, particularly as neither matornical feature has been analysed in terms of its possible biological or functional significance. Although the parietal flange probably formed an attachment surface for the jaw adductor muscles (King, 1988), the topological structure of the tidge is essentially independent of the degree to which the extremely thin postorbital overlaps the parietal, and it is difficult to see any functional relevance in this variation. The posttemporal ridge rnight relate to the insertion of muscles onginating from the axis-atlas cornplex, as Cluver (197 1) reconstructed a muscle he termed the obliquus capitis cranialis as inserting near the posttemponl fenestra of Lystrosaztn~s

(though below, rather than above, this structure). However, there could hardly have been any direct functional connection between the posttemporal ndge and the parietal flange, which was associated with the musculature of the mandible rather than of the occiput.

Alternatively, the fact that the two features in question seem to be hnctionally unrelated but statistically non-independent might be taken to have some taxonomie relevance.

However, their non-independence mainly reflects the rarity of one combination of character states, and hence does not suggest a consistent difference that might be useful in diagnosing taxonomic groupings at either the specific or subspecific level. Ultimately a developmental or functional explanation seems most likely, but it is impossible to be more specific at the present time.

The link between depth of the labial fossa and distance separating the parietal flanges in the intertemporal region is sirnilarly difficult to explain, and for similar reasons. A functional correlanon between these suucnires again seems uniikeiy in the exireme: ihe labial fossa probably accornrnodated blood vessels innervating the snout (Cluver, 197 1) or even held a salivary gland (Ewer, 1961), whereas the parietal Ranges were (as explained above) probably involved in anchoring the jaw adductor musculature. A taxonomic interpretation is equally difficult, as the statistical non-independence of the two characters once again rests on the extreme rarity of a single combination. As in the preceding case, these features cannot provide a solid basis for recognizing taxonomic. sexual or ontogenetic differences within Diictodon; the nature of each association is too arnbiguous.

The only remaining point of potentially significant intrageneric variation is the frequently discussed distinction between tusked and tuskless specimens. The andysis presented above provides evidence that the presence of a pineal boss, the position of the postparietal bone, and the size of the skull are al1 statistically linked to the occurrence of tusks, suggesting that the difference between tusked and tuskiess individuais is in some way biologically meaningful. It appean that tusk occurrence is only the most obvious of a complex of linked characters that together serve to distinguish multiple "aitegories" of

Diictodon. One category includes individuals that have tusks, often have a pineal boss regardless of skuIl length, often have the postparietal bone positioned fully on the occipital surface, and tend to have large skulls. The other category includes individuals that lack tusks, usuaily lack the pineai boss (except in the largest specimens), generdly have the postparietal bone associated partially or entirely with the intertemporal bar, and tend to have smdl skulls.

There is universal agreement in the literature that the distinction between tusked and tuskless individuals in many dicynodont taxa relates to a difference between bioiogicaiiy important categories (eg.,Broom, 1935; Barry. 1957; King, 1988). These categones have been variously interpreted as entirely separate taxa (see Broom [1935] on the debate regarding the status of the tusked Dicynodon and the ~sklessOudenodon), ontogenetic stages (Barry, 1957) and opposite sexes (Broom, 1935; Toerien, 1953). The essence of the problem lies in deciding which of these possibilities represents the most parsimonious explanation for the anatomid differences between tusked and tuskless specimens and the observed distribution of the two morphotypes in time and space.

An ontogenetic explanation for the difference between tusked and tuskless skulls would seem to require a large and consistent size difference between the two morphotypes, in addition to a significant number of intermediate specimens with small or newly erupted tusks. Entirely tuskless specirnens would represent juveniles. Tusked and tuskless specimens would show almost identicai geographic and stratigraphie distributions, frequently occming together, as both are considered to be denved from a single population under this hypothesis.

Barry (1957) described a sample of 37 skulls, identified as "Dicynodon" (now

Diictodon) grimbeeki that met at Ieast some of these expectations. Only four specimens lacked tusks entirely, while seven had small or rudimentary ~sks,and most (perhaps dl) of the skulls were from a single area (the farm Leeukloof). However, Barry (1957) did not give a clear indication of the sizes of the specimens he studied, and his illustrations

(Barry. 1957, Fig. 13) show at best a vague correlation between skull size and the development of the tusk. They also reved (if the palatal rim was accurately drawn) that al1 five of the figured specimens lack the distinctive palatal notch of Diictodon and thus may be misidentified skulls of the generally sirmlar but al ways tusked Dicynodon. if this is the case then the overwhelming prevalence of tusked specimens (33 out of 37, cornpared to 28 out of 49 in the present sample) would be exactly as expected.

Given these uncertainties, the work of Barry (1957) failed to establish that tusked

Diictodon skulls represent adults, and tuskless ones juveniles, and evidence from the present sample suggests that this explanation is unlikely. There is considerable size overlap between the larger tuskless skulls and the smaller tusked skulls, although tusked skulls are slightly larger on average. Among specimens complete enough to be included in the principal components analysis, thret of the 12 tuskless skulls actually scored hi@ enough on the first principal component to exceed the average for tusked specimens

(0.035), and three of the 16 tusked specimens conversely fell below the tuskless average

(-0.046). Furthemore, the distribution of specimens dong the first principal component

(Fig. 17) reveals no obvious separation between the two morphs. These resulu suggest that the size difference between tusked and tuskless specimens is too small for these categories to represent adults and young juveniles respectively, and the tuskless skulls are generdy (apart from the smallest specimen, SAM-PK-K5204; skull length 4.7 cm) not characterized by open sutures or other obviously "juvenile" features. In fact, both tusked and tuskless skulls show varying ossification, with some (presumably younger) having smooth bone surfaces and clear suture lines, and others (presumably older) having a more rugose texture and Iess distinct sutures.

If the tuskless specimens are viewed as older juveniles or sub-adults, the near-absence

of specimens with extremely diminutive or newly empted tusks is alrnost inexplicable.

Only one skull (SAM-PK-10078) shows evidence of tooth replacement, and the tusks of

the shortest tusked skull (SAM-PK-K697Y; skull iengrh 7.4 cm) are if anything unusudiy

long and well developed. It is possible that a few of the very smallest specimens in the

sample (particularly SAM-PK-K5204) are sub-adult, but the fact that the tuskless skulls

are generally well ossified and nearly as large on average as their tusked counterparts

implies that the majonty of them must represent tuskless adults. The smallest tusked

specimen reported by Smith (1989) had a skull length of 6.2 cm. and as his specimens

were cataiogued in the field (i.e.,without prepantion) the tusks rnust already have been

large in this case in order to be visible. It seems likely that tusks normally enipted, when

at all, sometime between the attainment of skull lengths of 4.7 cm (the length of the only

apparent juvenile in the present sample) and 6.2 cm (the length of the smallest known

specimen with well developed tusks).

If this is the case, and if the eruption of tusks can be viewed as an indicator of

maturity, then al1 specimens in the present sample other than SAM-PK-K.5204 are large

enough to be considered adult. As growth was likely more or Iess indeterminate in

Diictodon, very small tusked individuals generally represent the youngest adults of the

tusked category (whether a sex or a species). Very smdl tuskless skuils are much more

cornmon because tuskless individuais, king slightly smaller on average, must have

generally grown more slowly - at least beyond a certain point - and would therefore occupy the extreme low end of the adult size range for a longer period of time. Even if one or two small individuals have been inadvertantly grouped with the tuskless specimens simply because their tusks remained unerupted at the time of death, the existence of both tusked and tuskless adults is clearly indicated by the data and must be explained. The distinction between the two morphs cannot be fully explained in ontogenetic texms.

It is more difficult to discriminate between the two rernaining possibilities, that the tusked and tuskless specimens represent different species or alternatively that they represent different sexes within a single species. Both hypotheses agree that the presence or absence of tusks would likely correlate with other morphological features (either secondary sexual characteristics, or apomorphies of one of the two species), and both cm accommodate the observed difference in average skull size (either sexual size dimorphism, or the difference between a smaller species and a larger). King (1993b) invoked the occurrence of tusked and tuskless forms in approximately equd numbea at one panicular site as evidence for the sexual dimorphism hypothesis, but there is no reason to believe that males and females would necessarily be equally numerous in

Diictodon populations (although they might be). By contrast, Barry (1957) argued that sexual dimorphism was improbable because cases of tooth dimorphism in living animals usually involve differences in tooth size, as in the larger canines of many male primates and the Iarger incisor tusks of male elephants, rather than the presence of a tooth in one sex and its total absence in the other. The value of these highly denved synapsids as living analogues for a dicynodont is questionable, as they share many denved characteristics that were demonstrably absent in dicynodonts, and if one must play the game of haphazard neontological cornparisons it is equdy valid to point out that a few mammals do show sexually dimorphic differences in dental formula. The female dugong lacks incisors, which are well developed in the male (Mitchell. 1973); the male narwhal has an extended incisor tusk, while the incisors of the female do not erupt (Fraser. 1938, contra Barry. 1957); canine teeth are normal in male homes, but not in females (e.g.,

Shine, 1989); and in beaked (ziphiid) whales, antenor mandibular teeth occur oniy in the male (Mead and Payne, 1975). In at least some of these cases vestigial teeth do form within the jaw of the female, but they do not erupt and hence are not externally visible. It therefore seems entirely possible that the presence of tusks might be limited to one sex in

Diictodon, but the alternative view that tusks were present in only one of two separate

Diictodon species is also plausible.

However, the "species difference" and "sex difference" hypotheses differ in their predictions regarding the probable distribution of tusked and tuskless specimens. If the difference is sexual, then tusked and tuskless individuals coexisted as part of a single

population in life and should frequently occur in proxirnity to one another. If the two

morphotypes represent separate species, however. some degree of geographic or temporal

partitioning would be highly likely. Tusked and tuskless specimens are so similar in size

and morphology that they probably could not have coexisted sympatrically without one extirpating the other by cornpetitive exclusion, or at least becoming much more abundant.

While it is possible that some behavioural factor (such as different foraging times or

prefemd diets) might allow sympatry, the balance of probability favours a more

allopatric distribution. A second difference between the hypotheses lies in the nature of

the feahires that would be expected to comlate with the occurrence of tusks in each case. Assuming the difference to be sexual, the correlating characters should be structural features such as skull crests or bosses that might act as conspicuous signals to the opposite sex or useful weapons against rivals; in fact, the structure of the nasal boss may be dimorphic in Aulacephdodon (Tollrnan et al., 1980) and the presence of supraorbital bosses may characterize males of Lystrosaurus (Thackeray et al., 1998). If on the other hand the variable occurrence of tusks is taxonornic. the correlating features mght equaliy well be differences in sutural patterns or other osteologica1 characters that have little effect on the external fonn of the head.

Applying these criteria to the present sample suggests that the sexual hypothesis is more likely to be correct. Smith's (1989) field data indicates that tusked and tuskiess specimens cornrnonly occur together, and many individual localities yield approximately equal numbers of the two morphotypes. This distribution provides circumstantial evidence in favour of the hypothesis of sexual dimorphism, as does the fact that the tusked morph (presumably male) is slightly larger and more likely to be equipped with a pineal boss (paralleling the possibly dimorphic occurrence of skull bosses in

Lystrosaums). The difference in the position of the postparietal between the two morphotypes is less readily interpreted, but seems to be partly linked to the larger size of tusked specimens. While these differences could conceivably be interspecific rather than intersexual, the conspicuous nature of the pineal boss and the dimorphic nature of skull bosses in other dicpodonts imply that the positive correlation between the pineal boss and the canine tusk may be interpreted as strongly suggestive of sexuai dimorphism.

In sum, the existence of two CO-occurringmorphotypes, differing chiefly in feams cornmoniy associated with sexual dimorphism in other taxa (size, tusks, and skull bosses), can be readily interpreted in terms of a single dimorphic species. Based on the available evidence, this explanation is more plausible than the alternative possibility that the morphotypes represent sympatric sister species that happen to differ only in features that might be expected to characterize opposite sexes.

Ecological Implications of Sexud Dimorphism

Among living animals. sexual dimorphism is a widespread phenomenon in both vertebrates and invertebrates, and has clearly evolved independently in numerous taxa

(eg, Shine, 1989). In amphibians, for instance, the most common intersexual ciifference is perhaps sexud size dimorphism (SSD),with female anurans and urodeles tendinp to be larger than conspecific males (Shine, 1979). The reasons for this may be complex, but the association between large size and enhanced female fecundity must provide at lest a partial explanation. However, in some species the male is the larger sex, and this reversed pattern shows a clear statistical correlation with the occurrence of combat between rival males (Shine, 1979). In murans, such combat may also lead to the development in the male of enhanced mament (i.e., structures that are pnmarily used, if not for actud fighting, then to threaten and intirnidate rivals) such as tusks or spines. Although other factors (e.g., the ability of a male to maintain control of the female during courtship and mating) may favour the evolution of large male body size, the presence of male combat seems to be the major determinant in most cases, and certainly in those involving weaponry as well as SSD.

In Iiving reptiles, dimorphism in features other than size and body proportions is oiten difficult to detect, especidly in the skeleton (e.g., Dodson, 1976). However, the association between larger male size and the occurrence of male combat is statistically apparent at least in snakes (Shine, 1994) and certain herbivorous lizards (Carothers,

1984). In turtles, Berry and Shine (1980) found that males tend to be larger not only in species displaying male combat (which are usually terrestrial) but also in species in which rape is cornmon (which are usudly semi-aquatic). Male head size is proportionately greater in many lizards that show maie combat (Carothers, i964; Shine, i989j, presumably because of the offensive advantage of a powerful bite. However, head size dimorphism may dso be associated with differences in feeding niche between the sexes

(Shine, 1989).

Among birds, the most obvious manifestations of sexual dirnorphism involve differences in plumage, which is often more colourful and elaborate in males. Sexual differences in body size and bill shape exist in some species, but generally seem to relate to differing male and female foraging patterns rather than male combat or other forms of sexual selection (Shine, 1989). In marnrnals, by contrast, SSD and intersexuai differences in marnent are widespread and often extreme, particularly in artiodactyls, pinnipeds, elephants and primates (Rails, 1977). Al1 of these taxa show some variability in the pattern of sexud dimorphism, but extreme cases generally involve large, well-med males and small, poorly-armed fernales. The association between large maie body size, large male canine size, and a polygynous mating system is well documented in a wide range of primates (e.g., Leutenegger, 1982), and polygyny is thought to promote cornpetition and hence aggression arnong rival males because it introduces large potential disparities in male reproductive success. In cervids and bovids a high degree of sexual dimorphism in size and armament (with the male being usually Iarger, and always better armed, where differences exist) generally implies a polygynous mating system and a high incidence of antagonistic interactions among mdes; they may either compete directly for females or compete for spacious, resource-rich temtories that are likely to attr;ict females

(Jannan, 1983). Less dimorphic forms are usuaily either small, cryptic, monogamous species in which cornpetition for mates is minimal (various small deer and antelope) or

Iarger species in which femaies are frequently required to fend off predators and unwanted male suitors and must therefore be large and well-armed (e.g., oryx and wildebeest). In pinnipeds there is a direct correlation between degree of sexual dimorphism and average harem size (Wiig, 1985).

Although the ecological and behavioural factors that give rise to sexual dimorphisrn are certainly cornplex. in mammals as in other taxa (Rdls. 1977; Shine, 1989) the occurrence of male intrasexual conflict in species in which the male is significantly larger and better medthan the femaie seems to be an almost univenal pattern. Although some animals that are minimally dimorphic, such as the gibbon, show a high incidence of aggression between males (Kinzey, 1972), there are few if any cases in which dimorphism in size and (particularly) armament occurs in the absence of such aggression.

Furthemore, dthough size differences reflect a multiplicity of evolutionary influences and may occur in either direction, the author is unaware of any tetrapod in which the fernale is better armed than the male. This pattern must reflect fundamental differences in the reproductive and behavioural roles of males and females within tetrapod populations, but for present purposes its evolutionary causes are of less interest than its empirical reliability. Placing Diictodon into the context established by these neontological cornparisons dlows two major conclusions to be drawn regarding its sexual dimorphism. The first of these is that the tusked sex was almost certainly the male, as a case in which the fernale was equipped with tusks and (cornmonly) a pineal boss while the male lacked these features would evidently be hiphly unusual and probably unique among tetrapods. The

Iarger skull size of tusked specimens, whether or not retlected in the size of the body as a whole. is also consistent with this conclusion. Secondly, the correlation between male armament and mde aggression is prevalent enough in living tetrapods to suggest diat it might also apply to Diictodon, asuming that the rusks and pineal boss were used to intirnidate and perhaps actually atrack nvals. In modem groups as disparate as anurans, primates, ungulates and pinnipeds, which differ radically in size and mode of iife, a high degree of dimorphism in size and weaponry is genenlly reflected in a high incidence of male combat and often a polygynous mating system. It is therefore probable that the primary function of the tusks of Diictodon lay in intrasexual competition over temtory or females, involving either combat or antagonistic visual dispiays, though a secondary role in predator defense and perhaps even burow excavation (see Smith, 1987) certainly cannot be mled out. It is unlikely that tusks had any essential role in feeding, as females clearly must have been able to feed in the absence of tusks and the masticatory apparatuses of the two sexes are otherwise so similar as to render unlikely any suggestion that their diets were significantly different.

Finally, it should perhaps be noted that the observed dimorphism in tusks and pineal bosses in Diictodon is among the most "marnmdian" features occurring in dicynodonts, given that differences in marnent are quite common in rnammals and extremely rare in other extant tetrapod groups (apart from anurans, which clearly are not closely related to dicynodonts). It is of coune unclear how closely this degree of dimorphism may be expected to correlate with other rnarnrnalian characteristics such as endothermy and lactation, but it is notable that extensive dimorphism and its presurned behavioural correlates seem to have developed in at least one relatively basal therapsid clade rather than being resmcted to denved, mammaiian synapsids.

DICYNODONT XNTERRELATIONSHIPS

The most recent, comprehensive, and detailed phylogeny presently available for dicynodonts is that of Cluver and King (1983), subsequently expanded by King (1988).

An important feature of this phylogeny (Fig. 18) is the appearance of the relatively recently discovered Eodiqnodon (Bq,1974) as the most basal dicynodont, followed by Endothiodon. More derived genera are divided into two major clades. labelled "J" and

"V" on the dadogram. Clade J notabiy includes Prisrerodon, Dicynodon, Oridenodon,

Arilaceplialodon and nearly ail Triassic dicynodonts; the Triassic genera form a clade, with Dicynodon as their sister-taxon. Clade V is an assemblage of relatively small fonns. from the basal Kingotfa to more derived genera such as Emydops. Diictodon and

Robertia, sharing (among other features) a unique and highly specialized feeding apparatus, appear as sister-taxa within this clade. The phylogeny was interpreted by

Cluver and King (1983) as indicating nurnerous independent losses of postcanine teeth, as

Watson (1948) implied when he suggested that the notorious wastebasket genus

Dicynodon probabl y had multiple evolutionary origins (Le ., was polyphy letic). Despite the importance of this phylogeny in advancing a clear and generally well

founded hypothesis of dicynodont relationships, the fact that it was produced "by hand"

rather than by a cornputer prograrn such as PAUP (Swofford, 1993) represents a potential

weakness. Unlike PAUP,a human cladist is not guaranteed to select the most

parsimonious pattern of relationships indicated by the data; although the cladogrm

certainly represents one plausibie hypothesis, the possïoiiity remains that some more

parsimonious alternative exists. The problem is highlighted by cornparison with an

alternative (and less detailed) phylogeny (Fig. 19) produced on the basis of feauires of the jaw and palate, apparently by similar methods (Cox, 1998). Although the two cladograms

are clearly incompatible, there is no way to determine which (if either) represents the

most parsimonious phylogenetic hypothesis for dicynodonts without recourse to a

cornputer-assisted cladistic analysis incorporating as many characten and taxa as

possible. In addition to its taxonomie significance, a reliable dicynodont phylogeny

would provide a helpful framework for discussions of specific aspects of dicynodont

evolution, such as recent studies of the development of the feeding system by King et al.

(1989) and Cox (1998). The present cladistic analysis of dicynodont interrelationships

thus represents an attempt to improve on previous studies by applying a rigorous

parsimony algonthm rather than subjectively assessing the importance of putative

characters. Although it incorporates fewer genera than were considered by Cluver and

King (1983), it includes members of al1 major lineages postulated by these authors,

facilitating cornparisons between their results and those of the present study. However,

the point of main interest is the phylogenetic position of Diictodon with respect to

broadly similar taxa such as Dicynodon, Robenia, and Oudenodon, as the new anatomical Figure 18. Dicynodont interrelationships according to King (1988). Note that individual genera have been substi tuted for the higher-level terminal taxa appearing on the original cladogram, and that al1 genera not included in the present snidy have been omitted; however, the positions of al1 taxa are those assigned by King (1988). Clade designations outside parentheses are those used in the present study; those within parentheses follow

King (1988).

Figure 19. Relationships among major dicynodont lineqes. according to Cox ( 1998).

Among genera included in the present anal ysis, Pristerodon and Endothiodon are included by Cox within the endothiodontoids; Robenia and Diicrodon within the robertoids: Kingoria and Eniydops within the emydopoids: and Oudenodon,

Aulacephalodon, Diqnodon, and Lystrosaurus within the dicynodon toids. data on Diictodon presented above (see Osteological Description) have more bearing on this problem than on other aspects of dicynodont interrelationships.

Methods

The phylogenetic analysis included 11 dicynodont taxa: Eodiqnodon (Rubidge, 1990;

Cluver and King, 1983; King and Rubidge, 19931, Endoriiiodon iCox, 1964; Ciuver ma

King, 1983; Latimer et al.. 1995). Pristerodon (Cluver and King, 1983; King and

Rubidge. 1993), Robenia (Cluver and King, 1983). Diictodon (Cluver and Hotton,

198 l), Diqnodon (Cluver and Hotton, 198 l), Oudenodon (Keyser, 1975; Cluver and

Hotton, 1981), Aiilacephalodon (Tollman et al.. 1980; Cluver and King, 1983).

Lystrosaiinis (Cluver, 1971), Kingoria (Cox, 1959; Cluver and Hotton, 198 1). and

Emydops (Cluver and King, 1983; King and Rubidge, 1993; Fourie. 1993). It will be noted that only Permian forms were considered, apart from the single Triassic genus

Lystrosaums, as most of the remaining Triassic dicynodonts appear to form a well defined clade with a close relationship to Lystrosaiirus (King, 1988). The Permian taxa chosen were intended to be representative of the major groups identified by Cluver and

King (1983) and Cox (1998). Two relatively well known basal anomodonts,

Patranomodon (Rubidge and Hopson, 1990) and Siiminia (Rybczynski. 1996). were used as outgroup taxa. As most dicynodonts are known mainly from skull material, al1 49 characten used in the analysis were cranial (see Appendix 2 for character definitions, and

Appendix 3, Table A2 for the data rnatrix). Unfortunately, well preserved and fully prepared specimens were available for only two of the genen considered in the analysis,

Diictodon and Pnsterodon, and data relating to al1 other taxa were gathered from descriptions and illustrations in the literature (see above citations). Although good descriptions were available in most cases. the lack of necessary details for some taxa - most notably Aulacephalodon, Endothiodon, Robertia and, surprisingly, Dicynodon - limited the scope of the data matrix and forced a number of characters to be dropped from the analysis because they could have been confidently scored in only a few taxa. It is clear that further descriptive work, raking accounr of variation wiîhin species iuid pnera and concentrating particulaily on the postcranial skeleton, is needed in order to provide the raw data required for elucidation of the phylogeny.

Multistate characten were treated as unordered, and ail characters were equally weighted. Although most characters were qualitative. a few were quantitative measures of either a single distance (e.g., skull length) or the ratio of two distances (e.g., length of the vomerine septum divided by length of the interpterygoid vacuity). For quantitative characten the numerical boundaries between character States were chosen to occupy the

Iargest gaps in the distribution of observed values, converting the continuum of values into discrete clusten as naturally as possible.

AI1 terminal taxa were genen, but many (such as Diictodon) are monotypic. Although the ideal approach in coding a polyspecific genus is to attempt to infer the ancestral state on the basis of relationships within the genus (Wiens, 1998), the lack of complete descriptions of individual dicynodont species rendered this methodology impractical. The most primitive species of Lystrosauncr is thought to be L. cunatris (Cluver, 1971), and

Lystrosaums data were drawn from this species wherever possible; for other polyspecific genera, including Eodicynodon and Emydops, data had to be gathered more h ap hazardl y from the available descriptions. Polymorphism within a species presents a hindamentdly different problem, and was explicitly coded as polyrnorphism unless one character state was ovenvheImingly predominant (present in at lest 75% of known specimens).

MacClade 3.0 (Maddison and Maddison, 199 1) was used to constnict the data matrix and to examine character changes once the most pmimonious trees had been identified.

Tree searching was carried out using the branch-and-bound algori thm in PAUP 3.1.1

(Swofford, 1993). In order to evaiuate ciade support, the progrm TreeRot (Sorensùn,

1996) was used in combination with PAUP to determine Bremer decay indices (Bremer,

1988) for al1 clades appearing on a strict consensus of the most parsimonious trees discovered in the analysis; the decay index represents the number of extra steps needed to produce a tree on which the clade in question does not appear. Bootstrap values

(Felsenstein, 1985), representing the percentage of cases in which a clade appears if the analysis is repeated with a nndomly drawn set of characters, were calculateci on the basis of 200 heurisuc search replicates within PAUP.

Results and Discussion

Four most parsirnonious trees were found, each with a length of 123 steps and consistency index of 0.602 (though only 0.534 when uninformative characters are excluded). Although a strict consensus of the four trees (Fig. 20) shows several

interesting features, it is evident from the low consistency index that the phylogeny is not

a particularly strong one. This impression is confirmed by the decay indices (Table 4),

which show that dl clades (including the entire ingroup) can be collapsed with an

additional step or two, and by the generally low bootstrap values. The weakness of the

tree is a significant result in itself, insofar as it demonstrates the need for additional work in gathering anatomical data in dicynodonts and identifying possible synapomorphies. As noted above, several potentially relevant characters (such as the occurrence of a longitudinal median ndge on the anterodorsal surface of the premaxilla in at least some dicynodonts) were excluded from the data maûix because the best descriptions available for several genera did not indicate clearly which character state was present, and examination of large numbers of specimens would probably ailow the data matnx to be considerably expanded for this reason alone. In other cases, of course, insufficient information on a particular character in one or a few genera led to an "uncertain" character state being scored for the genera in question, without removal of the entire character. An additional factor responsible for the weakness of the tree is the prevalence of autapomorphic chmcten in the data matrix, as indicated by the drastic &op in the consistency index when uninformative characters are removed from consideration. Seven of the 49 characters probably represent autapomorphies of individual genera and therefore provide no phylogenetically useful information.

Because it is difficult to place much confidence in this phylogeny, it seems inappmpriate and unnecessary to analyze its structure in great detail, and still more so ta use it as a basis for a discussion of dicynodont tûxonomy or patterns of character evolution. However, a few clades are of potential phylogenetic interest and require brief consideration.

Apomorphic character changes associated with each clade are listed and comrnented on, but note that only apomorphies occuning on al1 four most parsimonious trees are considered. Apomorphies marked with an asterisk (*) have ambiguous histones, in that their distribution changes depending on whether the ACCTRAN or DELTRAN Figure 20. Pattern of dicynodont interrelationships obtained in the present analysis. Note

the unexpected occurrence of Endothiadon nther than Eodiqmodon as the most basal

dicynodont. the sister-group pairing of Robenia and Diictodon, and the fact that

Aulacephalodon, Dicynodon and Lys~osazînrrform a clade.

Table 4. Two alternative measures of support for clades obtained in the phylogenetic

analysis. Decay support is the number of extra steps (i.e. character transitions) needed to

generate a tree on which a aven clade does not appear; bootstrap support estimates the

percentage of tnals in which a random selection of characters will generate a tree on

which the clade does appear. Higher numben are indicative of greater support in both

columns. See Figure 20 for definitions of clades.

Clade Decay Support (#steps) Boo tstrap Support (%) Dic ynodontia 1 80 Clade A -3 68 Clade B I 35 Robertiidae 1 47 Clade C 1 45 optimization option is preferred. Each chancter change is listed explicitly, with an anow indicating the transition from the ancestral to the descendant node; when the character state at a particular node is ambiguous, a slash (I) is used to separate the alternative possi bilities.

Diqnodontia

1 [0+0/1]* Premaxillae fused. This is characteristic of al1 dicynodonts other than

Eodiqnodon, although the primitive condition (i.e. paired premaxillae) persists in more basal anomodonts. However, judging from Diictodon, even forms with supposedly fused premaxillae may retain a nmow partial median suture near the posterior end of the dorsal exposure of the premaxilla. As with many character transitions in this region of the cladogram, ambiguity arises because the apparently basal Endothiodon shows the derived condition in this character (fused premaxillae) while the apparently more derived

Eodicynodon retins the primitive condition. The character may therefore either reverse in Eodicynodon or advance convergently in Endothiodon and in higher forms.

10[0/1+L]* Septomaxilla recessed within extemal naris. This character could not be coded for some genera, as the septomaxilla is loosely attached in many dicynodonts and tends to be separated easily frorn the skull. In Sicminia and most other basal anomodonts the septomaxilla is incorporated into the laterai surface of the snout, having sutura1 contacts with the premaxilla, mailla and nasal (Rybczynski, 1996), but for

Patranomodon, Eodicynodon and Endothiodon no information is available and it is therefore not clear which character state is primitive for anomodonts. In most dicynodonts the septomaxilla is a thin, twisted strap-like bone resting inside the naris and contacting the surrounding bones only very loosely. Dicynodon and Lystrosaiinrs are exceptional in this regard, having extemally exposed septomaxillae like that of Suminia, but this apparentiy represents a reversal to the presumed primitive condition.

19[0+l] Temporal fenestra elongated. Primitively the temporal fenestra is no longer than the orbit, as in Patranomodon and Siminia. In dicynodonts the temporal fenestra tends to become much more elongate, probably to accommodate expanded and relatively horizontal jaw adductor muscles (King et al., 1989). The eventual reversal of this chancter in Lystrosauncs probably relates to another realignment of the musculature, necessary in order to maintain a useful gape despite the downtumed snout (King and

Cluver, 199 1).

20[0-10/1]* Postparietal enters intertemporal bar. In fatruno»iodon and Suminia the postparietal appears to be confined almost entirely to the occipital surface, whereas in many dicynodonts this presumably primitive configuration is altered and the postparietal extends forward and upward to contribute to the postenor part of the intertemporal bar.

Unfortunately, indications in the Iiterature were not clear enough to allow coding of this character for some genera (including Endothiodon; Eodicynodon exhibits the denved condition). Furthemore, at least one dicynodont (Diicrodon) shows intragenenc polyrnorphism in this character. Perhaps the position of the postparietal, a rather superficial elernent in any case, is prone to variation.

34[0/1+0/1]* Vorners hised. With uncertainty at both ancestor and descendant nodes, this character has a particularly ambiguous history. Fusion of the vomers is characteristic of al1 dicynodonts except Eodicynodon, whereas in Patranomodon the vomers are still paired, a condition generally interpreted as primitive (e.g., Cluver and King, 1983).

Suminia, however, had to be coded as polymorphic for this character; of the two known specimens in which this region is exposed, one has at lest a partial suture between the vomers, but the other clearly does not (Rybczynski, 1996). A larger sample of Suminia material might clarify this character.

3 1[0+0/1]* Interpterygoid vacuity bordered anteriorly by the vomer. The primitive condition is seen in Patranomodon and Suminia, in which the pterygoids almost completely enclose the interpterygoid vacuity, meeting anterior to it in the midline. In nearly al1 dicynodonts, including Eodicynodon, the antenor pterygoid rami are known to be wideiy separated, and the antenor margin of the interpterygoid vacuity is formed by the bifurcating vomenne septum. Ambiguity results from the uncertainty in coding

Endothiodon; it almost certainly displays the denved condition, but there appean to be no positive confirmation of this in the literature.

32[0+011]* Transverse flange of pterygoid absent. The transverse flange is a primitive feature retained only in Eodicynodon among dicynodonts. King et al. (1989) suggest that reduction of the flange was necessary in order to accommodate propalinal movements of the lower jaw and rearrangement of the pterygoideus musculature. Even in Eodicynodon and non-dicynodont anomodonts the pterygoid flange is less prominent than in more basal therapsids, but in most dicynodonts it is replaced entirely by a slender anterior ramus of the pterygoid that does not project ventrally.

JZ[O-+ 11 Lateral exposure of prearticular absent. Prirnitively (in Patranomodon and

Sruninia) the prearticular is visible at the posteroventrai corner of the mandibular ramus in lateral view. but in al1 dicynodonts it is concealed entirely behind the surangular and articular.

43[0+0/112]* Maximal skull length medium (80-200mm) or large (>ZOO mm). Small size is evidently primitive for anomodonts, as in Patranomodon and Suminia the length of the skuli does not exceed 60 mm. Al1 dicynodonts included in this analysis, with the exception of Emydops, are considenbly larger, with some toms (Endoihiodon,

Oudenodon, Aiilacephalodon and some species of Dicynodon) attaining skull lengths of hundreds of millimetres. The history of this character is somewhat ambiguous because of the large variability of skull length within populations; maximal rather than average size was chosen because of the difficulty of identifying "average" adults of a group that probably displayed indeterminate growth.

Cornments: This clade is equivalent to the "higher dicynodonts" of Cluver and King

(1983) and King (1988), who used Dicynodontia in a wider sense to incorporate animals usually viewed as basal anomodonts. The most surprising feature of this clade is its weakness, as only a single additional step is needed to collapse it (despite bootstrap support of 80%). This suggests that the demarcation between dicynodonts and more basal anomodonts may be relatively weak. On the one hand, the dicynodonts Eodiqnodon and

Endothiodon each retain a set of features regarded as primitive (see below), while on the other the outgroup taxon Suminia possesses the horizontdly expanded squarnosal seen in al1 dicynodonts and the lateral dentary shelf that occurs in sorne genera, in addition to fused vomers in at least one known specimen. Furthemore, the monophyly of dicynodonts rests on oniy two unambiguous synapomorphies: the elongate temporal fenestrae and the lack of a lateral exposure of the prearticular. It seems probable that the characteristic craniai features of advanced dicynodonts appeared individually and sporadicaily, and it is probably rnisleading to assume that the emergence of dicynodonts from among basal anomodonts involved a major structural, functional or evolutionary transition.

Clade A

3[O+ 11 Postcanine teeth confined to maxilla. Patranomodon, Srtmiltia and Endodziodon retain the presumed primitive condition, having relatively complete marginal tooth rows that extend ont0 the premaxilla. In other dicynodonts that retain postcanine teeth, including Eodiqnodon, the teeth are fewer in number and occur only on the maxilla, generally in a relatively media1 position. The more derived dicynodonts, such as

Dicynodon, Aulacephalodon, Lystrosaitnîs, Diictodon and Oitdenodon, tend to be entirely edentulous apart from the possible presence of large canine tusks. This trend toward elimination of the dentition can presumably be explained in terms of increasing reliance on a homy beak for oral processing of food.

5[0-ll] Palatal rim smoothly rounded. Primitively the palatal rim is more or less occupied by the marginal tooth row. Endothiodon, Patranomodun and Suminia retain ihis condition, whereas in other dicynodonts it either forms a sharp blade-like edge or else is smoothly rounded. The rounded shape characterizes relatively basal dicynodonts such as

Eodicynodon, Prisrerodon, Kingoria, and possi bl y Emydops, w hereas highly derived genera usually have a sharp rim. 6[0+0/1/2]* Canine tusks present, at least in some individuals. This character is a famously intractable one, as sexual dimorphism in the presence of canine tusks has been suggested for many dicynodont genera (see Introduction and Intrageneric Variation, above). For the more abundant genera it is generally possible to determine whether canines are present in no specimens, some specimens or al1 specimens, but at least two pairs of widely recognized genera (Dicynodon and Oudenodon. and Arilacephaiodon ana

Pelanomudon) have been occasionally interpreted as male and female morphs because tusks are always present in one genus and always absent in the other (Cluver and King,

1983). In any event the distribution of canine tusks in dicynodonts shows no obvious phylogenetic pattern, although the uniformity of the dentition in Patranornodon and

Suminia shows their absence to be primitive. Tusks are always absent in Endothiodon, always present in Eodicynodon, and generally present in at least some specirnens in more derived dicynodonts: among advanced forms considered here, Oudenodon is the only entirely tuskless genus.

8 [O+ 11 Caniniform process present. The canini fom process projecting downward from the maxilla and sunounding the base of the tusk is among the most characteristic derived features of dicynodonts, lacking only in Endothiodon (presumably because the tooth row and lateral maxillary groove leave no room for it) and Pristerudon. In Oudenodon and

Kingoria a narrow keel of bone, the postcaninifoim crest, extends postenorly from the caniniform process.

11[hl] Nasal bosses present (and fused). In most dicynodonts the nasals form a single prominent rnedian boss, though Endothiodon and basal anomodonts lack this feature and retain the primitive smooth snout. Although Endothiodon apparently does have longitudinal ridges of some sort on its nasals (Cox, 1964). the arnbiguous nature of these structures and the dificulty of comparing them to anything else in the literature prevented their inclusion in the data matrix as an additional character or character state. In

Lystrosaurus, Oudenodon and Airlacepltalodon the nasal bosses are paired, each nad forming a small laterally placed boss over the naris.

16[O+O/ 11- Pineal boss absent. Pnmitiveiy an eievated bony boss or "chimney" surrounds the pineal foramen. The pineal boss is definitely retained in some genera, such as Endothiodon and Aulacephalodon, while in others (including Kingoria, Emydops.

Pristerodon and Dicynodon) it seems to be absent. In Oudenodon. Robertia and

Eodicynodon the indications in the literature appear equivocai. In any case this structure is variably present in Diictodon, and it would not be surprising to find sirnilar variability in other genera as well.

20[0/1-t LI* Postparietal enters intertemporal bar (see under Dic ynodontia, above).

3 1[01b 1]* Interpterygoid vacuity bordered anteriorly by the vomer (see under

Dicynodontia, above).

40[0/2+2]* Laterd dentary shelf prominent and well developed. Suminin, Eodicynodon, and many advanced dicynodonts share a large horizontal flange of bone projecting from the dorsal border of the mandibular fenestra, though this structure is lacking in

Endothiodon and Patranomodon and its absence may be primitive. In Diictodon,

Robertia, and Oudenodon the mandibular fenestra is bounded dorsally by a low and rounded eminence, which may be homologous to the more prominent shelf seen in other genera. Latimer et al. (1995: pp. 75-76) stated that "a bulbous swelling of the dentary is present on the ventrolateral side of the jaw .. . antenor to the intramandibular fenestra [=mandibular fenestra]" of Endothiodon, but the position of this presumably autapomorphic structure shows that it is not equivalent to the lateral dentary shelf.

43[0/1/2+1]* Skull length medium (see under Dicynodontia, above).

44[0+lj Length of snout divided by skull length between 0.2 and 0.3. Patranomodon,

Suminia and Endothiodon are al1 compamtively long-snouted, with ratios in excess of 0.3, and this clearly represents the primitive condition. Aulacephalodon, with its charactenstically short and blunt skull. is the only genus exhibiting a value lower than

0.2; al1 other dicynodonts considered here fall between these extremes. As with most peculiarities of the dicynodont skull, the relatively short snout may relate to the unusuai feeding apparatus. King et al. (1989) argued that the expanded jaw musculature began to encroach on the orbital region at an early stage in dicynodont evolution, forcing the orbits fonvard and resulting in shortening of the antorbital region.

Comments: The numerous primitive features of Eodiqnodon Ied Cluver and King (1983: p. 198) to descnbe it as "unquestionably the most primitive South Afncan dicynodont known". and other authors (King, 1988; Rubidge, 1990; Cox, 1998) have generally agreed with this assessment. It is therefore surprising that Endothiodon emerges as the most basal dicynodont in the present phylogeny, leaving Eodicynodon to form a clade dong with the remaining genera. This clade requires hvo extra steps to collapse, appears in 68% of bootstrap replicates, and is supported by five unambiguous spapornorphies, as listed above. In al1 of these derived features, absent in Endothiodon, Eodicynodon resembles higher dicynodonts. Assertions of primitiveness for Eodicynodon either ignore these charactenstics or explain their absence in Endothiodon as a secondary reversal to the primitive condition, as Cox (1998) did for the extended tooth row.

However, Eodicynodon also has a nurnber of apparently primitive features, including paired premaxillae and vomers and well developed pterygoid Ranges. In the present phylogeny these characters are ambiguous in their distribution; they either assume the denved condition at the base of Dicynodontia, requinng their reversai in Eo~icynodon,or else remain primitive in Eodicynodon but attain the denved state autapomorphically in

Endothiodon, in paralle1 with more advanced dicynodonts. Although neither of these possibilities may be intuitively satisfying, the five characters supporting a basal position for Endothiodon must be sirnilarly accounted for if one prefers to make Eodicynodon basal, and in any case this alternative is Iess parsimonious. At the very least it is apparent that a fair degree of homoplasy characterises the early stages of dicynodont evolution, and that the status of Eodicynodon as the most primitive dicynodont is suspect.

One caveat concems the fact that three of the five unambiguous synapomorphies given above are related to the tooth row, and they may not be independent. Obviously the presence of a long row of postcanine teeth requires that they occupy the palatal rim, and it rnight be argued that they also leave no room for a caniniform process. To test the effects of these correlations, the analysis was repeated without the characters descnbing the condition of the caniniform process and palatal nm. The resulting strict consensus tree contained a trichotomy at the base of Dicynodontia, the three branches leading respectively to Eodicynodon, Endothiodon and al1 other dicynodonts. This may cast doubt on the basal position indicated for Endothiodon in the original analysis, but it also fails to confirm the basal position of Eodicynodon and implies that the question is very much an open one.

Clade B

22[0+1] Anterior palatal ridges present and paired. Al1 members of this clade have a pair of well defined ndges on the anterior part of the secondq pdiîtr, sornetimes extzniling posteriorly to contact the single posterior ridge. In primitive foms the antenor ridges do not occur, although the postenor ndge appears almost universal in dicynodonts.

However, the condition of the secondary bony palate in Eodkynodon is unusual: the posterior ndge widens as it extends anteriorly, fomiing a large elevated area that reaches the antenor rim of the palate. Depressed areas lie posterolateral to this "broad fan-shaped ridge" (Rubidge, 1990: p. 3).

37IO-t 11 Dentary tables present. In most dicynodonts, with the exception of KNlgoria

Emydops. and Endu~hiodon,the dentary fons large and slightiy concave expansions just posterior to the symphyseal area. These structures, generaily referred to as dentary tables, would have worked against the recessed areas of the palate laterai to the posterior median ndge on the premaxilla when processing food (Cox. 1998). The lack of dentary tables in

Emydops and Kingoria implies that their absence is primitive and that Eodicynodon developed its dentary tables in parallel with higher foms.

45[0+0/1]* Minimum interorbital width less than 20% of skull length. The frontals, which form the bulk of the media1 orbital margins, are relatively nanow in some rnernbers of this clade, including Pristerudon, Diictodon and Oudenodon. In Dicynodon and Aulacephalodon, however, they retain or retum to the greater relative width characteristic of Emydops and Kingofia (i.e. between 20% and 30% of skull length) whereas in the short-skulled Lystrosauncs their width cornes to exceed 30% of the length of the skull. In more basal forms the interorbital width may be narrow (Suminia), very broad (Endothiodon) or moderate (Eodicynodon, Patranornodon) and the primitive condition is unclear.

Comrnents: This clade is weakly supported, requiring only a single extra step to collapse and occumng in only 25% of bootstrap replicates. and is supported by only two unambiguous synapomorphies. It is similar in its membenhip to "Clade J" (or

Pnsterodontoidea) of King ( 1988), apart from the inclusion of Robenia and Diictodon.

There is no support in the present phylogeny for "Clade V". the second major assemblage postulated by King (1988), which included Robenia, Diicfodon, Emyhps and Kingoria

(along with several other taxa not appearing in this analysis). The present phylogeny sugpsts that these four genen are not closely related to one other, though Robertia and

Diictodon are sister-taxa.

Altematively, the proximity of Robenia and Diictodon to such forms as Dicynodon.

Lysîrosanrus, Aulacephalodon and Oudenodon recalls Cox's (1998) suggestion that

"robertoid" and "dicynodontoid" lineages are to be regarded as sister groups (see Fig.

19), dthough the present phylogeny does not show their precise relationship. However,

Cox (1998) also postulated a sister group relauonship between "endothiodontoids"

(including Endothiodon and Pristerodon) and b'emydopoids" (including Emydops and

Kingorin), but the present study lends no support to either the existence of these clades or their supposed phylogenetic proximity. It is possible that the resemblances in jaw and palate structure that were used as the basis for these groupings (Cox, 1998) are primarily of functional rather than phylogenetic significance, and indicate adaptation to similar diets rather than shared cornmon ancestry.

Robertiidae

9[0+3] Maxillary nm notched. The most characteristic shedleature of Robenicl ünd

Diictodon is the sharp anterior edge of the caniniform process, which lies slightly medial to the palatal nm so that the two are separated by a srnail but distinct notch. No other dicynodont displays anything like this feature, and most in fact retain the unintempted palatal rirn of primitive forms. In Ernydops, however, the media1 face of the palatal nm is embayed just in front of the caniniform process, and in Eodiqnodon the rim itself is slightly emarginated in about the same position.

21[1+0] Contact between palatine and premaxilla absent. In most denved dicynodonts, the palatine makes such a large contribution to the surface of the secondary palate that it extends fonvard to contact the premaxilla. In Diictodon and Robenia, however. as in

Lystrosauncr and in the most basal dicynodonts (i.e. Eodicynodon and Endothiodon), the two bones are separated by a medial £lange of the maxilla. Palatine-premaxilla contact also occurs in Suminia, but not in Patranomodon, and the primitive state of this character is uncertain, though absence of contact seems more probable.

38[0+0/1]* Medial dentary blade present. In Diicrodon the medial edge of the dentary table forms an elevated cuttîng blade. The only presently available description of

Robertia's cranial anatomy (Cluver and King, 1983) does not clearly indicate whether this feature is present in Robenia as well, and it may be either a synapomorphy of

Robertiidae or an autapomorphy of Diictodon. Pnrnitively it is certainly absent.

40[1/2+1]* Lateral dentary shelf weak (see under Clade A). Uncertainty arises because the cladogram is equivocai as to whether or not Oudenodon, which also has a weak and rounded shelf, is the sister-taxon of Robertiidae.

46[0+1] Minimum intertemporal width between 80% and 120% of minimum inrerorbitül width. Primitively (e.g., in Patranomodon and Suminia ) the intertemporal region is substantially wider than the interorbital (the ratio exceeding 1.2). The condition is quite variable in dicynodonts, with the intertemporal region being substantially wider in some forms and the interorbital wider in others; in Diictodon and Robenia the two widths are nearly (within 208 of the interorbital width) the sarne.

Comrnents: Although this clade is not particularly strong (one extra step needed to collapse it; bootstrap support 47%), its existence represents an important point of agreement with previous snidies (Cherand King, 1983; King, 1988; Cox, 1998), and its narne was in fact assigned by King (1988). The palatal notch is the most obvious robertiid synapomorphy, and probably indicates that Robertia and Diictodon shared a unique feeding system. However, the fact that a few postcanine teeth are present in Rokrrni, whereas Diicrodon is edentulous apart frorn the occurrence of canine tusks in some individuals, presumably implies some significant difierence in feeding mechanics. Clade C

6[0/1/2+1]* Canine tusks present (see under Clade A, above). Dicynodon,

Aulacephalodon and Lysrrosaiirus are al1 characterized by the presence of canine tusks in both sexes.

10[1+0/1]* Septomaxilla exposed on exterior of snout (see under Dicynodontia, above).

In Dicynodon and Lystrosazinis the septomaxilla lies partiy wirhin the naris but is dso integnted into the extemal surface of the snout, fonning an extensive suture with the maxilla and also contacting the nasal. In Aulucephalodon, however, it is fully recessed, as in most other dicynodonts.

20[1+0/1]* Postparietal confined to occipital surface (see under Dicynodontia. above).

In Lys~rosaiinisthe postparietal is clearly confined to the occiput, but the position of this bone in Adaceplialodon and Dicynodon is difficult to determine from the available li terature.

34[0+ 11 Transverse ndge between tubera on basioccipital present. Often refened to as the intertuberal ridge, this structure represents a highly characteristic denved feature of

Dicynodon, Aulacephalodon and Lystrosaums and does not occur in other dicynodonts.

In illustrations (e.g., Cluver and King, 1983; Cluver, 1971) it appears as a tranverse prominence on the ventral surface of the basioccipital, perhaps less well-defined in

Aidacephalodon than in the other two genera.

45[0/1+0]* Interorbital width between 20% and 30% of skull length (see under Clade B, above).

46[0+2] Minimum intertemporal width less than 80% of interorbital widih (see under

Robertiidae, above). Comrnents: Although rather weak (one extra step needed to collapse it, bootstrap support

45%~)~and an unresolved trichotomy, this clade is compatible with earlier suggestions that

Dicynodon is the closest relative of Lystrosaurus and other Triassic dicynodonts (Cluver,

197 1; King, 198 1, 1988), and that Aillacephalodon is the sister-taxon of the clade including Dicynodon and Lystrosaunis (Cluver and King, 1983; &ng, 1988 j. The intertuberd ndge is a particularly unusuai feature, apparently unique to these three taxa among forrns included in the present analysis, and has previously been recognized as a point of resemblance arnong them (Cluver and King, 1983; King, 1988).

Phylogenetic Conclusions

Although the pattern of relationships obtained in the present study is not particularly

stable, it supports some results of previous studies while drawing attention to other areas

of the cladogram where relationships appear ambiguous and more work is clearly

necessary. The most important points of agreement with earlier phylogenetic hypotheses

are probabl y the appearance of Robertia and Diictodon as sister-taxa, con firming the

significance of the highly distinctive palatal notch, and the jgouping of Dicynodon,

Lystrosaun~sand Aidacephalodon. Further analysis of Dicynodon, which is rather poorly

characterized and still includes large numbers of probably invalid species (King, 1988),

might help to clarify relationships in this part of the cladograrn. Particularly interesting in

this connection is the unusual species Dicynodon tr+gonocepluilus, viewed by King

(1981) as a probable link between more primitive species of Dicynodon on the one hand

and Lystrosaums and other Triassic forms on the other. The most interesting and surprising discrepancy between the present phylogeny and previous studies is the position of Endothiodon, nther than Eodicynodon, as the most basal dicynodont. Although support for this arrangement is equivocd even within the bounds of the present analysis, it does show that patterns of character evolution in early dicynodonts and their closest anornodont relatives were probably more complex than has been generally supposed. with si,@ficant homoplasy. As with Dicynodon, funher anatomical study of these basal forms is necessary before a robust and well-supported phylogeny can be expected to emerge.

ACKNOWLJDGEMENTS

1 have received so much help with every stage of this project that 1 sometimes wonder if 1 actually did any of it rnyself. Roben Reisz went far above and beyond the cal1 of duty in his role as my primary supervisor and academic guardian angel; from the moment he fint suggested that 1 work on Diictodon to the present mad scrarnble of completion, he has been generous with advice, useful knowledge, scientific insight, constructive criticisms, occasional financial assistance. and Timbi ts. Diane Scott provided guidance and unfail ing encouragement with technical aspects of this project, helping me build my skills in fossil preparation and illustration from the ground up. She dso prepared a good part of

specimen R 97.2. and miscellaneous bits of a few other skulls. David Dilkes applied the

proverbial fine-tooth-comb to the chapter dealing with intragenenc variation, and offered

a number of vaiuable suggestions and criticisms; corne to think of it, the chapter could

hardly have been wntten in the fitplace without David's hints on multivariate statistics

and access to his copy of the cornputer program NTSYS-PC.Ham-Dieter Sues, my CO- supervisor, did a great deal of proof-reading, shared his encyclopedic knowledge of the pertinent literature, and provided helpful advice and discussion on a variety of topics.

1 am indebted to Gene Gaffhey at the AMNH, Sean Modesto at the BPI, and Roger

Smith, Sheena Kaal, and Hedi Stumrner at the SAM for hospitality and assistance during my visits to their respective institutions; Sem deserves an additional nod for allowing me to camp out in his kitchen in Johannesburg and for bnnging me up ro speed on the ia~est trends in dicynodont anatornical terminology (1'11 never have to ask what a "sticky-outy bone" is again!), while Roger gave me a lot of help with Karoo stratigraphy. Mike Raath

(BPI), Charlotte Holton (AMNH), Bob Purdy (USNM), and Derek Ohland (SAM) ailowed me to examine and (in al1 cases but the fint) bomw Diictodon specimens in their care. David Jowen laid some of the groundwork for the phyiogenetic analysis in an undergraduate project of his. Various memben of my family, but particularly Ken,

Dorothy, and James Scott, provided unflagging mord (and occasionally logisticai) suppon. While completing this project 1 was supported mainly by funding from the

Natural Sciences and Engineering Research Council of Canada, and 1 certainly hope it hasn't gone to waste.

LITERATURE CITED

Agnew, I. D. 1959. Cranio-osteological studies in Diqnodon grimbeeki with special reference to the sphenethmoid region and cranial kinesis. Palaeontologia Africana 6: 77-

107.

Agresti, A. 1990. Categoricat data analysis. Wiley, New York, 558 pp. Allin, E. F. 1975. Evolution of the mammalian middle ear. Journal of Morphology 147:

403-408.

Barry, T. H. 1957. On the significance of tuskless specimens of Diqnodon grimbeeki

Broom. Palaeontologia Afncana 5: 57-66.

Barry, T. H. 1974. A new dicynodont mcestor from the Upper Ecca (Lower Middie

Pennian) of South Afnca. Annals of the South Afncan Museum 64: 117-136.

Berry, J. F. and Shine, R. 1980. Sexual size dimorphism and sexual selection in turties

(order Testudines). Oecologia 44: 185- 19 1.

Boonstra, L. D. 1948. Anornodont reptiles from the Tupinocephalus zone of the Karroo system. In du Toit, A. L. (ed), Special publications of the Royal Sociery of South Africa.

Robert Broom cornmernorative volume. Royal Society of South Afnca, Cape Town, pp.

57-64.

Bremer, K. 1988. The limits of amino acid sequence data in Angiosperm phylogenetic

reconstruction. Evolution 42: 795-803. Broili, F. and Schroder, J. 1937. Beobachtungen an Wirbeltieren der Karrooformation.

XXW. Üûer einige neue Anomodontier aus der Tapinocephalus-Zone. Sitzungsbenchte der Bayenschen Akademie der Wissenschaften zu München 1937: 118- 168.

Broom, R. 1905a. On the use of the term Anomodontia. Records of the Albany Museum

1 : 266-269.

Broom, R. 1905b. Notice of some new fossil reptiles from the Karroo beds of South

Afnca. Records of the Albany Museum 1: 33 1-337.

Broom, R. 1912. On some fossil reptiles from the Permian and Triassic beds of South

Africa. Proceedings of the ZooIogicd Society of London 1912: 859-876.

Broom, R. 1913. On some new genera and species of dicynodont reptiles, with notes on a

few others. Bulletin of the Amencan Museum of Naturd History 32: 441-457.

Broom, R. 1921. On some new genera and species of anornodont reptiles from the Karroo

beds of South Africa. Proceedings of the Zoological Society of London 1921: 647-674.

Broom, R. 1932. The rnammal-like reptiles of South Aficn and the origin of mummols.

Witherby, London, 376 pp. Broom, R. 1935. A new genus and some new species of mammal-like reptiles. Annals of the Transvaal Museum 18: 1-13.

Broom, R. 1936. On some new genera and species of Karroo fossil reptiles, with notes on some others. Annals of the Transvaai Museum 18: 349-386.

Broom, R. 1940. On some new genera and species of fossil reptiles from the Karroo beds of Graaff-Reinet. Annals of the Transvaal Museum 20: 157-193.

Broom, R. and Haughton, S. H. 1913. Two new species of Dicynodon. Annals of the

South African Museum 12: 36-39.

Broom. R. and Haughton, S. H. 19 17. Some new species of Anomodontia (Reptilia).

Annds of the South African Museum 12: 119-125.

Carothers, J. H. 1984. Sexual selection and sexual dimorphism in some herbivorous

lizards. American Naturalist 124: 244-254.

Chinsamy, A. and Rubidge, B. S. 1993. Dicynodont (Therapsida) bone histology: phylogenetic and physiological implications. Palaeontologia Afncana 30: 97-102. CIuver, M. A. 1970. The palate and mandible in some specimens of Dicynodon testudirostris Broom & Haughton (Reptilia: Therapsida). Annals of the South African

Museum 56: 133-153.

Cluver, M. A. 197 1. The cranial morphology of the dicynodont genus Lystrosaiinis.

Annals of the South African Museum 56: 155-274.

Cluver, M. A. and Hotton, N. III. 1977. The dicynodont genus Diictodon (Reptilia,

Therapsida) and its significance. Proceedings and Papen of the 4' rUGS Gondwana

Symposium, 1977, Calcutta: 176- 183.

Cluver, M. A. and Hotton, N. III. 198 1. The genera Diqnodon and Diictodon and their bearing on the classification of the Dicynodontia. Annals of the South African Museum

83: 99- 146.

Cluver, M. A. and King, G. M. 1983. A reassessment of the relationships of Permian

Dicynodontia (Reptilia, Therapsida) and a new classification of dicynodonts. Annais of the South African Museum 83: 99-146.

Cox, C. B. 1959. On the anatomy of a new dicynodont genus with evidence of the position of the tympanum. Roceedings of the Zoolo@cai Society of London 132: 321-

367. Cox. C. B. 1964. On the palate, dentition and classification of the fossil reptile

Endothiodon and related genera. Amencan Museum Novitates 2 17 1: 1-25.

Cox, C. B. 1998. The jaw Function and adaptive radiation of the dicynodont mammal-like reptiles of the Kamo basin of South Afnca. Zoological Journal of the Linnean Society

132: 349-384.

Crompton, A. W. and Hotton, N. III. 1967. Functional morphology of the masticatory apparatus of two dicynodonts (Reptilia: Therapsida). Postilla 109: 1-5 1.

Dodson. P. 1975. Functional and ecological significance of relative growth in Alligator. journal of Zoology, London 175: 3 15-355.

Dodson, P. 1976. Quantitative aspects of relative growth and sexual dimotphism in

Protoceratops. Journal of Paleontology 50: 929-940.

Ewer. R. F. 196 1. The anatomy of the anornodont Daptocephalus leoniceps (Owen).

Proceedings of the Zoological Society of London 136: 375-402.

Felsenstein, J. 1985. Confidence limits on phylogenies: an approach using the bootstrap.

Evolution 39: 783-791. Fourie, H. 1993. A detailed description of the intemal structure of the skull of Emydops

(Therapsida: Dicynodontia). Palaeontologia Africana 30: 103-1 1 1.

Fraser, F. C. 1938. Vestigial teeth in the narwhal. Proceedings of the Linnean Society of

London 1938: 155-162.

Gibson, A. R., Baker, A. J., and Moeed, A. 1984. Morphometic variation in introduced populations of the common myna (Acridatheres trisris): an application of the jackknife to principal component analysis. Systematic Zoology 33: 408-421.

Haughton, S. H. and Brink, A. S. 1954. A bibliognphical list of Reptilia from the Karroo beds of Afnca. Palaeontologia Afncana 3: 1- 187.

Hotton, N. III. 1986. Dicynodonts and their role as pnmary consumers. In Hotton, N. DI.,

MacLean, P. D., Roth, I. J., and Roth, E. C. (eds.), nie Ecalogy and Biology of Mammal- like Reptiles. Smithsonian Institution Press, Washington, D.C.,pp. 7 1-82.

Huene, F. von 193 1. Beitrag zur Kenntnis der Fauna der südafrikanischen

Karrooformation. Geologische und palaontologische Abhandlungen 18: 159-227.

Huene, F. von 1942. Die Anomodontier des Ruhuhu-Gebiets in der Tübingen Sammlung.

Palaontographica A 94: 154- 184. Jarman, P. 1983. Mating system and sexual dimorphism in large, terrestrial, mammalian

herbivores. Biological Reviews 58: 485-520.

Jolicoeur, P. The multivariate generalization of the allometry equation. Biomevics 18:

497-499.

Kemp. T. S. 1982. Mammal-like reptiles and the origin of miunmals. Academic Press,

London, 363 pp.

Keyser, A. W. 1975. A reevaluation of the cranid morphology and systematics of some

tuskless Anornodontia. Memoirs of the Geological Survey of South Africa 67: 1-1 10.

King, G.M. 198 1. The functional anatomy of a Permian dicynodont. Philosophical

Transactions of the Royal Society Series B 29 1: 243-322.

King, G. M. 1988. Anomodontia. Encyclopedia of Paleoherpetology 17C: 1- 174.

King, G. M. 1990a. Dicynodonts and the end Permian event. Palaeontologia Africana 27:

3 1-39.

King, G. M. 1990b. nie dicynodonts: a study in paiaeobiology. Chapman and Hall,

London, 233 pp. King, G. M. 1992. The pdaeobiogeography of Pemüan anomodonts. Terra Nova 4: 633-

640.

King, G. M. 1993a. Species longevity and genenc divenity in dicpodont mammal-like reptiles. Palaeogeography, Palaeoclirnatology, Palaeoecology 102: 32 1-33?,

King, G. M. 1993b. How many species of Diictodon were there? Annals of the South

African Museum 102: 303-335.

King, G. M. and Cluver, M. A. 199 1. The aquatic Lystrosaiinis: an alternative lifestyle.

Historical Biology 4: 323-34 1.

King, G. M. and Rubidge, B. S. 1993. A taxonomie revision of srnall dicynodonts with postcanine teeth. Zoological Journal of the Linnean Society 107: 13 1-154.

King, G. M., Oelofsen, B. W., and Rubidge, B. S. 1989. The evolution of the dicpodont feeding system. Zoological Journal of the Linnean Society 96: 185-2 1 L.

Knzey, W. G. 1972. Canine teeth of the monkey, Callicebus moloch: lack of sexual dimorphism. Primates 13: 365-369.

Latimer, E. M., Gow, C. E.. and Rubidge, B. S. 1993. Dentition and feeding niche of

Endothiodon (Synapsida: Anornodonha). Pdaeontologia Africana 32: 75-82. Leutenegger, W. 1982. Scaling of sexual dimorphism in body weight and canine size in primates. Folia Primatologica 37: 163-176.

Maddison, W. P. and Maddison, D. R. 1992. MacClade: Analysis of phylogeny and character evolution. Version 3.0 Sinauer Associates, Sunderiand, Massachusetts.

Maisch, M. W. 1995. Taxonomie revision of Crytoqnodon pamhgtoni v. Huene 1942

(Therapsida, Dicpodontia) from the late Permian of Tanzania. Neues Jahrbuch fur

Geologie und Palaontologie, Monatshefte 10: 594-604.

iMarcus, L. 1990. Traditional morphometrics. In Rohlf, F. J. and Bookstein, F. L. (eds.),

Proceedings of the Michigan Morphometncs Workshop. University of Michigan Museum of Zoology, Special Publication 2, pp. 77-122.

Mead, J. G. and Payne, R. S. 1975. A specimen of the Tasman beaked whale, Tmacetzu shepherdi, hmArgentina. Journal of Marnmalogy 56: 2 13-2 18.

Mitchell, J. 1973. Determination of relative age in the dugong Dugong dugon (Müller) from a study of skuils and teeth. Zoological Journal of the Linnean Society 53: 1-23.

Momson, D. F. 1976. MtrZtivuriate statistical methods (second edition). McGraw-Hill,

New York, 415 pp. Mosteller, F., and Tukey,J. W.1977. Data analysis and regression: a second course in statistics. Addison-Wesley, Massachusetts. 5 88 pp.

Olson, E. C. 1944. The ongin of mammds based upon the cranial morphology of the therapsid suborders. Special Papers of the Geologicai Society of America 55: Li36

Olson, E. C. 1969. Sexual dimorphism in extinct arnphibians and reptiles. In

Westermann, G.E.G.,(ed.), Se-mal dimorphism in fossil metazoa and taronomic implications. International Union of Geologicai Sciences Series A, No. 1. pp. 223-225.

Osbom, H. F. 1903. On the primary division of Reptilia into two subclasses, Synapsida and Diapsida. Science 17: 275-276.

Owen, R. 1859. On the orden of fossil and recent Reptilia and their distribution in time.

Report to the British Association for the Advancement of Science 1859: 153466.

Owen, R. 1860. Palaeontology or a systematic summary of extincr animals and their geological relations. Adam and Charles Black, Edinburgh, 420 pp.

Owen, R. 1876. Descriptive and illustrated catalogue of the Fossil Reptilia of South

Afnca in the collection of the Brinsh Museum. British Museum, London. 172

Pearson, K. 1922. On the X' test of goodness of fit. Biometrika 14: 186-191.

Ralls, K. 1977. Sexual dimorphism in marnmals: avian models and unanswered questions. American Naturalist 1 11: 917-938.

Rayner, R. J. 1992. Phyllotheca: the pastures of the Late Perman. Pdaeogeograpny.

Palaeoclimatology, Palaeoecology 92: 3 1-40.

Rohlf, F. J. 1990. NTSYS-pc: numerical tawnomy and rnrd~ivariateanalysis system.

Exeter Software, Setauket, New York.

Rubidge, B. S. 1990. Redescription of the cranial rnorphology of Eodicynodon oosthuizeni (Therapsida: Dicynodontia). Navoninge van die Nasionde Museum

Bloemfontein 7: 1-25.

Rubidge, B. S. and Hopson, J. A. 1990. A primitive anornodont therapsid from the base of the Beaufort Group (Upper Permian) of South Africa. Zoological Journal of the

Linnean Society 117: 115-139.

Rubidge, B. S., Johnson, M. R., Kitching, J. W., Smith, R. M. H., Keyser, A. W., and

Groenewald, G. H. 1995. An introduction to the biozonation of the Beaufort Group. ln

Rubidge. B.S. (ed), Biostratigraphy ofrhe Beaufon Group (Karoo Supergroup). South

African Cornmittee for Stratigraphy, Biostratigraphic Series, No. l., pp 1-2. Rybczynski, N. 1996. Cranid anatomy and phylogenetic position of the basai anomodont

(Therapsida)Suminia geîmanovi. Unpublished M.Sc. Thesis, University of Toronto.

Seeley, H. G. 1898. On Oudenodon (Aidacocephalrrs)pithecops from the Dkynodun beds of East London, Cape Colony. Geoiogicai Maguine 5: 107- 1 10.

Shea, B. T. 1985. Bivariate and multivariate growth allometry: statistical and biological considerations. Journal of Zoology, London 206: 367-390.

Shine, R. 1979. Sexual selection and sexual dimorphism in the Amphibia. Copeia 1979:

297-306.

Shine, R. 1989. Ecological causes for the evolution of sexual dimorphism: a review of the evidence. Quarterly Review of Biology 64: 4 19-46 1.

Shine, R. 1994. Sexual size dimorphism in snakes revisited. Copeia 1994: 326-346.

Smith, R. M. H. 1987. Helicai burrow casts of therapsid origin from the Beaufort Group

(Permian) of South Afnca. Pdaeogeography, Palaeoclimatology, Palaeoecology 60: 155-

170. Smith, R. M. H. 1989. Fluvial facies, vertebrate taphonomy and palaeosols of the

Teekloof Formation (Permian)nea. Beaufort West, Cape Province, South Afnca.

Unpublished Ph. D. Dissertation, University of Cape Town. 230 pp.

Smith, R. M. H. 1993a. Sedimentology and ichnology of floodplain paiaeosurfaces in the

Beaufort Group (Late Peman), Karoo sequence, South Afnca. Palaios 8: 33-37.

Smith, R. M. H. 1993b. Vertebnte taphonomy of Late Permian floodplain deposits in the southwestern Karoo basin of South Africa. Palaios 8: 45-67.

Sollas, W. I. and Sollas, 1. B. J. 1913. A study of the skull of a Dicynodon by means of seriai sections. Philosophical Transactions of the Royal Society Series B 204: 201-225.

Sollas, W. J. and Sollas, 1. B. 1. 1916. On the structure of the dicynodont skull.

Philosophical Transactions of the Royal Society Series B 207: 53 1-539.

Sorenson, M.D. 1996. TreeRot. University of Michigan, Ann Arbor.

Strauss, R. E. 1987. On allometry and relative growth in evolutionary studies. Systematic

Zoology 36: 72-75.

Sun, A. L. 1973. A new species of Dicynodon from Sinkiang. Vertebrata Palasiatica Il:

52-58, Swofford, D. L. 1993. PAUP: Phylogenetic anulysis using parsimony . Version 3.1.1.

Minois Natural History Survey, Champaign, Illinois.

Thackeray, J. F., Durand, J. F., and Meyer, L. 1998. Morphometric analysis of South

African dicynodonts attributed to Lystrosarinis murruyi (Huxiey, 1859j and i deciivis

(Owen. 1860): probabilties of conspecificity. Annals of the Transvaal Museum 36: 413-

430.

Toenen, M. 1. 1953. Evolution of the palate in some anomodonts and its classificatory significance. Palaeontologia Ahicana 1: 49- L 17.

Toenen, M. J. 1954. Lystrosarînrs primitivrts sp. nov. and the origin of the genus

Lystrosaunîs. Annals and Magazine of Natural History 7: 934-938.

Tollman, S. M., Grine, F. E., and Hahn, B. D. 1980. Ontopny and sexual dimorphism in

Aulacephalodon (Reptilia, Anomodontia). Annals of the South Afncan Museum 8 1: 159-

186.

Van Hoepen, E. C. N. 1934. Oor die indeling van die Dicynodontidae na aanleiding van nuwe forme. Paleontologiese Navorsing van die Nasionaie Museum, Bloemfontein 2: 67-

10 1. Watson, D. M. S. 1948. Dicynodon and its allies. Proceedings of the Zoological Society of London 118: 823-877.

Watson, D. M. S. 1960. The anornodont skeleton. Transactions of the Zoological Society of London 29: 13 1-208.

Watson, D. M. S. and Romer, A. S. 1956. A classification of therapsids. Bulletin of the

Museum of Comparative Zoology, Harvard 114: 35-98.

Wiens, J. J. 1998. The accuracy of methods for coding and sampling higher-Ievel taxa for phylogenetic analysis: a simulation study. Systematic Biology 47: 397-413.

Wiig, 0. 1985. Morphornetric variation in the hooded seal (Cystophora cristata).Journal of Zoology, London, 206: 497-508.

Zelditch, M. L., Debry, R. W., and Straney, D. 0. 1989. Triangulation measurement schemes in the multivariate analysis of size and shape. Journal of Mamrnalogy 70: 571-

579. DATA ON INDIVIDUAL DIICTODON SPECIMENS Table Al. General information on Diicrodon specimens used in this study, including morphometric and anatomicai data, localities and stratigraphie positions of occurrence, and principal component scores (where applicable). Specimen Locality Stratigraphie Zone Notes -- AMNH 199 1 About 20 miles West of Cistecephalus (or Mumysburg Tropidostoma?) AMNH 5308 Slachters Nek Pristerognathus Type of D. galeops AMNH 5534 Beaufort West Tropidostoma Type of D. psittacops AMNH 8188 Driehoeksfontein Cistecephdus BPI 3 14 Nobelsfontein Cistecephalus BPI 316 Beaufort West Tropidostoma "PI 352 13caufûri %'ai Tnpi0os:orn;r BPI 1974 Nobelsfontein Cistecephalus R 97.1 Driehoeksfontein CistecephaIus R 97.2 Driehoeksfontein Cistecephcilus R 97.3 Dne hoeksfontein Cistecephalus Post-orbital part of skull absent SAM-PK- 2354 Dunedin Tropidostomî Type of D. resrudirosrris SAM-pK-7420 Nobelsfontein Cistecephalus Type of D. sollasi SAM-PK- 10078 Dunedin Tropidostoma Possible tusk replacement S AM-PK- 10086 Dunedin Tropidostoma S AM-PK- 1034 1 Nobe tsfontein Cistecephalus SAM-pK- 10377 Nobelsfontein Cistecephdus One unerupted tooth bud S AM-PK- 10394 Highlands Tropidostoma Skuli roof very incornplete SAM-PK- 1 1563 Kroonplaas Tapinocephalus SAM-PK-K1633 Leeu kloof Tropidostomci SAM-PK-KS 1O5 Dunedin Tropidostoma SAM-PK-KS 189 Dunedin Tropidostoma SAM-PIS-K.5204 Hoe ksplaas Cistecephalus SAM-pK-K.5990 Dunedin Tropidostoma SAM-PK-K6009 Dunedin Tropidostoma SAM-PK-K60 17 Dunedin Tropidostoma SAM-PK-K6588 Amandelboom Tropidostoma SAM-PK-K6654 Willowdene Tropidostomri SAM-PK-K6724 Ammdelboom Tropidostoma SAM-PK-K6818 Matjieskloof Trop idos toma SAM-PK-K6838 Achterplmts Cistecephalus SAM-PK-K6873 Waterval Trop or Cist. SAM-pK-K692 1 Oukloof Pass Tropidostoma SAM-PK-K6929 OukIoof Pus Tropidostoma SAM-pK-K6950 Waterval Trop. or Cist. SAM-PK-K6979 Doornp1;tilts Dicynodon SAM-PK-K7028 Leeukioof Tropidostoma 3 associated specimens SAM-pK-K738 1 Kareebos Tropidostoma S AM-PIS-K7603 Leeuriver Pristerognathus SAM-pK-K7M3 La-de-da Pristerognathus SAM-PK-K7673 Leeukioof Tropidostoma SAM-pK-K7675 Leeu kIoo f Tropidostoma SAM-PK-K7730 Leeukloof Tmpidostoma SAM-pK-K7738 Leeukloof Tropidostomri S AM-pK-K7795 Meltonwold Pris terognathus USNM 24S3 Leeukloof Tropidostoma Onfy a cast examined UT Von Huene ? ? Table Al. Cont. Qualitative skuil characters (see Table 2a for numbering and codhg) S pecimen 1 2 3 4 5 6 7 8 9 10 AMNH 199 1 AMNH 5308 AMNH 5534 AMNH 8188 BPI 314 BPI 316 BPI 352 BPI 1974 R 97.1 R 97.2 R 97.3 SAM-pK-2354 SAM-PK-7420 SAM-PK- 10078 SAM-PK- 10086 SAM-PK- 103 4 1 SAM-PK- 10377 SAM-PK-IO394 S AM-PK-11563 SAM-PK-KI633 SAM-PK-K.5 105 SAM-pK-= LN SAM-pK-ES204 SAM-PK-KS990 SAM-PK-K6009 SAM-PK-K60 L7 SAM-PK-K6588 SAM-PK-K6654 S AM-PK-K6734 SAM-PK-K68 18 SAM-PK-K6838 SAM-PK-K6873 SAM-PK-K692 1 SAM-PK-K6929 SAM-PK-K6950 SAM-PK-K6979 SAM-pK-K7028 ( 1) SAM-pK-K7028 (2) SAM-pK-K7028 (3) SAM-pK-K728 1 SAM-pK-K7603 SAM-PK-K7643 SAM-PK-K7673 SAM-pK-K7675 SAM-pK47730 SAM-pK-K7738 SAM-pK-K7795 USNM 24643 UT Von Huene 1922 180 Table Al. Cont. Morphometric measurements in cm (see TabIe 2b for numbering) Specimen 1 2 3 4 5 6 7

AMNH 5308 AMNH 5534 AMNH 8 188 BPI 314 BPI 316 BPI 352 BPI 1974 R 97.1 R 97.2 R 97.3 SAM-PK-2354 S AM-PK-7 420 SAM-PK- 10078 S AM-pK- 10086 S AM-pK- 10341 SAM-PK- 10377 SAM-pK- 10394 SAM-PK- 11563 SAM-PK-KI633 SAM-PK-K5 IO5 S AM-PK-KS 189 SAM-pK-K52W SAM-PK-K5990 SAM-PK-K6009 SAM-PK-K60 17 SAM-PK-K6588 SAM-PK-K6654 SAM-PK-K6724 SAM-PIC-K68 18 SAM-PK-K6838 SAM-PK-K6873 SAM-PK-K6921 SAM-PK-K6929 SAM-PK-K6950 SAM-PK-K6979 SAM-PKX7028 ( 1) SAM-pK-K7028 (2) SAM-PK-K7028 (3) SAM-pK-K728 1 SAM-PK-K7603 SAM-PK-K7643 SAM-PIS-K7673 SAM-pK-K7675 SAM-pK-K7730 SAM-PK-K7738 SAM-pK-K7795 USNM 24643 UT Von Huene 1922 Table Al. Cont. ------. -- Scores on~upalcornponents obtaineci in the morphometrie analysis Soecimen Score on PC 1 Score on PC 2 Score on PC 3 AMNH 1991 AMNH 5308 R 97.1 R 97.2 SAM-PK-2354 SAM-PK-7420 SAM-PK- 10078 SAM-PK- LOO86 SAM-PK- 10341 SAM-PK- 1 1563 SAM-PK-K1633 SAM-PK-KS IO5 S AM-PK-K5 189 SAM-PK-K5204 SAM-PK-K5990 SAM-PK-K6009 SAM-PK-K6734 SAM-PK-K6873 SAM-PK-K692 1 SAM-PK-K6950 SAM-PK-K7028 (2) SAM-PK-K7028 (3) SAM-PK-K728 1 SAM-PK-K7643 SAM-PK-K7673 SAM-PK-K7675 SAM-PK-K773 8 USNM 24643 APPENDE 3

DEFWITIONS OF CHARACTERS USED IN THE PHYLOGENETIC ANALYSIS Note that al1 characters are equaily weighted and unordered. A general effort has been made to designate the most primitive state for each character with a "O", but character numbering should not be read as a definite hypothesis of polarity; a priori character polarities were not assigned.

1. Premaxillae paired (0); or fused medially (1) (Cluver and King, 1983).

2. Antenor edge of palatal margin straight or weakly ernbayed (O); or deeply emarginated

to receive tip of mandible (1) (Cluver and King, 1983).

3. Postcanine teeth in upper jaw extending ont0 premaxilla (0); confined to maxilla ( 1);

or absent (2).

4. Front of snout rounded (O); or bluntly transverse (1).

5. Palatal nm occupied by tooth row (0); gently munded (1); or sharp (2).

6. Canine tusks always absent (O); variably present (1); or always present (2).

7. Intersection of palatal rim and maxilla/premaxiIla suture higher than jaw articulation

(O); or lower than jaw articulation (1).

8. Caniniforni process absent or weak (O); present (1); or present with a postenor keel (a* 9. Maxillary rim unintempted (0); embayed medially (1); notched (2); or embayed

dorsally (3) (Cluver and King, 1983).

10. Septomaxiua exposed on exterior of snout (O); or ~cessedwithin naris (1) (Keyser,

1975). 11. Nasal bosses absent (O); present and fused (1); or present and paired (2) (Keyser,

1975).

12. Suborbital bar uniformly slender (O); greatly thickened (1); or forms distinct boss (2)

(Cluver and King, 1983).

13. Boss on prefmntal absent (O); or present (1) (Keyser. 1975).

11. Postfrontals present (O); or absent ( il.

15. Preparietal absent (O); present and adjacent to the pineal forarnen (1); or present and

lying anterior to the foramen without contacting it (3).

16. Pineal boss present (O); or absent (1) (Keyser. 1975).

17. Posterior rami of postorbitals widely separated. exposing parietals medially (O); or

close together or iouching (1) (Cluver and King, 1983).

18. Posterior part of zygomatic arch nmow (0); or horizontally expanded ( 1) (Cluver and

King, 1983).

19. Temporal fenestra shorter than orbit (O); or longer than orbit (1) (Cluver and King,

1983).

20. Postparietal confined to occipital surface (O); or enters intertemporal bar (1).

21. Contact between palatine and premaxilla absent (O); or present (1) (Toerien, 1953).

23. Antexior palatal ndges absent (O); present and paired (1); or present as a single wide

expansion of the posterior ridge (2) (Cluver and King, 1983).

23. Laterally positioned groove on ventral exposure of maxilla absent (O); or present (1)

(Cluver and King, 1983).

24. Vomers paired (0); or fused (1). 25. Width of vomerïne septum constant (O); steadily increases anteriorly (1); swells near

anterior end (2); or swells at midpoint of the vomer's length (3).

26. Trough on ventral face of vornenne septum present (O); or absent (1) (Cluver and

King, 1983).

27. Medial margin of palatine straight (O); or deflected medially (1).

1%. Laterai paiatai foramen (oetween ectopierygoid and pülüiine) przsznt (O); or absent

(1).

29. Ectopterygoid large, preventing contact between pterygoid and maxilla (O); small,

allowing contact (1); or absent (2).

30. Labid fossa absent (O); or present (1).

3 1. Interpterygoid vacuity bordered antenorly by the pterygoids (O); or the vomer (1)

(Cluver and King, 1983).

32. Transverse flange of pterygoid present (O); or absent (1) (Cluver and King, 1983).

33. Ventral boss on "body" of pterygoids absent (O); or present (1) (Cluver and King,

1983).

34. Transverse ridge between basioccipital wbera on basioccipital absent (O); or present

(1) (Cluver and King. 1983).

35. Stapes solid (O); or incised with a notch (1) (Cluver and King, 1983).

36. Teeth on dentary arranged in a single row (0); arranged in multiple rows (1); or absent

(2).

37. Dentary tables absent (O); or present (1) (Cluver and King, 1983).

38. Elevated blade-like structure on medial edge of dentary absent (O); or present (1)

(Cluver and King, 1983). 39. Dentary sulcus absent (O); or present (1) (Cluver and King, 1983).

40. Lateral dentary shelf absent (O); low and rounded (1); or prominent and well

developed (2) (Cluver and King, 1983).

41. Mandibular fenestra widely open (O); occluded by the lateral dentary shelf (1); or

greatly reduced or absent (3) (Cluver and King, 1983).

42. Lateral exposure of prearticular present (O); or absent (1) (Rybczynsia, i996j.

43. Maximal skull length less than 80 mm (O); between 80 and 700 mm (1); or more ihan

200 mm (2).

44. Length of snout (nasals plus premaxilla) divided by skull length more than 0.3 (0);

between 0.3 and 0.3 (1); or less than 0.2 (2).

45. Minimum interorbital width divided by skull length between 0.2 and 0.3 (0); less than

0.2 (1); or more thm 0.3 (2).

46. Minimum intertemporal width divided by minimum interorbital width more than 1.2

(O); between 0.8 and 1.2 (1); or less than 0.8 (2).

47. Maximal width across zygomatic arches divided by skull length less than or equal to

unity (O); or more than unity (1).

48. Length of vomenne septum divided by length of interpterygoid vacuity more than 1.5

(O); between 0.5 and 1.5 (1); or less than 0.5 (2).

49. Length from occipital condyle to interpterygoid vacuity divided by skull length more

than 0.3 (0); or less than 0.3 (1). APPENDU 3

DATA MATEU,Y FOR THE PHYLOGENETIC PrNALYSTS Table A2. Data matrix for the phylogenetic analysis of dicynodont intenelationships. See Appendix 2 for character definitions.

Charocter Taxon 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Patrarrontodoti O?O '?OOOOO?OOOO 1 00000 Simr iriiu 00000000000000000 1 O O Eodicyriodori O01 O 1 2 O 1 3 :' 1 OO02?01 1 1 Eridothiotlori 1 1 O000000 1 020 11 O O 1 1 ? Pristerodorr 1 O 1 O 1 1 O00 1 1 0001 1 O 1 1 1 Robertin 1 O 1 1 220 1 2 '1 7 O O 1 1 ? O 1 1 y Diictodon 102 12 1 O 1 7 1 1 000 1 0&1 1 1 1 0&1 Dicynor/o/i 102 1 2 2 O 1 O O 1 O O O 1 1 1 1 1 ? Olïderiohri 1 O 2 1 20020 1 2 O 1 O 1 ? O 1 1 I Aulacephulotlorr 102 1 2 2 O 1 O *? 2 0&1 1 1 1 O O 1 1 1 Lystroslrrrr7rs 102 1 -3 2 1 1 002000 1 1 O 1 O O Kirigoria 1 O20 1 1 030 I I O 0 1 1 1 O 1 1 1 Er~ly~lops 101 1 '1 1 O 1 1 9 ?O001 1 O 1 1 O l-' 00 00

- - -p. ------Charactcr Taxon 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 Patrartonr otloti Stirriitr in Eotlicyn otfmr E~idotliiotlon Pri,r;erotfutt Robertia Diictodorr Dicyriodori Ouderiohri Aii/acep~iolotlort Lyst rostiii rus Kirrgorin