The morphology and systematics of Mammalodon colliveri (Cetacea: Mysticeti), a toothed mysticete from the Oligocene of Australiazoj_

Transcription

1 Zoological Journal of the Linnean Society, 2010, 158, With 48 figures The morphology and systematics of Mammalodon colliveri (Cetacea: Mysticeti), a toothed mysticete from the Oligocene of Australiazoj_ ERICH M. G. FITZGERALD 1,2,3 * 1 School of Geosciences, Monash University, Victoria 3800, Australia 2 Museum Victoria, GPO Box 666, Melbourne, Victoria 3001, Australia 3 Division of Mammals, National Museum of Natural History, Smithsonian Institution, NHB MRC 108, PO Box 37012, Washington, DC , USA Received 25 January 2009; accepted for publication 6 February 2009 Mammalodon colliveri is an unusual toothed archaic mysticete (Cetacea) from the Upper Oligocene Jan Juc Formation of south-east Australia. The morphology of the holotype skull and postcrania are described in detail. Superimposed on the generally plesiomorphic archaeocete-like morphology of Mammalodon are subtle mysticete synapomorphies. Derived features of Mammalodon include a short and bluntly rounded rostrum, reduced premaxillae, and anterodorsally directed orbits. Within Mysticeti, this suite of features is unique. The aberrant rostral morphology of Mammalodon suggests specialization for suction feeding. Janjucetus hunderi is placed in an expanded family Mammalodontidae. Phylogenetic analysis corroborates the monophyly of Basilosauridae, Neoceti, Odontoceti, and Mysticeti, and yields a novel hypothesis of toothed mysticete relationships: a basal clade of undescribed toothed mysticetes from South Carolina, a Llanocetidae + Mammalodontidae clade, and monophyletic Aetiocetidae are posited as successive sister taxa to edentulous baleen whales (Chaeomysticeti). Toothed archaic mysticetes clearly employed diverse prey capture strategies, with exaptations for filter feeding evolving sequentially in stem group Mysticeti. A stratigraphically calibrated phylogeny implies that the initial diversification of Mysticeti occurred during the Late Eocene.. doi: /j x ADDITIONAL KEYWORDS: evolution Mammalodontidae phylogeny whale. INTRODUCTION Renowned palaeocetologist Remington Kellogg (1969: 1) once remarked that discovery of additional mysticete skulls in Oligocene marine formations would furnish a much more authoritative basis for speculative opinions relative to their early geological history. At the time of Kellogg s writing, most fossil mysticetes ( baleen whales ) were Miocene and Pliocene in age, species of essentially modern morphology and inferred feeding habits. These mysticetes were edentulous, presumably had baleen, and were thought to filter feed as do extant mysticetes (Werth, 2000). In 1966, Emlong * efitzgerald@museum.vic.gov.au described Aetiocetus cotylalveus (Late Oligocene, Oregon), which although originally diagnosed as an archaeocete because of its possession of teeth, was correctly recognized by Van Valen (1968) as an archaic mysticete. The mysticete relationships of Aetiocetus were later confirmed by Barnes & Mitchell (1978), Barnes & McLeod (1984), and Barnes (1987). Since then, several Oligocene toothed archaic mysticetes have been described from Australia, Canada, Japan, New Zealand, and the USA (Fordyce, 1992, 2003a; Barnes et al., 1995; Fitzgerald, 2006), with one latest Eocene species from Antarctica (Mitchell, 1989; Fordyce, 1992, 2003a), and perhaps three species from the Early Miocene of California (Calvano et al., 2008; Staley & Barnes, 2008) (Figs 1 2; Tables 1 3). These archaic forms represent intermediate evolutionary 367

2 368 E. M. G. FITZGERALD Age (Ma) Epoch M I O C E N E O L I G O C E N E L E Stage Burdigalian E Aquitanian Chattian Rupelian EOCENE Priabonian Polarity Chron C5C C5D C5E C6 C6A C6AA C6B C6C C7 C7A C8 C9 C10 C11 C12 C13 C15 Plankt. Foram. M5 M4 P21 M3 M2 M1 P22 P20 P19 P18 b a P17 P16 Australia Victoria South Australia Jan Juc Formation Gambier Limestone Ruwarung Mbr., Port Willunga Formation New Zealand Kokoamu Greensand Ototara Limestone Amuri Limestone NW Pacific Morawan Fmn. Ashiya Group NE Pacific Vaqueros Formation Yaquina Fmn. Pysht Formation Hesquiat Fmn. Makah Formation El Cien Formation Alsea Fmn. NW Atlantic Chandler Bridge Fmn. Ashley Formation Antarctica La Meseta Formation Figure 1. Geochronology and correlation of key stratigraphical units that have produced fossils of toothed archaic mysticetes, or cetaceans that were formerly considered archaic mysticetes. Note that the vertical range of stratigraphical units represents the current best estimates of geological age maxima and minima, not discrete time-spans of deposition. The geological time scale is after Gradstein et al. (2004). Stratigraphy and geological ages of units from Applegate (1986), Fordyce (1991b, 2003a), Alley et al. (1995), Barnes et al. (1995), Barnes (1998), Prothero et al. (2001), Sanders & Barnes (2002a), Holdgate & Gallagher (2003), Cooper (2004), Fitzgerald (2004), McGowran et al. (2004), Calvano et al. (2008), and Staley & Barnes (2008). Abbreviations: Fmn., Formation; Mbr., member; NE, north-east; NW, north-west; Plankt. Foram., planktonic Foraminifera. and morphological grades between raptorial feeding using teeth (primitive) and filter feeding using baleen (derived). Despite increased knowledge, most published information on toothed mysticetes refers to one family, the diverse Aetiocetidae from the margins of the North Pacific (Emlong, 1966; Barnes et al., 1995; Fordyce & Muizon, 2001; Deméré et al., 2008). Other described toothed mysticetes in the families Llanocetidae and Mammalodontidae are more poorly known than aetiocetids. These less thoroughly documented taxa represent a more archaic grade of mysticetes than the aetiocetids. Thus, they have the potential to illuminate the origins and earliest evolutionary history of mysticetes, soon after their divergence from the Odontoceti. Despite being reported in an issue of Nature (Anonymous, 1939), Kellogg never referred to a small fossil whale discovered in Victoria, Australia, in This fossil was briefly described in 1939 by Pritchard as a new genus and species: Mammalodon colliveri. Subsequently, Ma. colliveri was included in a major bibliography (Camp, Welles & Green, 1949), and in a classification of Cetacea in Romer s (1966) Vertebrate paleontology. Originally considered an archaeocete, it was not until 1981 when R. E. Fordyce prepared the type specimen and reassessed its morphology (Fordyce, 1982a), that the mysticete affinities of Mammalodon were recognized. Fordyce (1984) published a more detailed synopsis of Ma. colliveri, but this species, which is represented by relatively complete cranial material, has never been adequately described and analysed. This article describes and discusses the cranial and postcranial morphology of the holotype specimen of Ma. colliveri, representing the basis for further interpretation and analysis. A secondary aim is to investigate the relationships of Ma. colliveri, within the broader context of testing the phylogeny and evolution of toothed archaic mysticetes through comprehensive sampling of stem mysticete taxa. Thirdly, the anatomy of Ma. colliveri will be compared with that of archaeocetes, archaic odontocetes, and several other

3 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 369 Basilosaurus isis Zygorhiza kochii Janjucetus hunderi Mammalodon colliveri Aetiocetus cotylalveus Aetiocetus weltoni Aetiocetus polydentatus Chonecetus goedertorum Figure 2. Skulls of archaeocetes and toothed mysticetes in dorsal (upper) and right lateral (lower) views. Skulls are scaled to the same length. Dashed lines represent reconstructed regions of cranial anatomy. Basilosaurus isis modified from Stromer (1908); Zygorhiza kochii modified from Kellogg (1936); Janjucetus hunderi modified from Fitzgerald (2006); Mammalodon colliveri by E. M. G. F.; Aetiocetus cotylalveus modified from Emlong (1966) and Barnes et al. (1995); Aetiocetus weltoni dorsal view modified from Deméré & Berta (2008) and lateral view by E. M. G. F.; Aetiocetus polydentatus modified from Barnes et al. (1995); Chonecetus goedertorum modified from Barnes et al. (1995). mysticete taxa. Fourth, a preliminary analysis of the form and function of the feeding apparatus of Mammalodon will be presented. Finally, potential evolutionary implications will be discussed. MATERIAL AND METHODS Anatomical nomenclature follows that of Kellogg (1936), Sisson & Grossman (1953), Fraser & Purves (1960), Kasuya (1973), Novacek (1986), Rommel (1990), Schaller (1992), Evans (1993), Fordyce (1994a, 2002a), Oishi & Hasegawa (1995), and Smith & Dodson (2003). Photographs in Figures 6, 10, 11, 12, 13, 14, 15, 16, 18, and 36A B were taken by Mr Rodney Start (Museum Victoria) using a Hasselblad medium format camera. All other photographs were taken by the author using a Nikon D70 digital SLR camera, with a 60 mm macro lens). All line drawings were prepared by the author.

4 370 E. M. G. FITZGERALD Table 1. Classification of cetacean taxa referred to in the text, with an emphasis on toothed Mysticeti Order Cetacea Brisson, 1762 Family Kekenodontidae Mitchell, 1989, incertae sedis Kekenodon onamata Hector, 1881 Suborder Archaeoceti Flower, 1883 Family Protocetidae Stromer, 1908 Georgiacetus vogtlensis Hulbert et al., 1998 Family Basilosauridae Cope, 1868 Basilosaurus isis Beadnell in Andrews, 1904 Dorudon atrox Andrews, 1906 Zygorhiza kochii Reichenbach in Carus et al., 1847 Undetermined rank Neoceti Fordyce & Muizon, 2001 Suborder Odontoceti Flower, 1865 sensu Flower, 1867 Family Simocetidae Fordyce, 2002a Simocetus rayi Fordyce, 2002a Family Waipatiidae Fordyce, 1994a Waipatia maerewhenua Fordyce, 1994a Suborder Mysticeti Gray, 1864 sensu Cope, 1869 Willungacetus aldingensis Pledge, 2005, incertae sedis Family Llanocetidae Mitchell, 1989 Llanocetus denticrenatus Mitchell, 1989 Family Mammalodontidae Mitchell, 1989 Mammalodon colliveri Pritchard, 1939 Janjucetus hunderi Fitzgerald, 2006 Family Aetiocetidae Emlong, 1966 Aetiocetus cotylalveus Emlong, 1966 Aetiocetus weltoni Barnes & Kimura in Barnes et al., 1995 Aetiocetus tomitai Kimura & Barnes in Barnes et al., 1995 Aetiocetus polydentatus Sawamura in Barnes et al., 1995 Chonecetus sookensis Russell, 1968 Chonecetus goedertorum Barnes & Furusawa in Barnes et al., 1995 Ashorocetus eguchii Barnes & Kimura in Barnes et al., 1995 Morawanocetus yabukii Kimura & Barnes in Barnes et al., 1995 Undetermined rank Chaeomysticeti Mitchell, 1989 For classification within Chaeomysticeti see Rice (1998), Fordyce & Muizon (2001), Geisler & Sanders (2003), and Steeman (2007). INSTITUTIONAL ABBREVIATIONS ALMNH, Alabama Museum of Natural History, Tuscaloosa, USA; AMP, Ashoro Museum of Paleontology, Ashoro-cho, Japan; CGM, Cairo Geological Museum, Cairo, Egypt; ChM PV, Charleston Museum, Charleston, USA; GS, New Zealand Geological Survey, New Zealand; GSM, Georgia Southern Museum, Statesboro, Georgia, USA; KMNH, Kitakyushu Museum of Natural History, Fukuoka, Japan; LACM, Natural History Museum of Los Angeles County, Los Angeles, USA; MNB, Museum für Naturkunde der Humboldt- Universität, Berlin, Germany; MUGD, University of Melbourne School of Earth Sciences, Melbourne, Australia, including the collections of the former Melbourne University Geology Department; CMNFV, Canadian Museum of Nature Vertebrate Fossil Collection, Ottawa, Canada, including the collections of the former National Museum of Canada; NMV C, Museum Victoria Comparative Anatomy Collection, Melbourne, Australia; NMV P, Museum Victoria Palaeontology Collection, Melbourne, Australia; OU, Geology Museum, Department of Geology, University of Otago, Dunedin, New Zealand; SAM P, South Australian Museum Palaeontology Collection, Adelaide, Australia; UCMP, University of California Museum of Paleontology, Berkeley, USA; UM, University of Michigan Museum of Paleontology, Ann Arbor, USA; USNM, Collections of the National Museum of Natural History, Smithsonian Institution, Washington, D.C., USA, including the collections of the former United States National Museum; ZMT, Canterbury Museum Fossil Mammals, Christchurch, New Zealand.

5 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 371 Table 2. Described toothed archaic mysticete taxa including their respective holotype specimens, age, locality and main references Species Holotype Age Locality References Aetiocetus cotylalveus Emlong, 1966 Aetiocetus tomitai Kimura & Barnes in Barnes et al., 1995 Aetiocetus weltoni Barnes & Kimura in Barnes et al., 1995 Aetiocetus polydentatus Sawamura in Barnes et al., 1995 Chonecetus sookensis Russell, 1968 Chonecetus goedertorum Barnes & Furusawa in Barnes et al., 1995 Ashorocetus eguchii Barnes & Kimura in Barnes et al., 1995 Morawanocetus yabukii Kimura & Barnes in Barnes et al., 1995 Willungacetus aldingensis Pledge, 2005 Mammalodon colliveri Pritchard, 1939 Janjucetus hunderi Fitzgerald, 2006 Llanocetus denticrenatus Mitchell, 1989 USNM 25210: skull, postcrania Early Late Oligocene Oregon (USA) Emlong (1966), Barnes et al. (1995) AMP 2: partial skull, postcrania Late Oligocene Hokkaido (Japan) Barnes et al. (1995) UCMP : skull, postcrania Early Late Oligocene Oregon (USA) Barnes et al. (1995) AMP 12: skull, postcrania Late Oligocene Hokkaido (Japan) Barnes et al. (1995) CMNFV 64443: partial skull, vertebrae Late Oligocene British Columbia (Canada) Russell (1968), Barnes et al. (1995) LACM : skull, postcrania Late Oligocene Washington (USA) Barnes et al. (1995) AMP 3: fraction of skull Late Oligocene Hokkaido (Japan) Barnes et al. (1995) AMP 1: partial skull, postcrania Late Oligocene Hokkaido (Japan) Barnes et al. (1995) SAMP40034: partial skull, vertebra NMV P17535, P199986: skull, partial postcrania NMV P216929: skull, partial postcrania USNM : skull, mandible, teeth, postcrania Early Oligocene South Australia (Australia) Pledge (2005) Late Oligocene Victoria (Australia) Pritchard (1939), Fordyce (1984), Fitzgerald (2006) Late Oligocene Victoria (Australia) Fitzgerald (2006) Late Eocene Seymour Island (Antarctica) Mitchell (1989), Fordyce (2003a, 2003b)

6 372 E. M. G. FITZGERALD Table 3. Toothed archaic mysticetes that have been recorded, but remain undescribed, or are represented by material insufficiently complete to serve as satisfactory holotypes of new taxa Taxon Specimens Age Locality References Llanocetus sp. (= proto-squalodontid of Keyes, 1973) GS 10897: teeth, partial skull Late Eocene Early Oligocene North Otago (New Zealand) Keyes (1973), Fordyce (1991b, 2003a, b)?mysticeti gen. et sp. indet. ZMT 62: partial mandible Early Oligocene North Canterbury (New Zealand)?Mammalodontidae gen. et sp. indet. NMV P48794: partial skull, vertebrae Fordyce (1989a) Late Oligocene Victoria (Australia) Fitzgerald (2005) Mysticeti gen. et sp. nov. NMV P216928: partial skull Late Oligocene Victoria (Australia) Fitzgerald (2005), Piper, Fitzgerald & Rich (2006), Uhen, Fordyce & Barnes (2008) Aetiocetidae gen. et sp. indet. KMNH VP 000,008: mandible Mysticeti gen. et sp. nov. ChM PV5720: partial skull, postcrania Mysticeti gen. et sp. nov. ChM PV2778: partial skull, postcrania Mysticeti gen. et sp. nov. ChM PV4745: partial skull, postcrania Early Late Oligocene North Kyushu (Japan) Okazaki (1987, 1995) Late Oligocene South Carolina (USA) Barnes & Sanders (1996), Geisler & Sanders (2003) Late Oligocene South Carolina (USA) Barnes & Sanders (1996), Geisler & Sanders (2003) Early Oligocene South Carolina (USA) Barnes & Sanders (1996) See text for discussion of additional undescribed toothed mysticetes.

7 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 373 SYSTEMATIC PALAEONTOLOGY CETACEA BRISSON, 1762 NEOCETI FORDYCE & MUIZON, 2001 MYSTICETI GRAY, 1864 SENSU COPE, 1869 MAMMALODONTIDAE MITCHELL, 1989 Janjucetidae Fitzgerald (2006: 2955) Type genus: Mammalodon Pritchard, Included genera: Mammalodon Pritchard, 1939 and Janjucetus Fitzgerald, Definition of family: The clade Mammalodontidae includes Mammalodon and all other genera more closely related to Mammalodon than to any other mysticete. Emended diagnosis of family: Relatively small mysticetes (condylobasal length < 500 mm). More archaic than described species of Chaeomysticeti: mammalodontids are differentiated from all these mysticetes in possessing mineralized rostral and mandibular teeth as adults. Mammalodontids are placed in the Mysticeti because they possess: a maxilla with a dorsoventrally thin lateral edge, a laterally projecting antorbital process with a steep face on its anterior edge clearly distinguishing it from the rostral part of the maxilla, and a posteriorly elongated infraorbital process developed ventral to the supraorbital process of the frontal; upper cheek teeth that lack an entocingulum; and transversely thickened basioccipital crests. Primitive archaeocete-like features of Mammalodontidae include: a relatively small skull and inferred body size, functional heterodont teeth, a short rostral portion of the maxilla (as a proportion of condylobasal length), a rostrum that, although shelved, is not dorsoventrally flattened, a posteriormost upper tooth that lies posterior to the level of the antorbital notch, bony external nares that open at a level well anterior to the antorbital notch, elongate and dorsoventrally thin nasals, an anteroposteriorly elongated optic infundibulum, parietals extensively exposed on the dorsal surface of the cranium within an elongated intertemporal constriction, a braincase that is not inflated, a supraoccipital shield with a semicircular outline, a supraoccipital that is (secondarily) not thrust forward anteriorly (i.e. no occipital component of cranial telescoping), a short anterior process of the periotic, a tympanic bulla with a distinct median furrow on its ventral surface and with a bilobed posterior edge, a posterior half of the alveolar margin of the mandible forming an angle with the ventral margin of the mandible, a large mandibular foramen and associated pan bone forming its lateral wall, and a high and elongate coronoid process of the mandible (see Fig. 2 for skull characters). Mammalodontidae are differentiated from the toothed mysticete family Llanocetidae in having the following derived features: relatively and absolutely small lower cheek teeth; posterior lower cheek teeth with two roots joined below the crown base by a transversely narrow isthmus; lower cheek teeth are closely spaced along the alveolar margin without elongated intervening diastemata; mandible has a salient lateral edge to the alveolar margin such that lower cheek teeth are implanted within an alveolar groove. Mammalodontids differ in morphology from described aetiocetid mysticetes in possessing the following derived characters: an extremely short rostral part of the maxilla (length of rostral part of maxilla is < 43% of condylobasal length); presence of a triangular wedge of frontal separating the posteromedial margin of the ascending process of maxilla from the posterolateral margin of the nasal; a linguiform preorbital process of the frontal; a low-profile braincase; a V-shaped frontoparietal suture on the dorsal surface of the cranium; a cranium with secondarily reduced occipital telescoping; a secondarily semicircular supraoccipital shield; lower teeth implanted within an alveolar groove that has a salient, ridge-like, lateral edge; cheek teeth have salient longitudinal ridges developed on both the buccal and lingual surfaces of the crown enamel; and posterior upper and lower cheek teeth with two distinct roots joined by an isthmus for part or all of their length (see Fig. 2 for skull characters). In summary, Mammalodontidae is diagnosed by the following synapomorphies: a short rostral portion of the maxilla, which is < 43% of condylobasal length minus the premaxillae; the preorbital process of the frontal has a linguiform outline with a rounded-off anterior edge; a triangular wedge of frontal separates the posteromedial edge of the ascending process of maxilla from the posterolateral margin of the nasal; viewed laterally, the dorsal edge of the braincase is low to flat, its dorsal profile at an angle of < 10 to the lateral edge of the rostrum; a V-shaped frontoparietal suture on the dorsal surface of the skull; the skull lacks cranial telescoping, with the anterior edge of the supraoccipital at a level posterior to the anterior edge of the squamosal fossa (homoplasious); the anterior edge of the supraoccipital has a semicircular outline (homoplasious); and the posterior upper cheek teeth are double rooted, with the roots being joined by an isthmus for part or all of their length. Remarks: Mitchell (1989: 2231), in a nondifferential diagnosis, characterized a monotypic family Mammalodontidae on the basis of the following characters: (1) small size; (2) long and low skull displaying little cranial telescoping ; (3) broad and flat palate; (4)

8 374 E. M. G. FITZGERALD externally convex upper tooth row; (5) no vascular grooves in palate; (6) loosely sutured rostral bones; (7) lack of a bony mandibular symphysis; (8) a heterodont but not polydont dentition; (9) 11 mandibular teeth; (10) large, closely appressed lower cheek teeth; (11) double-rooted lower cheek teeth with roots fused for most of length; (12) lower cheek tooth crowns possess multiple accessory denticles; (13) cheek tooth crown enamel with finely fluted or wrinkled ornamentation; (14) lower cheek teeth reclined posteriorly in a rake of about 20 ; (15) a mandible with a welldeveloped coronoid process; (16) an enlarged mandibular foramen and canal; and (17) a mandibular condyle projecting posteriorly from the longitudinal axis of the mandible. Subsequent to Mitchell s (1989) establishment of Mammalodontidae it has been noted that this family was not diagnosed by synapomorphies (Fordyce & Barnes, 1994; Barnes et al., 1995). Of Mitchell s (1989) 17 diagnostic characters listed above: characters 2, 5, 12, 13, 15, 16, and 17 likely represent mysticete symplesiomorphies; characters 7, 8, 9, and 11 are based on spurious interpretations of anatomy; and characters 3 and possibly 6 are synapomorphies of a more inclusive clade (i.e. Mysticeti: Fitzgerald, 2006). Character 4 is an autapomorphy of Ma. colliveri. Character 1, small size, is difficult to assess. It is unclear whether Mitchell (1989) was referring to relatively small skull size or inferred small body size, compared to that of extant Mysticeti. It remains possible that small size is a synapomorphy of Mammalodontidae. Character 10 is inadequately defined, lacking a quantitative description of relative tooth size and spacing between lower cheek teeth. Character 14 is a potential synapomorphy of Mammalodontidae, as both Ma. colliveri and Janjucetus hunderi possess posteriorly reclined lower cheek teeth. MAMMALODON PRITCHARD, 1939 Type and only included species: Mammalodon colliveri Pritchard, Diagnosis: As for the type species Ma. colliveri, at present the only included species. Figure 3. George Baxter Pritchard and Frederick Stanley Colliver (second and third from the left, respectively) in the field at Jan Juc, date unknown, but probably between 1920 and Note the ladder used to access cliff exposures of the Jan Juc Formation. Unknown photographer. Image from the Museum Victoria archives. Remarks: In his original description of Ma. colliveri Pritchard (1939: 157) submitted the following marked features as being diagnostic of the species Mammalodon colliveri:... (1) very small crown to length of root, (2) (lower) teeth set very close together in groups with a very definite rake, (3) each molar and praemolar (sic) distinctly medially grooved indicating double fanged teeth, and (4) very large countersunk holed (sic) for the anterior teeth. (5) Jaw bones very flat and straight. Of these features: (1) is impossible to verify, because the tooth crowns in the holotype of Ma. colliveri are heavily worn; (2) is a potential mammalodontid synapomorphy; (3) (doublerooted lower cheek teeth) is a neocete plesiomorphy; (4) is presumably referring to the alveoli for the lower incisors, canine, and p1 and is thus a spurious interpretation of anatomy; and (5) is perhaps an autapomorphy of Ma. colliveri. MAMMALODON COLLIVERI PRITCHARD, 1939 Mammalodon colliveri Hills, 1958: 101. Mammalodon pritchardi Pledge & Rothausen, 1977: 286. [lapsus calami] Holotype: NMV P (formerly MUGD 1874), an incomplete skull including the right periotic, stapes and tympanic bulla, right mandible with p2 m3 in situ, loose left upper or right first lower incisor, right thyrohyoid, manubrium of sternum, axis vertebra, and rib fragments; NMV P17535, left lower cheek tooth (associated with NMV P and certainly representing the same individual). Collected by George Baxter Pritchard, Frederick Stanley Colliver, and Alan Frostick, early January 1932 (Fig. 3). Note that Mahoney & Ride (1975: 160) considered only NMV P the holotype specimen, whereas they referred to NMV P17535 as a syntype, an opinion subsequently followed by Mitchell (1989: 2231). I follow Pritchard s (1939) original referral of NMV P17535 to the holotype, which has been reiterated in subsequent publications (e.g. Fordyce, 1982a, 1984; Fitzgerald, 2006). The degree of dental wear, fusion of cranial sutures, and fusion of the posterior epiphysis to the axis vertebra, indicates that the holotype specimen of Ma. colliveri represents a mature adult individual.

9 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 375 N NEW SOUTH WALES Spring Creek 36 S N Murray River 36 S Type locality for NMV P199986, Mammalodon colliveri TORQUAY Point Danger VICTORIA AUSTRALIA MELBOURNE JAN JUC Half Moon Bay Bird Rock Jan Juc Beach N Bass Strait 40 S 40 S Bass Strait 1000 km km 144 E TASMANIA 148 E Bells Beach 5 km Figure 4. Map of Victoria, south-east Australia, showing the type locality for Mammalodon colliveri. Referred specimens: NMV P173220, Mammalodon sp. cf. Ma. colliveri, an incomplete left periotic lacking the posterior process; NMV P199587, Mammalodon sp. cf. Ma. colliveri, a partial skeleton including the left tympanic bulla, right mandible with m2 in situ, seven loose teeth, one cervical vertebra, seven thoracic vertebrae, two lumbar vertebrae, ribs, radius, and left metacarpal II. An associated ulna (NMV P198871) undoubtedly represents the same individual as NMV P Other specimens: NMV P48794,?Mammalodontidae gen. et sp. indet., an incomplete cranium with the left periotic in situ, and associated thoracic vertebrae and rib fragments; NMV P16417, Mysticeti gen. et sp. indet., an incomplete right mandible consisting of the symphysis and body of the mandible including the alveoli for i1-p3. Etymology: Mammal from the English mammal and odon from the Greek odontos meaning tooth (hence mammal tooth ), an allusion to the supposed similarity of the cheek teeth of the holotype to the postcanine teeth of terrestrial placental mammals ( It is in my opinion, a very early type of Tertiary whale, showing the closest approach to descent from a mammalian type of ancestor. ; Pritchard, 1939: 158); the species name colliveri honours Mr Frederick Stanley Colliver who discovered the holotype specimens (Pritchard, 1939). Locality, horizon, and age: The holotype specimen of Ma. colliveri was discovered...in the cliff face about 12 feet above the level of the beach...barely a hundred yards around [i.e. south-west of] the Bird Rock [a prominent rock stack] corner... (Pritchard, 1939: 151), Jan Juc Beach, south-west of Torquay in central coastal Victoria, south-east Australia (near S, E) (Fitzgerald, 2004) (Fig. 4). In his original description, Pritchard (1939: 159) did not explicitly determine the unit from which the holotype specimen was obtained, instead noting that the horizon was Jan Jukian and that the age of the rocks was Eocene. Subsequently, Singleton (1945: 284) reported that NMV P and P17535 were derived from the Lower beds, Bird Rock cliffs and... about a foot above the Spring Creek ledge [a prominent hard band] at its extreme S.W. margin (F. S. Colliver, personal communication, ) and is thus from within the Glycymeris beds. According to Singleton (1941: 39) the Glycymeris beds are sandy marls with abundant fossils of the pelecypod Glycymeris ornithopetra. These beds lie immediately below a fine-grained quartz calcarenite hard band laterally correlative with the top of the Bird Rock stack (i.e. the Bird Rock Cap; marker bed E of Raggatt & Crespin, 1955) (Fig. 5). The Bird Rock Cap horizon occurs about 1.8 m below the top of the Jan Juc Formation, the latter being conformably overlain by the Lower Miocene Puebla Formation. The matrix associated with NMV P consists of richly fossiliferous grey, sandy, glauconitic marl, consistent with the Jan Juc Formation, but inconsistent with the Puebla Formation, which is sparsely fossiliferous, grey, clayey calcareous silt with sparse glauconite (Raggatt & Crespin, 1952, 1955; Abele, 1979; Abele et al., 1988; Webb, 1995; Holdgate & Gallagher, 2003). These data confirm that the holotype of Ma. colliveri was collected from within the upper 5 m of the Jan Juc Formation, in a section bounded above by the Bird Rock Cap hard band, and below by the Spring Creek Ledge hard band. This section, which corresponds to Abele s (1979) unit BR-5, is 2 m below the top, and about 5 m above the base, of the Jan Juc Formation,

10 376 E. M. G. FITZGERALD Top of section not shown m G G G G G G G G G G G G G G G G G G clayey silt G sandy marl 23.9 Ma Bird Rock Cap Mammalodon colliveri horizon Spring Creek Ledge 25.7 Ma 27.3 Ma glauconite representing a thickness of approximately 3.30 m (Abele, 1979; Webb, 1995; Holdgate & Gallagher, 2003) (Fig. 5). The geological age of the onshore Jan Juc Formation has, until recently, been quite contentious with age determinations vacillating between Eocene (Hall & Pritchard, 1896, 1902; Raggatt & Crespin, 1952, 1955), Late Oligocene (Li, Davies & McGowran, 1999; Holdgate & Gallagher, 2003; McGowran et al., 2004), and Late Oligocene to Early Miocene (Singleton, 1941; Siesser, 1979; Darragh, 1985; Abele et al., 1988; Boreen & James, 1995; Webb, 1995; Nicolaides & Wallace, 1997). These widely varying estimates of the formation s age have inevitably been reflected in disparate citations of the geological age of Ma. colliveri: Eocene (Anonymous, 1939; Pritchard, 1939; Romer, 1966); Early Late Oligocene (Hills, 1958; McLeod, Whitmore & Barnes, 1993; McKenna & Bell, 1997); G Jan Juc Formation Puebla Formation Late Oligocene Early Miocene calcarenite hard band Figure 5. Generalized stratigraphical section at the type locality of Mammalodon colliveri, being a composite section of Bird Rock stack and the cliffs opposite and immediately west-south-west of Bird Rock. 87 Sr/ 86 Sr dates from Dickinson (2002) and Holdgate & Gallagher (2003). This section is based on Raggatt & Crespin (1955), Abele (1979), and Holdgate & Gallagher (2003). Late Oligocene (Fordyce, 1982a, 1984, 1985a, b; Fitzgerald, 2004, 2005, 2006); Late Oligocene to Early Miocene (Fordyce, 1978, 1987, 1991a, 1992, 2006; Fordyce & Barnes, 1994; Barnes et al., 1995; Domning, 1996; Milinkovitch, 1997; Fordyce & Muizon, 2001); and Early Miocene (Singleton, 1945). Li et al. (1999) have demonstrated that the Jan Juc Formation Puebla Formation contact is the local manifestation of the Oligocene Miocene boundary. They base their conclusion on the first occurrence of the planktonic foraminifer Globoquadrina dehiscens immediately above the Jan Juc Formation Puebla Formation contact, as well as other locally significant biofacies change evidence (Li et al., 1999). Globoquadrina dehiscens first occurs in southern Australian sequences at the base of the Early Miocene, indicating planktonic foraminifer zone M1 (Li et al., 1999; McGowran et al., 2004). This implies that sediments below the Jan Juc Formation Puebla Formation contact are pre-miocene in age, and more specifically, that the onshore Jan Juc Formation was deposited during planktonic foraminifer zones P21b P22 and is thus entirely Chattian in age (Li et al., 1999; Holdgate & Gallagher, 2003; Gradstein, Ogg & Smith, 2004) (Fig. 1). This biostratigraphical evidence is corroborated by 87 Sr/ 86 Sr isotope ratios from foraminifer tests in sediments directly above the Jan Juc Formation Puebla Formation contact, which yielded a date of 23.9 Mya (Kelly, Webb & Maas, 2001). Dickinson (2002) measured 87 Sr/ 86 Sr ratios from calcitic bioclasts and phosphate intraclasts and concretions from the cliff section near Bird Rock. Her study recovered dates of 23.9 Mya from 10 m up-section (just below the top of the Jan Juc Formation), 25.7 Mya from immediately below the Spring Creek Ledge horizon (about 4 m up-section), and a date of 27.3 Mya from 3 m up-section (dates also cited in Holdgate & Gallagher, 2003). These analyses show that the onshore Jan Juc Formation is entirely Chattian (Late Oligocene) in age, approximately spanning Mya (following Gradstein et al., 2004). The 87 Sr/ 86 Sr dates and their relative heights within the section suggest that Ma. colliveri was collected from a horizon that may have a minimum age of 23.9 Mya and a maximum age of 25.7 Mya (Fig. 5). In this case, Ma. colliveri has a stratigraphical distribution restricted to the late Chattian. Emended diagnosis of Mammalodon colliveri: Mammalodon colliveri is a mammalodontid mysticete distinguished from J. hunderi by the following autapomorphic characters: rostrum has a bluntly rounded apex; viewed dorsally, the rostrum has a gently convex lateral profile; alveoli for the upper incisors are coalesced; body of the premaxilla is gracile and foreshortened; premaxilla is dorsoventrally flattened; nasal

11 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 377 expands in width towards its anterior end; five dorsal infraorbital foramina in facial fossa; ascending process of the maxilla is transversely narrow and linguiform; posterior edge of the ascending process of maxilla lies in transverse line with the posterior edge of the nasal; orbit directed further anteriorly and anterodorsally; anteriormost point on the posterior edge of the supraorbital process of the frontal is laterally positioned; in cross-section, the parietals have a gently convex dorsal profile, with no salient sagittal crest along the midline of the braincase; nuchal crest projects anterodorsally and anterolaterally; alisphenoid forms most of the roof of the pterygoid sinus fossa, with the superior lamina of the pterygoid limited to the anteromedial corner of the sinus fossa; involucrum of the tympanic bulla bears a transverse groove on its dorsal surface, which divides it into a wider posterior part and a narrower anterior one; in dorsal or ventral view the mandible is straight; body of the mandible is relatively gracile; mental foramina relatively large; and three upper and four lower molars present (i.e. polydont lower dentition). MORPHOLOGICAL DESCRIPTION SKULL General features of the skull Mammalodon colliveri is represented by generally well preserved cranial material (Figs 6 24, 28 32, 35 38; measurements in Tables 4 8). Nevertheless, several elements of the Ma. colliveri holotype skull (NMV P199986) are not preserved: distal end of right postorbital process of frontal; both lacrimals and jugals; most of the palatines (a fragment of palatine is preserved in the right palatal region, but is anatomically unrevealing; Figs 8, 9); much of the lateral wall of the left side of the braincase; basicranial exposure of the vomer; mesethmoid; apex of the right zygomatic process of squamosal; entire left squamosal; majority of the pterygoids; left half of the basioccipital; ventrolateral margin of the right exoccipital; entire left exoccipital; posteroventral portion of the left half of the supraoccipital; left periotic and tympanic bulla; all middle ear ossicles (except right stapes); entire right upper dentition; and first through third left upper molars. Post-mortem distortion of the skull is generally limited to the rostrum (Figs 6, 8). Both maxillae have disarticulated from the frontals, been displaced anteriorly and dorsally, and rotated laterally. The left maxilla is more notably dislocated out of its sutural contact with the frontal than is the right maxilla. The left maxilla has been displaced anteriorly, dorsally, and medially, such that the ascending process of maxilla overlaps the left nasal, obscuring the latter from view for almost its entire length (Figs 6, 7). The right premaxilla has detached from the maxilla. In the cranium, the right periotic and tympanic bulla are disarticulated from the braincase (Fig. 21). Mammalodon colliveri is amongst the smallest described mysticetes, the condylobasal length of the adult skull being about 450 mm. This adult skull length approximates that of J. hunderi (460+ mm; Fitzgerald, 2006), but is less than that of Aetiocetus cotylalveus (630 mm; Emlong, 1966), and the smallest Recent mysticete Caperea marginata Gray, 1846 (1450 mm; Baker, 1985). There are two aspects of overall skull form that immediately draw attention: the short rostrum relative to total skull length; and the rounded, externally convex, profile of the rostrum when viewed dorsally (Figs 2, 6). In the latter distinctions, Mammalodon diverges from the skull morphology of all other known mysticetes. The foreshortened rostral bones of Ma. colliveri appear to have articulated loosely with the cranium. The orbit is relatively large and positioned high on the skull, its dorsal edge above that of the rostrum (Figs 2, 13). The braincase is elongated, with the frontals and parietals widely exposed in a distinct intertemporal constriction. In lateral view (Fig. 13), the cranium has a long and low profile. The topography of bones in the skull is generally similar to the condition seen in basilosaurid cetaceans (Kellogg, 1936): the bony external nares are located well anterior to the antorbital notch; and the apex of the supraoccipital is at a level posterior to the anterior edge of the squamosal fossa (Fig. 2). This contrasts with the so-called telescoping of the cranium in crown group and most fossil Mysticeti (Fig. 2), whereby the supraoccipital is thrust forward to override the parietal and frontal bones, thus excluding most if not all of the latter two bones from exposure on the dorsal surface of the skull (see Miller (1923) for a detailed discussion of cetacean skull telescoping). The dentition is heterodont and polydont (there is a fourth lower molar). Rostrum Compared to most Mysticeti, the rostrum is extremely short (its length representing about 30% of condylobasal length), and broad-based (Figs 2, 6, 7). The curved lateral margin of the left maxilla suggests that, in dorsal view (Figs 2, 6), the rostrum has a rounded outline. This is quite unlike all other known mysticetes, which possess a rostrum with a sharp, triangular outline (Fig. 2). In addition, the apex of the rostrum was rounded off and blunt. The premaxillae make a relatively minor contribution to the rostrum, and project 15 mm beyond the anterior edge of the maxillae. The premaxilla is splint-like, being dorsoventrally thin and transversely narrow for most of its length (Figs 6, 17). However, the apex of the premaxilla is dorsoventrally and laterally expanded for accommodation of the incisors (Fig. 17).

12 378 E. M. G. FITZGERALD Figure 6. Mammalodon colliveri, NMV P199986, dorsal view of holotype skull. Skull whitened with ammonium chloride.

13 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI mm premaxilla P2 P1 C1 position of mesorostral groove vomer palatine P3 P4 maxilla dorsal infraorbital foramina roots of P4 facial fossa ascending process of maxilla nasal antorbital notch antorbital process of maxilla preorbital process of frontal supraorbital margin supraorbital foramina frontal postorbital process of frontal orbitotemporal crest intertemporal constriction temporal fossa subtemporal crest external occipital protuberance parietal broken base of the zygomatic process of squamosal external occipital crest supraoccipital squamosal squamosal fossa dorsal intercondylar notch nuchal crest occipital condyle exoccipital Figure 7. Mammalodon colliveri, NMV P199986, dorsal view of holotype skull, based upon Figure 6, showing main anatomical features. Diagonal solid lines indicate major breaks; diagonal dashed lines indicate matrix.

14 380 E. M. G. FITZGERALD Figure 8. Mammalodon colliveri, NMV P199986, ventral view of holotype skull. Skull whitened with ammonium chloride.

15 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI mm premaxilla C1 P1 P2 vomer P3 P4 antorbital process of maxilla maxilla major palatine foramen infraorbital process of maxilla postorbital process of frontal palatine?palatal crest frontal opening of the common nasal meatus into the fundus of the nasal fossa frontal foramina postorbital ridge pterygoid orbitosphenoid presphenoid pterygoid sinus fossa cranial foramen ovale subtemporal crest sulcus for ophthalmic artery basisphenoid broken base of the zygomatic process of squamosal squamosal parietal alisphenoid external acoustic meatus periotic fossa basioccipital supraoccipital basioccipital crest sulcus for hypoglossal nerve exoccipital occipital condyle Figure 9. Mammalodon colliveri, NMV P199986, ventral view of holotype skull, based upon Figure 8, showing main anatomical features. Diagonal solid lines indicate major breaks; diagonal dashed lines indicate matrix.

16 382 E. M. G. FITZGERALD Figure 10. Mammalodon colliveri, NMV P199986, anterior view of holotype skull. Skull whitened with ammonium chloride. The dorsal presentation of the maxilla increases in breadth posteriorly towards the antorbital notch (Figs 6, 7). As a result of breakage of the maxilla, the antorbital notch is imprecisely localized. Nevertheless, in J. hunderi the antorbital notch is located at a level slightly anterior to the anterolateral edge of the antorbital process of the maxilla (Fig. 2). Thus the antorbital notch of Ma. colliveri is inferred to be within millimetres of this point (Fig. 7). The rostrum is widest at the antorbital notch. Anteromedial to the antorbital notch, the facial region of the maxilla is dominated by the multiple dorsal infraorbital foramina and their associated sulci (Figs 6, 7, 11, 12, 18). Viewed dorsally (Figs 6, 7), the posterior wall of the antorbital notch (formed by the antorbital process of maxilla) forms a ~90 angle with the sagittal plane. In lateral view (Fig. 13), the anterior third of the rostrum appears somewhat flattened (although this is exaggerated by the disarticulation of the maxillae from the cranium). The lateral edge of the rostrum s anterior half in cross-section forms an angle of (Fig. 10), and posterior to the bony external nares becomes higher as the maxilla increases in depth. The rostrum has a dorsoventrally thin lateral edge. Owing to the dislocation of the maxillae it is difficult to judge the shape of the ventral profile of the rostrum, in lateral view (Fig. 13). However, it does not appear to have been markedly arched dorsally or convex, and was probably relatively straight. The transverse profile of the palate is flat to slightly concave (Fig. 10). The mesorostral groove forms the median part of the rostrum, anterior to the nasals, and located medial to both premaxillae and maxillae (Figs 6, 7). The mesorostral groove was probably open anteriorly as seen in J. hunderi. The olfactory cavity is elongated and roofed by the frontals and nasals. The bony external nares open at a level approximately 35 mm anterior to the antorbital notch. Cranium The cranium of Ma. colliveri is elongate, comprising about 70% of the condylobasal length. In lateral view (Fig. 13), the dorsal profile of the cranium ascends posteriorly from the horizontal frontal shield, rising gradually to the anterior edge of the supraoccipital. Although most of the ventral surface of the cranium is corroded or missing (Figs 8, 9), it is likely that in lateral view (Fig. 13), the ventral profile of the cranium was straight. Viewed laterally (Fig. 13), the orbit opens anterolaterally and dorsolaterally, with its dorsal edge at a level well dorsal to the lateral edge of the rostral base (at the antorbital notch). The ascending process of maxilla extends posteriorly to a point level with the mid-point (anteroposteriorly) of the orbit. In dorsal (Figs 6, 7), ventral (Figs 8, 9), and lateral view (Fig. 13), the intertemporal constriction is prominent, with the frontals and parietals extensively exposed on the lateral (Fig. 13) and dorsal (Figs 6, 7) surfaces of the braincase. The anterodorsal border of the temporal fossa is defined by a well-developed orbitotemporal crest (Figs 7, 11, 14). Dorsally, the orbitotemporal crest overhangs the anteromedial temporal fossa, such that the lateral wall of the intertemporal constriction is pinched ventrally (Figs 6, 7). The lateral wall of the braincase (formed primarily by the parietal and squamosal) is smoothly convex with no notable inflation. The posterodorsal boundary of the temporal fossa is formed by a salient nuchal crest (formed by parietal anterolaterally and supraoccipital posteromedially) that slightly overhangs the posteromedial temporal fossa. The posteroventral border of the temporal fossa, defined by the subtemporal crest, is concave in dorsal view (Fig. 6). The supraoccipital shield has a semicircular anterior edge, primitively similar to that of archaeocetes (Kellogg, 1936) (Fig. 2), but distinct from the more triangular supraoccipital shield of aetiocetids (Fig. 2) and baleen-bearing mysticetes (Miller, 1923; Kellogg, 1928; Barnes & McLeod, 1984), although Titanocetus sammarinensis Capellini, 1901 from the middle Miocene of Italy has a semicircular supraoccipital (Bisconti, 2006). Premaxilla The premaxilla is limited to the rostrum, anterior to the antorbital notch (Figs 6, 7). Mammalodon colliveri exhibits a derived condition where there is limited sutural contact between the medial edge of the premaxilla and lateral edge of the nasal (cf. the premaxilla nasal suture in Basilosauridae, J. hunderi, and Aetiocetidae) (Fig. 2). In dorsal view, the apex of the premaxilla is slightly inflected medially (Figs 6, 17). Posterior to the anterior edge of the maxilla, the premaxilla is parallel to the sagittal plane

17 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 383 Figure 11. Mammalodon colliveri, NMV P199986, anterodorsal view of holotype skull. Skull whitened with ammonium chloride. Diagonal solid lines indicate major breaks; diagonal dashed lines indicate matrix. posteriorly to the level of P2. At this point, it is inflected medially towards the nasal, which it lies lateral to for the remainder of its length. The transverse width of the premaxilla decreases posteriorly from the apex (at the level of C1 it is at maximum width). Posterior to the tooth-bearing apex, both premaxillae are gracile splints of bone, being dorsoventrally thin (< 7 mm thick for most of the length), and in transverse cross-section having an elliptical outline (Fig. 10). Posterior to its medial inflection, the premaxilla terminates on the rostrum at a point just posterior to the anterior edge of the nasal, at about the level of P4. The posterior edge of the premaxilla is attenuated (Fig. 17). The premaxilla articulates laterally and posterolaterally with the maxilla, medially with the vomer, and posteromedially with the nasal. A large cavity in the apex of the premaxilla, partially filled by matrix, represents the rather indistinct alveolus for incisors (Figs 11, 17). There is no evidence for the presence of interalveolar septa, suggesting that the alveoli were confluent, the small incisors (see description of dentition below) sharing

18 384 E. M. G. FITZGERALD Figure 12. Mammalodon colliveri, NMV P199986, holotype skull: A, oblique anterolateral (right side); and B, oblique anterolateral (left side) views. Skull whitened with ammonium chloride. a common alveolus that is 25 mm wide and 10 mm high. In dorsal view, there is a distinct groove between the premaxilla and maxilla (Figs 6, 7). Although the right premaxilla has disarticulated from the maxilla (in part because of breakage of the premaxilla), the incisivomaxillary suture appears to be relatively tight. Towards its apex, and lateral to the alveolus for the incisors, an articular process of the premaxilla extends laterally into the anterior edge of the maxilla corpus, nearly contacting the root of the upper canine (Figs 10 12). Maxilla The maxilla forms most of the rostrum, and is the primary tooth-bearing bone in the upper jaw. In lateral view, the maxilla has a flattened profile (Fig. 13). The lateral edge of the maxilla is dorsoventrally thin, as in all described Mysticeti, but is not developed into a distinct flange as in aetiocetids and Chaeomysticeti. The facial region anteromedial to the antorbital notch is broad and notably concave, rising medially and posteriorly towards the ascending process of the maxilla. The upper tooth row is externally convex, such that the posterior cheek teeth are implanted further laterally in the edge of the maxilla than the canine and the anterior premolars (Figs 6, 7). The maxilla articulates anteriorly and anteromedially with the premaxilla, medially with the vomer, posteromedially with the nasal, posteroventrally with the palatine, and posteriorly with the frontal and lacrimal. In the maxilla there are alveoli for a canine and seven cheek teeth. By comparison with basilosaurid archaeocetes (Kellogg, 1936; Uhen, 1998, 2004), the cheek teeth are identified as P1 P4 and M1 M3. Unlike Basilosauridae, the maxilla of Ma. colliveri bears three alveoli posterior to P4 (preserved in the left maxilla of NMV P199986). Upper teeth are only preserved in the left maxilla, and these are described in further detail below. All alveoli open ventrolaterally, and alveoli for C1 P2 also open slightly anteriorly. The alveoli are separated from one another by diastemata, which increase in length posteriorly along the tooth row to P4. Posterior to P4, poor preservation means that it is unknown whether alveoli for M1 M3 were separated by diastemata. There are no embrasure pits, for insertion of the lower teeth, as possessed by Basilosauridae (see Kellogg, 1936; Uhen, 2004). The ultimate alveolus in the upper tooth row, for M3, lies about mm posterior to the level of the antorbital notch, and hence posterior to the level of the anterior edge of the orbit (Fig. 36B). Alveolar juga occur adjacent to the alveoli for P3 and P4. The alveoli for M1 and M2 are more oblate in outline than those of C1 P4, and the alveolus for M2 is markedly smaller than that of more anterior teeth. This suggests that M2 (and presumably M3) was substantially smaller than M1 as in basilosaurid archaeocetes and other basal mysticetes (e.g. Janjucetus). The ventral surface of the maxilla, or palatine process of the maxilla, is flat in transverse profile, anterior to P2, becoming more concave posteriorly towards the level of the antorbital notch. Near the medial edge of the palatine process of the maxilla, a major palatine foramen occurs at the level of P3 (Figs 8, 9, 36C). The major palatine foramen opens anteriorly into a shallow sulcus. The major palatine foramen represents the point of exit of the major palatine nerve (V) into the palatine canal. Apart from the single major palatine foramen, there are only small, scattered, foramina on the palatine process of the maxilla. A few distinct foramina occur medial to the alveoli for cheek teeth (Fig. 36C). These are interpreted as nutrient foramina, which supplied the gingival tissue. None of the latter foramina open into sulci or grooves, and are thus not homologous to the nutrient foramina for rami of the superior alveolar artery and nerve (Walmsley, 1938), such as occur in some Aetiocetidae and Chaeomysticeti (Fitzgerald, 2006; Deméré et al., 2008).

19 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 385 Figure 13. Mammalodon colliveri, NMV P199986, holotype skull: A, lateral (right side); B, lateral view (right side) of holotype skull, based upon (A), showing main anatomical features; and C, lateral (left side) views. Skull whitened with ammonium chloride. Diagonal solid lines indicate major breaks. The maxilla does not appear to have made a significant contribution to the surface of the orbital wall. However, there is a dorsoventrally thin, plate-like, infraorbital process of maxilla located ventral to the preorbital process of the frontal (Figs 8 9, 13, 19). The infraorbital process of the maxilla projects about 36 mm posterior to the anterior edge of the orbit. In Ma. colliveri the alveolus for M3 occurs at the level of the infraorbital process of the maxilla (Fig. 36A B), which was thus tooth bearing for part of its length as in basilosaurid archaeocetes (Fig. 2). A plate-like and edentulous infraorbital process of the maxilla extensively developed ventral to the frontal occurs in J. hunderi (Fitzgerald, 2006: fig. 1d), Aetiocetidae (e.g. Deméré & Berta, 2008: fig. 1B), and Chaeomysticeti. The dorsal surface of the maxilla may be divided into a rostral and cranial portion. On the dorsal surface of the rostral maxilla (Figs 6, 7), the most salient features are observed anteromedial to the antorbital notch, within the facial fossa (Fig. 18). Amongst mysticetes, Ma. colliveri is distinguished by the development and distribution of the dorsal infraorbital foramina in its maxilla. In all, there are five distinct dorsal infraorbital foramina clustered in the facial fossa on the maxilla (Figs 6, 7, 12A, 18). Four of the foramina are arrayed in a linear fashion: in dorsal view, a line drawn

20 386 E. M. G. FITZGERALD Figure 14. Mammalodon colliveri, NMV P199986, oblique posterolateral view (right side). Skull whitened with ammonium chloride. Figure 15. Mammalodon colliveri, NMV P199986, posterodorsal view of holotype skull. Skull whitened with ammonium chloride. across the apertures of the foramina forms an approximately 45 angle with the sagittal plane. A fifth (and largest) dorsal infraorbital foramen is located posterodorsal to the anterior four foramina, occupying a position near the angle between the antorbital notch and base of the ascending process of the maxilla. The anterior four foramina open anteriorly and are separated from one another by a thin septum of bone. Sulci in the surface of the maxilla run anteriorly from the apertures of four foramina. The fifth dorsal infraorbital foramen opens dorsally. Posterolateral to the dorsal infraorbital foramina, the high, laterally projecting, antorbital process of the maxilla forms the posterolateral wall of the antorbital notch, the latter of which is a cetacean apomorphy. The antorbital notch defines the base of the rostrum, and marks the path of the facial nerve where it emerges onto the facial region, providing motor innervation of nasofacial muscles (Schulte & Smith, 1918; Breathnach, 1960; Pabst, Rommel & McLellan, 1999). The anterior face of the antorbital process is steep, clearly separating this process from the rostral portion of the maxilla (Figs 6, 7, 11, 13, 18). Viewed dorsally (Figs 6, 7), the anterior edge of the antorbital process is at a right angle to the sagittal plane. In lateral view (Fig. 13), the anterior edge of the antorbital process of the maxilla forms a 45 angle with the lateral edge of the rostrum. By comparison with J. hunderi, the antorbital process probably wrapped around the anterior edge of the lacrimal at a planarcurved suture. The posterodorsal edge of the antorbital process of the maxilla declines steeply posteroventrally towards the infraorbital process (or infraorbital plate) of the maxilla. On the dorsal surface of the skull, the ascending process of the maxilla represents the cranial part of the maxilla. Dorsally (Figs 6, 7), the ascending process of the maxilla is elongate, with a transversely narrow linguiform outline. It contacts the nasal medially, and the frontal laterally and posteroventrally. Indeed, the primary suture between the maxilla and frontal seems to have been via the ascending process of the maxilla. The ascending process is not laterally expanded, and does not override the supraorbital process of the frontal. In transverse profile (Figs 11, 15, 18), the ascending process is U-shaped, with a transversely concave dorsal surface. The posterior termination of the ascending process of the maxilla is rounded-off, and in line with the posterior edge of the nasal (level with the mid-point, anteroposteriorly, of the orbit). At their posterior edges, the maxilla and nasal are separated for a short distance by a triangular wedge of frontal (Figs 6, 7). Vomer The vomer is exposed on the rostrum (dorsal view, Figs 6, 7), displaced anteriorly and dorsally. The vertical part of the vomer forming the mesorostral groove is rotated approximately 90 to the right, such that the left lateral side of the bone is visible in dorsal view (Figs 6, 7). Ventrally (Figs 8, 9), the vertical plate of the right half of the vomer is preserved posteriorly to its contact with the presphenoid. Nasal The nasal is an elongated and mediolaterally narrow bone that roofs the nasal cavity, and has a length equal to 77% of the rostral length (Figs 6, 7). The dorsal surface of the nasal is approximately planar. The nasal articulates with the premaxilla at its ante-

21 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 387 Figure 16. Mammalodon colliveri, NMV P199986, posterior view of holotype skull. Skull whitened with ammonium chloride. Diagonal solid lines indicate major breaks; diagonal dashed lines indicate matrix. rolateral extremity, and with the maxilla along most of its lateral margin. Posteriorly, the nasal contacts the frontal. The anterior edge of the nasal, at the opening of the external bony nares, has a dorsally arched inverted U-shaped cross-section (Figs 10 12). The external bony nares appear to have opened at the coronal level of P3 or at the level of the diastema between P3 and P4. Posterior from its anterior edge, the nasal narrows in width, but does not become markedly attenuated towards its termination. The lateral aspect of the nasal is planar to slightly rounded where it contacts the ventromedial edge of the ascending process of the maxilla (Fig. 11). The posteriormost 18 mm of the length of the nasal is bounded laterally and medially by a thin lamina of frontal. The posterior edge of the nasal is squared-off, and lies at a point level with the mid-point (anteroposteriorly) of the orbit. Frontal The conjoined frontals form a broad horizontal roof to the orbits and the posterior nasal cavity (Figs 6, 7, 11). The anterior border of the frontal mainly contacts the maxilla, having more limited contact with the nasal near the midline of the skull. In dorsal view (Figs 6, 7), the frontomaxillary suture is sigmoidal in shape and forms the medial and anteromedial border of the preorbital process of the frontal. Viewed dorsally (Figs 6, 7), the preorbital process has a linguiform outline and its anterior edge is rounded-off. By comparison with the skull of J. hunderi, the anterior and anterolateral border of the preorbital process was

22 388 E. M. G. FITZGERALD Figure 18. Mammalodon colliveri, NMV P199986, oblique anterolateral view of right maxilla of holotype skull. Skull whitened with ammonium chloride. Figure 17. Mammalodon colliveri, NMV P199986, right premaxilla, whitened with ammonium chloride: A, dorsal; and B, lateral views. probably formed by its contact with the lacrimal at the frontolacrimal suture. Viewed laterally, the preorbital region of the frontal is a dorsoventrally thin curved plate with no preorbital ridge (Fig. 13). Such morphology of the preorbital region is also seen in J. hunderi (Fitzgerald, 2006: figs 1B, 2A) and some aetiocetids (Fig. 2). This is strikingly different from Figure 19. Mammalodon colliveri, NMV P199986, oblique ventrolateral view of right orbit of holotype skull. Skull whitened with ammonium chloride.

23 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 389 orbital fissure 50 mm orbitotemporal crest sulcus for ophthalmic artery frontal postorbital process of frontal postorbital ridge frontal foramina ethmoidal foramina parietal orbitosphenoid maxilla palatine presphenoid vomer optic foramen basisphenoid pterygoid inferred path of maxillary branch of trigeminal nerve Figure 20. Mammalodon colliveri, NMV P199986, oblique ventrolateral view of right orbit of holotype skull, based upon Figure 19, showing main anatomical features. Diagonal solid lines indicate major breaks. the preorbital process of basilosaurid archaeocetes and crown mysticetes, which have a dorsoventrally thickened preorbital process with a preorbital ridge. The ascending process of the maxilla interlocks with the anteromedial corner of the frontal. The postorbital process of the frontal is elongated posterolaterally and slightly ventrally, as in other mysticetes, but does not appear to have been as tapered and digitiform as in Aetiocetus spp. and J. hunderi (Fig. 2). The frontal is extensively exposed on the dorsal and lateral surface of the skull in the intertemporal region, posterior to the frontal shield (Figs 6, 7, 13). The orbit is bordered dorsally and anteromedially by the preorbital process of the supraorbital process of the frontal, and dorsally and posterolaterally, by the postorbital process of the supraorbital process of the frontal (Figs 19, 20). The orbit opens anterolaterally and slightly dorsal, with the supraorbital margin of the postorbital process of the frontal forming a 116 angle with the sagittal plane (Fig. 6). Viewed laterally (Fig. 13), the orbit is arched dorsally. The dorsal edge of the orbit is at a level higher than the dorsal edge of the rostrum at its base. The highest point on the frontal is located on the postorbital process at a level slightly posterior to the anteriormost point on the supraorbital margin of the postorbital process. The external surface of the frontal slopes away from this point in an anteromedial direction towards the frontomaxillary suture and central frontal shield, and posteroventrally, posterolaterally, and posteromedially towards the orbitotemporal crest on the posterior edge of the postorbital process. The supraorbital margin of the supraorbital process of the frontal is dorsoventrally thin (~3 mm thick) for most of its length, but thickens gradually towards the preserved posterolateral edge of the postorbital process (where it is 12 mm thick). Medial to the supraorbital margin, on the external surface of the supraorbital process (Figs 6, 7), are scattered and rather indistinct supraorbital foramina (= supraorbital canals of Schaller, 1992), which represent exits for blood vessels and nerves.

24 390 E. M. G. FITZGERALD Figure 21. Mammalodon colliveri, NMV P199986, ventral view of basicranium (right side) of holotype skull. Skull whitened with ammonium chloride. The articular facet for reception of the ascending process of the maxilla is visible on the left supraorbital process of the frontal (Figs 6, 7). The shape of the articular facet mirrors the corresponding ascending process of the maxilla. The surface of the facet is complex, consisting of a series of transversely narrow ridges (and adjacent depressions), with a prominent pair of ridges in the middle of the facet. The frontal appears to have been relatively loosely sutured with the maxilla (as evidenced by the disarticulation of the maxillae from the frontals). The orbitotemporal crest is present on the posterior edge of the postorbital process of the frontal, and forms the anterodorsal margin of the temporal fossa. The orbitotemporal crest continues posteromedially onto the dorsolateral edge of the intertemporal constriction, extending onto the anterolateral surface of the parietal. Where the orbitotemporal crest occurs

25 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI mm pterygoid pterygoid sinus fossa cranial foramen ovale broken base of the zygomatic process of squamosal alisphenoid squamosal basisphenoid fossa for sigmoid process of tympanic external acoustic meatus basioccipital periotic fossa exoccipital basioccipital crest jugular notch sulcus for hypoglossal nerve occipital condyle Figure 22. Mammalodon colliveri, NMV P199986, ventral view of basicranium (right side) of holotype skull, based on Figure 21, showing main anatomical features. Diagonal solid lines indicate major breaks; diagonal dashed lines indicate matrix. on the intertemporal constriction it overhangs the temporal fossa, such that the lateral side of the intertemporal constriction appears somewhat pinched ventrally (Figs 6, 7, 13, 14). Anterior to the orbitotemporal crest on the external surface of the postorbital process of the frontal is a band of rugose bone, with a consistent thickness (anteroposteriorly) of 6 7 mm. This band of rugose bone is present along the entire preserved posterior edge of the postorbital process, and represents the anterior origin of the superficial temporalis muscle. Posteriorly, the fused frontals articulate with the parietals at a long frontoparietal (or coronal) suture in the intertemporal region (Figs 6, 7, 11). In dorsal view (Figs 6, 7), the frontoparietal suture has an open V-shape, the apex of the V-shape being directed

26 392 E. M. G. FITZGERALD Figure 23. Mammalodon colliveri, NMV P199986, right periotic, whitened with ammonium chloride: A, ventral; B, ventromedial; C, medial; and D, cerebral views. Diagonal solid lines indicate major breaks.

27 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 393 Figure 24. Mammalodon colliveri, NMV P199986, right periotic, whitened with ammonium chloride: A, lateral; B, oblique anterolateral; C, anterior; and D, posterior views. Diagonal solid lines indicate major breaks.

28 394 E. M. G. FITZGERALD Table 4. Measurements (in mm) of the holotype skull of Mammalodon colliveri, NMV P Length of skull as preserved 470 Condylobasal length, measured from tip of rostrum to posterior edge of occipital condyles 452* Length of rostrum, from tip to line across posterior limit of antorbital notch 132* Width of rostrum at base, along line across posterior limits of antorbital notches 140* Height of rostrum at base, measured relative to the lateral edge of the rostrum 33* Projection of premaxilla beyond maxilla, measured from tip of rostrum to line across foremost tip of maxilla 8 visible in dorsal view Maximum width of premaxilla exposed on dorsal surface of skull 18 Length of rostral portion of the maxilla 125 Maximum preorbital width 119 Maximum width across posterior ends of nasals 17 Maximum height of dorsal edge of orbit, measured relative to the lateral edge of rostrum 46* Supraorbital width across middle of orbits 136 Preserved diameter of orbit, from apex of preorbital process of frontal to preserved apex of postorbital process 84 Maximum width across anterior edges of nasals 24* Maximum length of right nasal 95 Distance from posterior edge of nasals to anteriormost edge of supraoccipital 180 Maximum parietal width, within temporal fossa (measured from midline of skull to lateral edge of parietal 69 Maximum length of temporal fossa, measured from anteriormost point on orbitotemporal crest to external 190 margin of nuchal crest Maximum width across zygomatic processes of squamosals 268+ Maximum distance from midline of skull to preserved lateral edge of zygomatic process 134 Vertical external height of braincase from midline of basisphenoid to summit of supraoccipital 85 Distance from midline of skull to lateral edge of occipital condyle 50 Maximum width across occipital condyles 100* Maximum height of foramen magnum 28 Length of upper left tooth-row, from hindmost margin of most posterior alveolus to tip of rostrum 149 Distance from position of posteriormost alveolus in maxilla to antorbital notch 15* *indicates estimated measurement. Measurements are the standard skull measurements used by Perrin (1975), with additional measurements more applicable to archaic cetaceans. Table 5. Measurements (in mm) of the holotype right periotic of Mammalodon colliveri, NMV P Maximum length of the periotic, measured from the apex of the anterior process to posterior edge of the pars 38.7 cochlearis Maximum transverse diameter of periotic, measured from internal edge of pars cochlearis to apex of lateral 23.5 tuberosity Anteroposterior diameter of the anterior process of periotic 18.0 Transverse diameter of the anterior process of periotic, measured at the midlength of the anterior process 14.2 Dorsoventral diameter of the anterior process of periotic, measured at the midlength of the anterior process Anteroposterior diameter of the pars cochlearis 25.0 Transverse diameter of the pars cochlearis, measured from internal edge to the fenestra ovalis 12.0 Maximum dorsoventral diameter of the pars cochlearis 25.9 Distance between aperture for the cochlear aqueduct and the fenestra rotunda 10.6 Distance between aperture for the vestibular aqueduct and the fenestra rotunda 13.3 Distance between the fenestra ovalis and the fenestra rotunda 5.4 Minimum distance between the edge of fundus of the internal acoustic meatus and the aperture for the 5.9 vestibular aqueduct Minimum distance between the edge of fundus of the internal acoustic meatus and the aperture for the 3.1 cochlear aqueduct Measurements are the standard measurements introduced by Kasuya (1973), as well as additional measurements.

29 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 395 posteriorly. The apex of the V-shape, formed by the posteriormost extent of the frontals on the dorsal surface of the skull, is located 82 mm anterior to the margin of the supraoccipital. From this point, the frontoparietal suture courses anterolaterally away from the midline of the skull and sweeps laterally where it meets the orbitotemporal crest. The suture then reverses direction, turning posteromedially alongside the orbitotemporal crest on the dorsolateral edge of the intertemporal constriction, continuing to a point approximately 12 mm posterior to the level of the apex of the V-shaped frontoparietal suture on the dorsal surface of the skull. The frontoparietal suture then turns ventrally and slightly anteriorly at this point onto the lateral surface of the intertemporal constriction, until it reaches the posterior end of the postorbital ridge, dorsolateral to the foramen for the ophthalmic artery (Figs 13, 19, 20). Anterodorsal to this region, and posteromedial to the orbit, the orbitotemporal crest on the dorsolateral edge of the intertemporal constriction projects laterally, overhanging the anteromedial extremity of the temporal fossa. The exposure of the frontal in the orbital wall is bounded posteriorly by the parietal, posteriorly and ventromedially by the orbitosphenoid, and ventrally by the palatine (Figs 19, 20). The anterior border of the orbit was formed mainly by the lacrimal (by analogy with J. hunderi). The anatomy of the orbital wall of Ma. colliveri was interpreted using Kellogg (1936), Muller (1954), and Fordyce (2002a) as guides. Anteroventral to the optic infundibulum, the frontal forms the dorsolateral border of the opening of the common nasal meatus into the fundus of the nasal fossa (Figs 8, 9). The majority of the orbital exposure of the frontal is smooth and featureless (Figs 19, 20). At a point level with the postorbital process of the frontal, three frontal foramina open in the roof of the right orbit, and four frontal foramina open in the roof of the left orbit. Posteroventral to the frontal foramina are the paired ethmoidal foramina for the ethmoidal nerve and associated blood vessels. The more posterior foramen of the pair in each orbit is larger than the anterior foramen, and opens into a posteroventrally and laterally directed sulcus. The anterior ethmoidal foramen of the pair is separated from the posterior ethmoidal foramen by a 4 5 mm thick septum of bone, and opens into a shallow posterodorsally directed sulcus. Following Kellogg (1936), the posterior ethmoidal foramen is located immediately anterior to the orbitosphenoid frontal suture. The exact contact details between the orbitosphenoid and frontal are difficult to determine in the holotype of Ma. colliveri. Thus, the description of the contact relationship between the orbitosphenoid and frontal is tentative (see description of orbitosphenoid). Nevertheless, the position of the ethmoidal foramina represents the anterior limit of the optic infundibulum (Fig. 13B). The postorbital ridge of the frontal forms the posterior border of the orbit (Figs 19, 20). This ridge is most distinct posterodorsal to the ethmoidal foramina, becoming indistinct anteriorly and posteriorly, where it contributes to the dorsolateral border of the foramen for the ophthalmic artery, and hence dorsal border of the optic infundibulum. The remainder of the optic infundibulum occurs within the orbitosphenoid and is described under that element. Parietal The parietal is exposed over a large area of the cranium, representing the majority of the external surface of the intertemporal constriction, and the anterodorsal and lateral sides of the braincase. Viewed dorsally (Figs 6, 7), the parietals are bifurcated anteriorly by the central posterior extension of the frontals onto the dorsum of the intertemporal region. The interparietal is indistinct and apparently absent. The median, or sagittal, suture between the parietals is fused, forming a low, rounded-over, ridge but not a dorsally projecting sagittal crest as in archaeocetes and some other stem Mysticeti (Fitzgerald, 2006) (Fig. 2). This median ridge becomes indistinct about 18 mm posterior to the apex of the V-shaped frontoparietal suture. Approximately 11 mm lateral to, and on either side of, the sagittal suture is a longitudinal shallow groove. These shallow depressions are interpreted as the origins for the deep temporalis muscles. No parietal or postparietal foramina are present in the external surface of the parietal. Posterolateral and ventrolateral to the grooves for the deep temporalis muscle, the parietal slopes away to form the smooth posterolateral external surface of the intertemporal constriction (Figs 6, 7, 11, 12). Here, the parietal is bordered anteriorly by the frontal, and forms the dorsal edge of the optic infundibulum. Similar to an undescribed stem mysticete (NMV P216928) and Zygorhiza kochii (in Kellogg, 1936), the parietal contacted the alisphenoid in a region immediately posteroventral to the preserved ventral extremity of the frontoparietal suture on the lateral surface of the intertemporal region. Posterior to this point, details of the sphenoparietal suture are lost, although it is likely that the alisphenoid excluded the parietal from the subtemporal crest as in J. hunderi. The squamosal suture (or parietosquamosal suture) runs posterodorsally from the region of the subtemporal crest, extending across the lateral surface of the braincase (Figs 12A, 13A B). At a point approximately 20 mm posterior to the level of the anterior edge of the supraoccipital, the squamosal suture curves

30 396 E. M. G. FITZGERALD Table 6. Measurements (in mm) of the holotype right tympanic bulla of Mammalodon colliveri, NMV P Length of the tympanic bulla, measured from the anterior tip to the posterior end of the outer posterior 57.6 prominence Length of the involucrum, measured from the anterior tip of the involucrum to the posterior end of the inner 51.0 posterior prominence Maximum distance, measured from the posteroventral edge of the outer posterior prominence to the preserved 32.5 dorsal edge of the base of the sigmoid process Width of the tympanic bulla at the level of the sigmoid process 35.0 Maximum height of the tympanic bulla, measured from the preserved edge of the base of the sigmoid process 29.6 to the ventral edge of the outer lip Width across the inner and outer posterior prominences 31.0 Measurements are the standard measurements introduced by Kasuya (1973). Table 7. Measurements (in mm) of the holotype right mandible of Mammalodon colliveri, NMV P Maximum length of mandible 381 Length of lower tooth row, measured from the hindmost margin of the posteriormost alveolus to the tip of the 210 mandible Maximum preserved height of the mandible, measured at the level of the coronoid process 138+ Maximum length of the coronoid process 115 Height of the mandible at the level of the posterior margin of the alveolus for m3 61 Height of the mandible at the level of the anterior margin of the alveolus for c1 38 Length of the mandibular foramen (including fossa posterior to it), measured from the level of the preserved 83 dorsal margin of the mandibular foramen to the medial rim of the internal surface of the condyle Measurements are from Perrin (1975). Table 8. Measurements (in mm) of the holotype dentition of Mammalodon colliveri (NMV P199986), and holotype isolated left lower premolar (NMV P17535) Maximum mesiodistal diameter of the tooth crown (NMV P17535) 14.0 Maximum buccolingual diameter of the tooth crown (NMV P17535) 8.2 Maximum mesiodistal diameter of the tooth crown (isolated incisor) 7.0 Maximum buccolingual diameter of the tooth crown (isolated incisor) 5.3 Maximum mesiodistal diameter of the tooth crown (P1) 8.1+ Maximum buccolingual diameter of the tooth crown (P1) 7.5 Maximum mesiodistal diameter of the tooth crown (P2) 10.7 Maximum buccolingual diameter of the tooth crown (P2) 8.8+ Maximum buccolingual diameter of the tooth crown (P3) 6.9+ Maximum buccolingual diameter of the tooth crown (P4) 11.8 Maximum mesiodistal diameter of the tooth crown (p2) Maximum mesiodistal diameter of the tooth crown (p3) 15.4 Maximum buccolingual diameter of the tooth crown (p3) 7.0 Maximum mesiodistal diameter of the tooth crown (p4) 14.5 Maximum buccolingual diameter of the tooth crown (p4) 8.3 Maximum mesiodistal diameter of the tooth crown (m1) 14.1 Maximum buccolingual diameter of the tooth crown (m1) 8.4 Maximum mesiodistal diameter of the tooth crown (m2) 12.3 Maximum buccolingual diameter of the tooth crown (m2) 8.3 Maximum mesiodistal diameter of the tooth crown (m3) 11.2 Maximum buccolingual diameter of the tooth crown (m3) 7.1 Note that the mesiodistal diameters of P3 and P4, and buccolingual diameter of p2, could not be measured as a result of the crowns being incomplete.

31 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 397 posteriorly and turns ventrally until it intersects with the nuchal crest. In this region the nuchal crest appears vertical when viewed laterally (Fig. 13). From this vertical orientation, the nuchal crest then turns anteriorly, with its edge directed anterodorsally, such that it overhangs the dorsomedial extremity of the temporal fossa. The nuchal crest continues anteriorly and then curves medially towards the midline of the skull. The parietal forms the anterior and anterolateral surface of the nuchal crest, whereas the posterior and posteromedial surfaces are formed by the supraoccipital. The superficial temporalis muscle originates on the lateral surface of the cranium on the parietal, with the anterior edge of the nuchal crest delimiting the origin of the posteriormost muscle fibres. Supraoccipital The supraoccipital is a broad, flat, bone that forms the posterodorsal and posteromedial external surface of the braincase (Figs 15, 16). It slopes forward at an angle of about from the horizontal. In dorsal (Figs 6, 7) and posterodorsal views (Fig. 15), the supraoccipital has a hemispherical, or semicircular, outline, with a rounded anterior edge. This contrasts with most other Mysticeti, which have a supraoccipital with a sharply triangular shape. The anteriormost point on the anterior edge of the supraoccipital is posterior to the level of the anterior edge of the squamosal fossa. The supraoccipital articulates with the parietal anteriorly, with the lambdoid suture between these bones running parallel to the nuchal crest. Ventrolaterally, the supraoccipital fuses with the exoccipital via the intraoccipital synchondrosis. Anterodorsal to the dorsal edge of the foramen magnum, the surface of the supraoccipital is planar, becoming more concave laterally in the region of the intraoccipital synchondrosis. The posterodorsal surface of the supraoccipital is dished, such that in anterodorsal view (Fig. 11) its dorsal profile is concave. Immediately posterior to the anteriormost edge of the supraoccipital is a low external occipital protuberance on the mid-line of the skull. Extending posteroventrally from the external occipital protuberance is a medially located external occipital crest that is 43 mm long. At the apex of the supraoccipital the nuchal crest projects 5 mm above the parietals located anterior to it. Exoccipital The exoccipital forms the posteroventral and posterolateral surface of the cranium and the lateral and ventral margins of the foramen magnum (Figs 6 9, 14 16). The exoccipital is fused completely with the supraoccipital along its dorsal margin. Only the right occipital condyle is preserved. It is bulbous, and projects well posterior to the vertical part of the exoccipital lateral to the condyle. Viewed posteriorly (Fig. 16), the condyle has an almost hemispherical outline and is transversely broad. In lateral view (Fig. 13), the occipital condyle has a hemispherical posterior profile. The foramen magnum appears to have had an oval outline. The dorsal limit of the articular surface of the condyle is at a point opposite its ventral limit. The dorsal intercondylar notch is broad and shallow. Anterodorsal to the condyle the exoccipital is gently concave, with no distinct dorsal condyloid fossa. Anterior to the ventral limit of the articular surface of the condyle, there is a shallow ventral condyloid fossa. Lateral to the occipital condyle, the vertical part of the exoccipital meets the squamosal along its dorsal margin via the occipitosquamosal suture. This suture is a ventrolateral continuation of the lambdoid suture between the supraoccipital and parietal. The occipitosquamosal suture continues ventrally and ventromedially, and can be traced along the posterior margin of the roof of the periotic fossa on the ventral side of the skull (Figs 21, 22), where it then turns anterodorsally and becomes indistinct. At its lateral and ventrolateral extremity, the vertical part of the exoccipital forms the paroccipital process, which contributes to the posterolateral external surface of the cranium at least to the level of the lateral wall of the squamosal fossa (Figs 15, 16). It is likely that the paroccipital process projected posterolaterally and slightly ventrally, but to a point which lay anterior to the posterior edge of the occipital condyle. In ventral view (Figs 8, 9, 21, 22), the exoccipital contacts the basioccipital and squamosal anteriorly. As preserved, the paroccipital process is exposed in the wall of the posterior lacerate foramen posterodorsal and lateral to the basioccipital crest. Immediately anteromedial to this region the exoccipital bounds the jugular notch formed between the base of the paroccipital process and the posterior edge of the basioccipital crest (Figs 21, 22). Within the angle of the jugular notch lies the hypoglossal foramen for the hypoglossal nerve (CN XII), which opens ventrolaterally. In NMV P the ventral wall of the hypoglossal canal is missing, exposing the path of the hypoglossal nerve in the exoccipital as a deep sulcus. The hypoglossal canal courses along the anterior edge of the exoccipital at the posterodorsal base of the basioccipital crest. Basioccipital Only the right half of the basioccipital is present in NMV P199986, and it is incompletely preserved (Figs 21, 22). The basioccipital is a broad unpaired bone that forms the flat cranial base, being bordered posteriorly by the exoccipital, laterally by the aperture of the posterior lacerate foramen, and anteriorly by the basisphenoid. Viewed ventrally (Figs 8, 9), the

32 398 E. M. G. FITZGERALD complete basioccipital probably had a trapezoidal outline, with its anterior side being transversely narrower than its posterior side. The basioccipital may be divided into two parts: (1) the anteromedian horizontal basilar part, which represents the posterior floor of the braincase; and (2) the ventrolaterally protruding muscular tubercle of the basioccipital, which is developed into a prominent basioccipital crest. The anterior side of the basilar part is formed by the suture with the basisphenoid, which occurs at a level immediately anterior to the anterior edge of the basioccipital crest (Figs 8, 9). The suture between the basioccipital and basisphenoid was fused via the sphenoccipital synchondrosis. Posterior to its anterior side, the basilar part is planar along the mid-line of the skull, and becomes gently concave towards the posterior lacerate foramen and basioccipital crest. At a point level with the posterior base of the falciform process of the squamosal, the basilar part turns ventrolaterally and expands to form the basioccipital crest. The basioccipital crest represents the ventralmost and most posterolateral point on the basioccipital. It is at the level of the crest that the basioccipital is broadest. In ventral view (Figs 21, 22) the basioccipital crest has an approximately rectangular outline, and its surfaces are flat to smoothly convex. The crest itself is bulbous and knob-like, being mediolaterally and dorsoventrally thickened (as in all mysticetes), relative to the condition in virtually all odontocetes where the crest is transversely thin and anteroposteriorly elongated. The crest is anteroposteriorly short, but elongated in the transverse plane, projecting ventrolaterally below the rest of the basicranium, and in ventral view (Fig. 21) overhangs the posterior margin of the posterior lacerate foramen: the latter foramen is thus obscured from view. The crest is orientated laterally in the same axis as the ventral surface of the external acoustic meatus, and the entire crest is located posterior to the level of the falciform process of squamosal. Towards its ventrolateral apex the crest narrows mediolaterally and its apex is rounded over and blunt. The ventromedial surface of the basioccipital crest is the site of insertion for neck flexor musculature (m. longus capitis) and the anterodorsal beginning of the pharyngeal constrictor muscles. Opposite the periotic fossa the dorsolateral surface of the basioccipital crest is convex, without any excavation for accommodation of a peribullary sinus, although this soft tissue feature may have been present despite the lack of an obvious osteological correlate. Anteromedially from the lateral side of the base of the crest the basioccipital forms a small component of the ventromedial wall of the posterior lacerate foramen. The posterior lacerate foramen itself is bounded by the exoccipital posteriorly, squamosal laterally, and is Figure 25.?Mammalodontidae gen. et sp. indet., NMV P48794, ventral view of left periotic and surrounding basicranium, whitened with ammonium chloride. confluent anteriorly with the foramen ovale. Posteromedially, the posterior lacerate foramen is presumably confluent with the jugular foramen. Although the confluent posterior lacerate foramen and jugular foramen of odontocetes has been referred to as the cranial hiatus by Fraser & Purves (1960), these authors stressed that mysticetes strictly lack a cranial hiatus because the periotic is not completely excluded from the cranial wall. The periotic of Mammalodon presumably retained contact with the squamosal, and thus the term cranial hiatus is not applied here. The jugular notch between the posterior edge of the basioccipital crest and the base of the paroccipital process of the exoccipital indicates the position of the jugular foramen. The posterior lacerate foramen (and confluent jugular foramen) probably transmitted the glossopharyngeal (IX), vagus (X), and accessory (XI) nerves, along with vessels of the sigmoid sinus (including the internal jugular vein: see Rommel et al., 2006). Squamosal The squamosal contributes to the external surface of the braincase, the posterior zygomatic arch, and accommodates the cranial half of the temporomandibular joint between the skull and mandible. The

33 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 399 alisphenoid inferred path of mandibular branch of trigeminal nerve position of external foramen ovale alisphenoidal infundibulum falciform process of squamosal broken base of anterior pedicle of the tympanic bulla fossa for tensor tympani mallear fossa internal acoustic meatus squamosal anterior process of periotic vascular foramen at ventral part of anteroexternal suclus mandibular fossa lateral tuberosity fossa for sigmoid process of tympanic stapes (in fenestra ovalis) fossa incudis fenestra rotunda fossa for stapedius muscle broken base of tympanohyoid postglenoid process anterior meatal crest external acoustic meatus spiny process of squamosal posterior meatal crest facial sulcus posterior bullar facet posterior process of periotic paroccipital process of exoccipital 10 mm Figure 26.?Mammalodontidae gen. et sp. indet., NMV P48794, ventral view of left periotic and surrounding basicranium, based on Figure 25, showing main anatomical features. Diagonal solid lines indicate major breaks. squamous part, or external surface, of the squamosal represents the dorsal part of the bone, with the zygomatic process of squamosal separating this dorsolateral exposure from the basicranial presentation of the squamosal. Within the temporal fossa, the squamosal is exposed over a broad area of the dorsal and lateral surface of the cranium (Figs 6, 7, 13). The posterior border of the temporal fossa is formed by the contact of the squamosal with the exoccipital at the occipitosquamosal suture. A depression in the dorsolateral external surface of the squamosal, the squamosal fossa, floors the temporal fossa and separates the external surface of the squamosal from the zygomatic process of the squamosal (Figs 6, 7). The anterior border of the squamosal fossa (and anteroventral margin of the temporal fossa as a whole) is formed by the dorsoventrally thin subtemporal crest. Posterodorsal to the subtemporal crest, the floor of the squamosal fossa is planar to shallowly concave (transversely), but narrows in width posteriorly. The lateral wall of the squamosal fossa is formed by the medial surface of the base of the zygomatic process of the squamosal. Most of the zygomatic process of the squamosal is missing, with approximately half of the base of the process preserved. The ventral, or basicranial, surface of the squamosal has lost many of its complex details through breakage and corrosion (Figs 8, 9, 21, 22; also see Figs 25, 26). Only the anteromedial surface of the mandibular fossa is preserved, with the position of the postglenoid process indicated by the preserved remnant of the external acoustic meatus and the presence of a small rectangular fossa for reception of the sigmoid process of the tympanic bulla (Figs 21, 22). The medial edge of the mandibular fossa occurs anterolateral to the fossa for the sigmoid process, and continues anteriorly as the base of the falciform process of the squamosal. There is no distinct tympanosquamosal recess anteromedial to the mandibular fossa and lateral to the base of the falciform process of squamosal. Anteromedial to the subtemporal crest and base of the falciform process is

34 400 E. M. G. FITZGERALD a region of broken cancellous bone, which presumably includes the position of the path for the mandibular branch of the trigeminal nerve (V 3) (see Figs 25, 26). The latter feature is primarily associated with the alisphenoid and is described in more detail below as part of that element. The sphenosquamosal suture between the ventral surface of the squamosal and the alisphenoid is located approximately within the path of the mandibular nerve or immediately anterior to it, about parallel with the posterior edge of the pterygoid sinus fossa. Posteromedial and dorsal to the base of the falciform process, the squamosal forms the surface of the periotic fossa, which receives the periotic excluding its posterior process (Figs 21, 22). The surface of the periotic fossa is generally smooth where it accommodated the dorsolateral edge of the periotic. The periotic in Ma. colliveri appears to have approximated the squamosal at the superior process, lateral tuberosity, and via a centrally located region on the dorsolateral surface of the anterior process. As preserved, the surface of the periotic fossa is essentially featureless, with two exceptions. Firstly, there is a shallow sulcus which courses dorsomedially and anteriorly towards the foramen ovale, and is located at the level of the base of the falciform process. This sulcus may represent the dorsolateral wall of a canal that transmitted a vascular feature, perhaps the middle meningeal artery (Fordyce, 1994a, 2002a) or the postglenoid vein (Geisler & Luo, 1998; Geisler & Sanders, 2003). The counterpart ventromedial wall of this canal is represented by a deep vascular sulcus in the dorsolateral surface of the anterior process of the periotic (i.e. the anteroexternal sulcus). The second exception is a small foramen located at the posteromedial extremity of the periotic fossa, at the level of the fossa for the sigmoid process of the tympanic bulla (Fig. 21). The homology of this small foramen is uncertain. Periotic The periotic (= petrosal or petrous portion of temporal in terrestrial mammals), via its tympanic surface, forms the roof of the middle ear cavity. Internally, the periotic houses the inner ear, which includes the organs of hearing and equilibration. As in other Neoceti (see Figs 25, 26), the periotic of Ma. colliveri (Figs 23, 24) is not firmly ankylosed to the surrounding basicranium, but is a separate element partially excluded from the wall of the braincase (although not to the extent seen in crown Odontoceti), closely approximating the squamosal and exoccipital via relatively loose articulations. This loose contact with the rest of the braincase is evidenced by the fact that the right periotic of NMV P is preserved separate from the skull. The periotics of archaic mysticetes may be divided into four parts: pars cochlearis (accommodating the cochlear duct); body (including superior process, and the pars vestibularis); anterior process (anterior extension of the body); and the posterior process (Figs 25, 26). The pars vestibularis cannot be accurately distinguished from the pars cochlearis based on external morphology alone. Nevertheless, the approximate boundary between the pars cochlearis and pars vestibularis is indicated by the position of the fenestra ovalis on the tympanic surface, with the pars cochlearis occurring medial to the level of the fenestra ovalis. In NMV P the entire posterior process and much of the dorsolateral surface of the posterior process are not present because of breakage. The periotic of Ma. colliveri presents five surfaces: tympanic (ventral surface facing the tympanic bulla) (Fig. 23A), cerebral (surface bordering the cranial cavity and posterior lacerate foramen, and including the internal acoustic meatus) (Fig. 23D), medial (surface facing the basioccipital crest) (Fig. 23C), lateral (surface facing the periotic fossa on squamosal) (Fig. 24A), and posterior (surface facing the paroccipital process of exoccipital) (Fig. 24D). The anterior process of the periotic, epitympanic recess, pars cochlearis, and fossa for the stapedius muscle can be seen in direct ventral view (Fig. 23A). The stapes is in situ, obscuring the aperture of the fenestra ovalis from view. In ventral view (Fig. 23A), the anterior process is triangular in outline, with a pointed apex. The anteroventral angle, or apex, of the anterior process is approximately at the same level as the ventral edge of the pars cochlearis. In contrast with most other mysticetes, Mammalodon has a periotic with a short anterior process, its length being less than 100% of the length of the pars cochlearis. Furthermore, the anterior process is clearly demarcated at its base from the body of the periotic, contra the condition in Chaeomysticeti (where the anterior process passes into the body of the periotic without a distinct boundary; e.g. Sanders & Barnes, 2002a, b), but similar to the condition in basilosaurids (Kellogg, 1936; Uhen, 2004) (Fig. 27), other toothed mysticetes (e.g. J. hunderi; Fitzgerald, 2006) (Figs 25, 26), and stem odontocetes (Fordyce, 2002a). With the periotic in ventral view (Fig. 23A), the angle between the medial surface of the anterior process and the anterior edge of the pars cochlearis is about 110. In anterior view (Fig. 24C) the anterior process has a pyriform outline. At mid-length, the anterior process has an ovoid cross-section, its greatest depth being at this point. This is because of an internally elongated anterodorsal angle (the vertex of which is not preserved), which forms the dorsal edge of the anterior process near its junction with the cerebral

35 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 401 Figure 27. Basilosaurus isis, cast of UM 97507, left periotic, whitened with ammonium chloride: A, ventral; B, lateral; C, medial; D, anterior; E, cerebral; and F, posterior views. surface of the periotic body. Between the anterodorsal and anteroventral angles is a sharp crest, the anterior keel, forming the anterior edge of the anterior process. In medial view (Fig. 23C), the ventral profile of the anterior process is convex and the anterior edge of the anterior process is squared-off. Viewed medially (Fig. 23C), the ventral and anterior edges of the anterior process form an angle of about 75 at the anteroventral angle. The ventral surface of the anterior process of the periotic is smoothly convex, with no evidence of an anterior pedicle for the tympanic bulla or a fovea epitubaria for reception of an independent accessory ossicle (Fig. 23A, B). A distinct anterior pedicle of the tympanic bulla (perhaps including part of the accessory ossicle) is fused to the anterior process of an isolated Mammalodon periotic (NMV P173220), immediately anterior to the mallear fossa. The medial surface of the anterior process bears no anatomical features of note, although a fossa of uncertain homology occurs in the angle between the anterior process and the anterior edge of the pars cochlearis (Fig. 23C). The lateral and ventrolateral surface of the anterior process of the periotic is traversed by the anteroexternal sulcus, trending dorsoventrally in orientation (Fig. 24A, B). The anteroexternal sulcus has a consistent width of about 3 mm along its entire length. In direct ventral view (Fig. 23A), the sulcus begins on the ventrolateral surface of the anterior process of the periotic as a longitudinal depression laterally adjacent to the midline of the anterior process. This shallow sulcus becomes more distinct posteriorly towards the base of the lateral tuberosity. In ventral view (Fig. 23A), the anteroexternal sulcus forms a notch in the lateral surface of the base of the anterior process adjacent to the lateral tuberosity. From this point the anteroexternal sulcus turns dorsolaterally onto the lateral surface of the anterior process. With the anterior process of the periotic in lateral view (Fig. 24A), the sulcus descends across the dorsolateral

36 402 E. M. G. FITZGERALD surface of the anterior process to its preserved termination immediately posterior to the level of the apex of the anterior keel. With the periotic occluding the periotic fossa on the squamosal, the anteroexternal sulcus forms the ventromedial wall of a vascular canal (the dorsolateral wall being formed by the squamosal), which opens ventrally via a foramen between the periotic and squamosal, located immediately anterior to the level of the lateral tuberosity (Figs 25, 26). The anteroexternal sulcus on the periotic (and the canal between the periotic and squamosal) has been interpreted as the pathway of either the postglenoid vein (Geisler & Luo, 1998) or a ramus of the middle meningeal artery (Fordyce, 1994a). The lateral tuberosity has a triangular outline and projects anterolaterally from the body of the periotic (Fig. 23A, B). A depression in the ventral surface of the lateral tuberosity represents the anterior edge of the articular facet for the spiny process of the squamosal, which by analogy with J. hunderi and NMV P48794 (Figs 25, 26) primarily articulated with the periotic at the hiatus epitympanicus (not preserved in the holotype of Ma. colliveri, but probably occurred posterolateral to the epitympanic recess at the level of the fenestra ovalis). Medial to the lateral tuberosity, the mallear fossa (for reception of the head of the malleus) is distinct from the rest of the epitympanic recess, whereas the fossa incudis (for reception of the crus breve of the incus) is relatively indistinct and lies adjacent to the posterior margin of the mallear fossa, level with the anterior edge of the fenestra ovalis (Fig. 23A, B). The ventral foramen of the facial canal is posterodorsal to the fossa incudis and opens into the facial sulcus, lateral to the fenestra ovalis. The facial nerve (CN VII) enters the middle ear cavity via the ventral foramen of the facial canal and passes onto the lateral surface of the skull via the shallow facial sulcus (Fig. 23A). A low transverse ridge divides the facial sulcus from the fossa for the stapedius muscle, which forms a deep pocket located slightly dorsal to the facial sulcus and immediately posterior to the fenestra ovalis. At its deepest point, the dorsal surface of the fossa for the stapedius muscle is at a level dorsal to the ventral margin of the fenestra rotunda (Fig. 24D). The posterior origin of the tensor tympani muscle is indicated by a deep anteroposteriorly orientated groove located medial and anteromedial to the mallear fossa (Fig. 23A, B). The anterior termination of this groove is at a point level with the preserved apex of the lateral tuberosity (Fig. 23B). The lateral edge of the groove for the tensor tympani is formed by a low rounded-over ridge on the anterolateral margin of the ventral surface of the pars cochlearis. In extant balaenopterids (e.g. USNM , Balaenoptera bonaerensis Burmeister, 1867; USNM , Balaenoptera physalus Linnaeus, 1758; USNM 21492, Megaptera novaeangliae Borowski, 1781) the origin for the tensor tympani is represented by a broad and shallow groove anteromedial to the mallear fossa that is continuous anteriorly with a broad triangular fossa that extends onto the medial surface of the anterior process of the periotic. The aperture of the fenestra ovalis is located at the fundus of a distinct vestibular fossula. In ventral view (Fig. 23A), the pars cochlearis has a semicircular outline, but is not inflated with smoothly rounded anterointernal and posterointernal angles (sensu Fordyce, 1994a: 161) and the pars cochlearis as a whole appears somewhat transversely compressed. Such pars cochlearis morphology differs from that of described basilosaurid archaeocetes (wherein the pars cochlearis has a rectangular outline in ventral view) (Fig. 27A), and from basal odontocetes (e.g. Fordyce, 1994a, 2002a) where the pars cochlearis is inflated and appears globular in ventral view. The most convex region on the pars cochlearis occurs on its ventral surface posteromedial to the vestibular fossula and anteroventral to the fenestra rotunda. The ventromedial surface of the pars cochlearis bears a distinct eminence located at the level (dorsoventrally) of the fenestra rotunda. There are no unambiguous vascular sulci or grooves traversing the surface of the pars cochlearis. The fenestra rotunda, located 6 mm posteromedial to the fenestra ovalis, has a suboval outline and a maximum diameter of 4 mm. The fenestra rotunda is at the interface between the inner ear and the middle ear cavity and in vivo was covered by the secondary tympanic membrane. Ventrolateral to the fenestra rotunda, the caudal tympanic process of the periotic forms a sharply triangular crest, such that in ventromedial view its ventral and posterior edges form a 90 angle (Fig. 23B). The cerebral surface of the pars cochlearis is elongated towards the cranial cavity, as seen in the periotics of adult individuals of Recent Balaenopteridae. This elongation of the pars cochlearis results in the cerebral edge of the lateral wall of the internal acoustic meatus being well dorsal to the adjacent suprameatal area on the body of the periotic (Fig. 23C). This contrasts with the condition of Basilosauridae wherein the lateral wall of the internal acoustic meatus is low and does not protrude internally from the surrounding pars cochlearis and suprameatal area (Fig. 27C F). Thus the lateral wall of the internal acoustic meatus in Ma. colliveri forms a longitudinal ridge almost equal in length to the pars cochlearis itself. In medial view (Fig. 23C), it can be seen that the cerebral edge of the medial wall of the internal acoustic meatus is at a level about 8 mm ventral to the cerebral edge of the lateral wall. Additionally, the cerebral edge of the lateral wall of the internal acoustic meatus increases in internal elon-

37 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 403 gation towards the level of the posterior edge of the internal acoustic meatus, where the pars cochlearis is at its maximum dorsoventral thickness (26 mm). The most prominent feature on the cerebral (or internal) surface of the pars cochlearis is the internal acoustic meatus (IAM) itself (Figs 23D, 25, 26). The IAM is a large recess into which the canals for the facial and vestibulocochlear nerves open. The foramina of the mysticete IAM have a different pattern of distribution to that in the periotics of terrestrial mammals, archaeocetes (Fig. 27E), and most odontocetes. Most notably, the area nervus facialis (into which the internal foramen of the facial canal opens; sensu Whitmore, 1953) is isolated from the area cochleae [into which the canal for the vestibulocochlear nerve (CN VIII) opens; sensu Whitmore, 1953] in the fundus of the IAM by an internally elongated and anteroposteriorly thickened transverse crest. Repenning (1972: ) used the term vestibulocochlear fossa as a synonym of the area cochleae in the petrosal of phocid pinnipeds. Micromysticetus rothauseni Sanders & Barnes, 2002b, a stem baleenbearing mysticete, possesses a low transverse crest, and thus an IAM similar to that of Basilosauridae. All more crownward Mysticeti possess an IAM where the area nervus facialis is isolated by the transverse crest from the area cochleae. As a result of this configuration of IAM foramina, anatomical terminology has been incorrectly used when applied to mysticete periotics. The most common inaccuracy is use of the term internal acoustic meatus to describe the area cochleae only, to the exclusion of the area nervus facialis (see Kellogg, 1928; Geisler & Luo, 1996; Sanders & Barnes, 2002a). In Ma. colliveri the area nervus facialis and area cochleae are arranged linearly within the internal acoustic meatus, with the area nervus facialis located directly anterior to the area cochleae (Fig. 23C, D). The cerebral apertures of the area nervus facialis and area cochleae have a subequal maximum diameter (about 7 mm) and in cerebral view (Fig. 23D) have an ovoid outline. Viewed medially (Fig. 23C), the medial walls of the area nervus facialis and area cochleae have U-shaped cerebral margins. The internal foramen of the facial canal opens into the fundus of the area nervus facialis and is subcircular in outline. The area nervus facialis is funnel-like, such that its diameter is significantly greater at its cerebral rim than at its fundus. A transversely and anteroposteriorly thickened transverse crest forms the posterior and anterior walls of the area nervus facialis and area cochleae, respectively. In medial view (Fig. 23C), the transverse crest has an acutely triangular outline. The area cochleae is more tubular in cross-section than the area nervus facialis, the diameter at its fundus being subequal to its diameter at the cerebral rim. Two foramina open within the fundus of the area cochleae: (1) the cochlear foramen, which transmits the cochlear branch of the vestibulocochlear nerve, and (2) the vestibular foramen, which transmits the vestibular branch of the vestibulocochlear nerve. Within the area cochleae the vestibular foramen is positioned anterior to the cochlear foramen, with its medial half located lateral to the centre of the cochlear foramen. The cochlear foramen has a diameter approximately equal to that of the internal foramen of the facial canal (about 2.75 mm), whereas the diameter of the vestibular foramen is less (about 1.5 mm). Both the cochlear and vestibular foramina are subcircular in outline. A triangular internally projecting wedge of bone, the pyramidal process, occurs posterolateral to the internal acoustic meatus (Figs 23C, 24A). Posteroventral to the pyramidal process the apertures for the cochlear and vestibular aqueducts open onto the cerebral surface of the pars cochlearis of the periotic (Fig. 23D). The aperture for the cochlear aqueduct is located 3.5 mm posterior to the cochlear foramen and 11 mm dorsal to the fenestra rotunda. The cochlear aqueduct communicates between the scala tympani and the subarachnoid space, transmitting the perilymphatic duct (MacPhee, 1981). Located directly lateral to the aperture for the cochlear aqueduct is the mediolaterally elongate aperture for the vestibular aqueduct (Fig. 23D). Sanders & Barnes (2002a) referred to this structure in Eomysticetus as the subarcuate fossa. The subarcuate fossa is a primitive feature for all mammals, housing the paraflocculus of the cerebellum (MacIntyre, 1972; Wible, 1990; Evans, 1993). In most eutherians, the subarcuate fossa is a hollow space encased by the three semicircular canals (Wible, 1990), but in Neoceti the semicircular canals are greatly reduced in size (Spoor et al., 2002) and the subarcuate fossa is subsequently obliterated as a distinct feature. In Ma. colliveri a septum of bone arises from the anterior wall of the aperture for the vestibular aqueduct, partially dividing the foramen. Ventrolateral to the internal acoustic meatus, the suprameatal area is smooth apart from numerous small nutrient foramina in its surface. The suprameatal area of the periotic body is not excavated by a concave fossa. A distinct dorsally projecting superior process of the periotic (as seen in basilosaurid periotics; Fig. 27B D) appears to have been absent, as indicated by another Mammalodon periotic (NMV P173220) with a complete body. The posterior surface of the pars cochlearis and body of the periotic are smoothly convex, with no stylomastoid fossa preserved posterolateral to the posterior edge of the fossa for the stapedius muscle (Fig. 24D).

38 404 E. M. G. FITZGERALD Figure 28. Mammalodon colliveri, NMV P199986, right tympanic bulla, whitened with ammonium chloride: A, ventral; and B, dorsal views. Diagonal solid lines indicate major breaks. Stapes The oval footplate of the stapes occupies the fenestra ovalis (Fig. 23A). An ovoid stapedial foramen appears to be patent between the tall anterior and posterior crura (Fig. 23B). A well-developed muscular tubercle for attachment of the stapedius muscle is present on the posteromedial surface of the neck of the stapes (Fig. 24A). Tympanic bulla The cetacean tympanic bulla is a highly specialized component of the middle ear, being markedly divergent in morphology from the tympanic of terrestrial mammals. Ridewood (1923), Eales (1950), and Luo (1998) have presented data supporting an ectotympanic origin for the cetacean tympanic bulla, with no contribution from the entotympanic to its development. Fordyce (1988) has previously mentioned and figured the tympanic bulla of Ma. colliveri. The holotype right tympanic bulla of Ma. colliveri (Figs 28, 29) lacks its posterior process, the anterodorsal extremity of the outer lip, and the sigmoid process. In dorsal or ventral view (Fig. 28), the tympanic bulla has a rhomboid outline, with a convex lateral margin and slightly concave medial edge of the involucrum. The tympanic bulla has a squared-off anterior edge and a bilobed posterior surface: a result of the well-developed interprominential notch sepa-

39 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 405 Figure 29. Mammalodon colliveri, NMV P199986, right tympanic bulla, whitened with ammonium chloride: A, posterior; B, medial; and C, lateral views. Diagonal solid lines indicate major breaks. rating the broad and squared-off inner posterior prominence from the rounded, transversely narrower and dorsoventrally deeper outer posterior prominence. The inner and outer posterior prominences extend posteriorly to the same level (Fig. 28). A transverse ridge on the posterior surface of the involucrum (Fig. 29A) represents a continuation of the low ventromedial keel (Fig. 29B) along the edge of the involucrum, forming an isthmus between the inner and outer posterior prominences, and bridging the interprominential notch. The interprominential notch is continuous anteriorly with a distinct median furrow on the ventral surface of the tympanic bulla (Fig. 28A). The median furrow becomes indistinct on the ventral surface of the tympanic at a point just posterior to the level of the lateral furrow on the outer lip. Anterior to the median furrow, the ventral surface of the tympanic is essentially featureless apart from the presence of numerous small nutrient foramina. As viewed dorsally (Fig. 28B), the involucrum is divided into a transversely thicker posterior part and a narrower anterior part by a prominent transverse groove. The transversely narrow anterior part of the involucrum has a lateral edge that is abruptly

40 406 E. M. G. FITZGERALD depressed into the tympanic cavity. A series of transverse creases in the dorsal surface of the anterior part of the involucrum terminate in five distinct tubercles arrayed linearly in an anteroposterior orientation. The more posterior, transversely thick, part of the involucrum has a smoothly convex dorsal surface. An anteroposteriorly broad transverse ridge arises from the lateral surface of the involucrum and divides the tympanic cavity into a smaller anterior space and larger posterior space. Viewed medially (Fig. 29B), the dorsal and ventral profiles of the involucrum converge anteriorly, with the ventral edge of the inner posterior prominence being dorsal to the ventral edge of the outer posterior prominence. The dorsal part of the outer lip is lost (Fig. 29C). A shallow U-shaped notch in the anterior edge of the outer lip at the anteromedial corner of the tympanic cavity represents the Eustachian outlet (homologous with the musculotubal canal of terrestrial mammals), which passes the Eustachian tube (Figs 28B, 29B). In lateral view (Fig. 29C), the apex of the outer posterior prominence has a smoothly rounded profile, whereas the outer lip has a convex ventral profile. Only the base of the sigmoid process is preserved, with the lateral furrow immediately anterior to it. Viewed posteriorly (Fig. 29A), a patent elliptical foramen forms a deep and narrow notch between the broken base of the outer posterior pedicle of the posterior process of the tympanic (on the outer lip) and the broken base of the inner posterior pedicle of the posterior process of the tympanic (on the involucrum). A distinct and patent elliptical foramen in Ma. colliveri implies the presence in vivo of a posterior sinus, which originates from the tympanic cavity via the elliptical foramen (Fraser & Purves, 1960; Kasuya, 1973). Pterygoid The pterygoid is exposed in the ventral surface of the skull (Figs 8, 9, 21, 22), lining the anterior and anteromedial extremities of the pterygoid sinus fossa. The hamulus and most of the delicate laminae are lost, with a remnant of the medial lamina of the pterygoid being preserved. The pterygoid articulates with the basisphenoid medially, and the alisphenoid dorsally and posterolaterally. The pterygoid is present in the braincase as far anteriorly as the level of the posterior edge of the optic foramen (Figs 13A B, 19, 20). The pterygoid sinus fossa is a relatively small triangular excavation in the ventral surface of the pterygoid anteriorly and alisphenoid posteriorly, located anterior to the cranial foramen ovale (Figs 21, 22). The pterygoid sinus fossa appears to have been limited to the region anterior to the cranial foramen ovale, with no extension into the orbital region. The preserved remnant of the medial lamina of the pterygoid borders the lateral edge of the basisphenoid posteriorly to the level of the anteriormost edge of the foramen ovale. There is no indication that a lateral lamina of the pterygoid contributed to formation of the subtemporal crest, although a well-developed lateral lamina of the pterygoid occurs in the related stem mysticete J. hunderi. A transversely narrow and dorsoventrally thin wedge of dorsal (or superior) lamina of the pterygoid forms the extreme anterior and anteromedial roof of the pterygoid sinus fossa, underlying the more extensively exposed alisphenoid which forms the posterior and posterolateral roof of the sinus fossa. Alisphenoid The alisphenoid is preserved on the ventral surface of the cranium (Figs 8, 9, 21, 22). By analogy with basilosaurids, Janjucetus, and an undescribed stem mysticete (NMV P216928), the alisphenoid was probably exposed on the external surface of the cranium forming the lateral wall of the orbital fissure (including the foramen rotundum), and perhaps contributing to formation of the subtemporal crest. Ventrally (Figs 21, 22), the alisphenoid forms the majority of the roof of the pterygoid sinus fossa, with the dorsal lamina of the pterygoid anterior and ventral to the alisphenoid at the extreme anteromedial corner of the sinus fossa. Posteromedially, the alisphenoid contacts the squamosal although exact sutural relationships are indistinct. Within the pterygoid sinus fossa, the alisphenoid is bordered anteromedially by the medial lamina of the pterygoid. The cranial foramen ovale marks the posterolateral border of the alisphenoid. Posteromedial to the cranial foramen ovale, the suture with the basisphenoid is obliterated. The mandibular branch of the trigeminal nerve (CN V 3) exits the braincase via the cranial foramen ovale, which in Mammalodon forms a notch in the posterior edge of the pterygoid sinus fossa. From the cranial foramen ovale, the mandibular branch of the trigeminal nerve presumably travelled anterolaterally along a hollow in the surface of the alisphenoid and/or squamosal, the alisphenoidal infundibulum (sensu Fraser & Purves, 1960: 35, 100) (see Figs 25, 26). It is presumed that the mandibular branch of the trigeminal nerve exits the skull from an external foramen ovale, as seen in Janjucetus and other archaic mysticetes (see Figs 25, 26). Presphenoid The presphenoid is an anteroposteriorly elongated median bone in the anterior half of the cranium that was probably not exposed on the external surface of the skull. The body of the presphenoid is roughly cylindrical in overall shape, with an ovoid crosssection. Anteriorly, the presphenoid articulates with the vomer (Figs 8, 9). At the level of the orbit the presphenoid is sutured dorsolaterally to a lamina of the frontal, and posterior to the level of the ethmoidal

41 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 407 foramina, articulates dorsomedially with the orbitosphenoid (Figs 19, 20). In extant mysticetes, the presphenoid fuses with the orbitosphenoid early in ontogeny, with sutures between these bones obliterated in neonates. In Ma. colliveri the suture between the presphenoid and orbitosphenoid is not entirely obliterated but remains relatively indistinct (see description of orbitosphenoid for further details of contact relationships). The presphenoid articulates posteriorly with the basisphenoid, although the suture between these bones is poorly preserved in NMV P In the skull of the neonatal minke whale Balaenoptera acutorostrata Lacépède, 1804 (e.g. NMV C24936; also see Ridewood, 1923) the posterior edge of the fused presphenoid and orbitosphenoid is at the level of the foramen rotundum, with the orbitosphenoid forming the anterior half of the foramen. The suture between the presphenoid and basisphenoid in Ma. colliveri is interpreted as being at the level of the foramen rotundum, about 60 mm anterior to the level of the anterior edge of the cranial foramen ovale (Figs 8, 9). Basisphenoid The median anteroventral surface of the braincase is formed by the basisphenoid, which articulates anteriorly with the presphenoid, laterally with the pterygoid and alisphenoid, and posteriorly with the basioccipital (Figs 8, 9). The posterior edge of the basisphenoid is indicated by a transverse region of broken bone immediately anterior to the level of the basioccipital crest (Figs 21, 22). This transverse area of broken bone probably includes the suture between the basisphenoid and basioccipital, although details of the suture itself are poorly preserved. There is no foramen preserved in the basisphenoid that may be unambiguously interpreted as the carotid foramen. In foetal and neonatal Recent Balaenoptera skulls the carotid foramen in the basisphenoid is present and patent (Ridewood, 1923), albeit apparently vestigial, and may transmit an internal carotid artery (Melnikov, 1997); in adult Recent mysticetes the carotid foramen probably does not transmit an internal carotid artery (Fraser & Purves, 1960; Vogl & Fisher, 1981). Fordyce (2002a) identified a carotid foramen in the basisphenoid of the stem odontocete Simocetus rayi. In Ma. colliveri there is a foramen in the basisphenoid at the level of the cranial foramen ovale that opens into a posterolaterally orientated sulcus (Figs 21, 22). The homology of this foramen is uncertain. Orbitosphenoid The orbitosphenoid is extensively exposed in the orbital wall, where it encircles the foramina of the anteroposteriorly elongated optic infundibulum (Fig. 13). The elongated morphology of the optic infundibulum in Ma. colliveri is similar to that of stem Cetacea, but contrasts with the anteroposteriorly compressed morphology of the optic infundibulum in living Neoceti. The orbitosphenoid articulates anteriorly and dorsolaterally with the frontal, ventrally with the presphenoid and frontal, posterolaterally with the parietal, and posteriorly and posteroventrally with the pterygoid, alisphenoid, and basisphenoid (Figs 19, 20). Most sutures between the orbitosphenoid and surrounding bones are rather indistinct. Nonetheless, it is possible to delimit the external surface of the orbitosphenoid within the orbit. The anterior edge of the orbitosphenoid is indicated by the position of the paired ethmoidal foramina (Figs 19, 20). The suture between the orbitosphenoid and frontal may be traced posteriorly from the ethmoidal foramina to where it continues along the postorbital ridge. The orbitosphenoid continues posteroventrally forming the medial wall of the foramen and canal for the ophthalmic artery. At its posterior edge, the orbitosphenoid forms the anteromedial wall of the orbital fissure and confluent foramen rotundum. Anterior to the foramen rotundum, the ventrolateral edge of the orbitosphenoid articulates with the pterygoid and, from the level of the optic foramen onwards, the presphenoid. Anteriorly from a point level with the optic foramen, a triangular wedge of frontal separates the orbitosphenoid from the presphenoid. Overall, the external surface of the orbitosphenoid has a rhomboid outline (Figs 19, 20). Within the optic infundibulum the optic foramen opens anteriorly at approximately the same height as the ethmoidal foramina, which lie about 37 mm anterior to the optic foramen (Figs 19, 20). The optic foramen opens anterolaterally into a shallow sulcus for transmission of the optic nerve (cr. II), with a sharp longitudinal ridge forming the foramen s ventral edge. Posteromedial and slightly ventral to the level of the optic foramen is the deep orbital fissure, including the foramen rotundum. The foramen rotundum opens anteriorly into the broad (dorsoventral diameter ~ 17 mm) groove for the maxillary nerve (CN V 2 the maxillary branch of the trigeminal nerve), which traverses the orbitosphenoid posteroventral to the optic foramen (Figs 19, 20). MANDIBLE The mandible (= dentary) of Ma. colliveri (Figs 30 32, 34B) may be divided into three distinct regions: the body of the mandible, which includes the horizontal tooth-bearing part of the bone (the alveolar portion of the mandible); the mandibular ramus, including the dorsally directed coronoid process; and the mandibular neck, which connects the mandibular condyle to the mandibular ramus. The body accounts for about 54% of the preserved length of the mandible in NMV P

42 408 E. M. G. FITZGERALD Figure 30. Mammalodon colliveri, NMV P199986, holotype mandible (right side), whitened with ammonium chloride: A, lateral; B, dorsal; and C, medial views. Salient plesiomorphic (i.e. basilosaurid-like: see Fig. 34) features of the mandible include: alveoli for functional teeth; strongly developed coronoid process that is both high and anteroposteriorly broad; a dorsoventrally high and anteroposteriorly elongated mandibular foramen and mandibular canal; and a dense mediolaterally thin pan bone forming the lateral wall of the mandibular foramen. Salient apomorphic features of the mandible include: anteroposteriorly short mandibular symphysis; straight body of the mandible

43 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 409 A coronoid process mandibular neck condyloid crest mandibular condyle m4 m3 m2 m1 p4 p3 p2 p1 c1 i3 i2 i1 position of pan bone mental foramina B coronoid crest 100 mm mandibular notch p2 p3 p4 m1 m2 m3 mandibular foramen inferred position of symphysis broken margin of mandibular foramen fossa posterior to mandibular foramen C 100 mm Figure 31. Mammalodon colliveri, NMV P199986, holotype mandible (right side): A, lateral; and B, medial views, based on Figure 30, showing main anatomical features. C, reconstruction of left and right mandibles in dorsal view. Diagonal solid lines indicate major breaks; diagonal dashed lines indicate matrix.

44 410 E. M. G. FITZGERALD Figure 33. Mysticeti gen. et sp. indet., NMV P16417, incomplete right mandible, whitened with ammonium chloride: A, lateral; and B, medial views. C, symphysis in medial view. Figure 32. Mammalodon colliveri, NMV P199986, holotype mandible (right side), whitened with ammonium chloride: A, anterior; and B, posterior views. C, alveoli for lower incisors and canine in dorsal view. D, symphyseal region in medial view, area enclosed by dotted line indicating the inferred position of the symphysis. (it is slightly recurved medially at the level of the mandibular symphysis: Fig. 30B); alveoli for 12 mandibular teeth (cf. 11 alveoli in Basilosauridae); alveoli without wide diastemata between them; no embrasure pits; and relatively large mental foramina. The mandible of NMV P is missing its anterior apex, most if not all of the mandibular symphysis, much of its ventral margin, parts of the dorsal margin of the mandible in the mandibular notch, and the posteromedial margin of the mandibular foramen. The lingual surface (Fig. 30C) of the mandibular body is mediolaterally crushed where it forms the medial wall of the mandibular canal, between the level of the alveolus for i2 and the posterodorsal edge of the mandibular foramen. Anteriorly, the apex of the mandible appears to have been rounded off but not dorsoventrally expanded. The mandibular symphysis of NMV P is enigmatic (Fig. 32D): there is no preserved trace of either a rugose region suggesting a bony sutural symphysis between the left and right mandibles (cf. archaeocetes and most Odontoceti) or a longitudinal groove correlated with possession of a loose fibrocartilaginous symphysis (characteristic of Chonecetus goedertorum, Aetiocetus spp., and most Chaeomysticeti). What then was the nature of the mandibular symphysis in Ma. colliveri? A right mandible referred to Mammalodon sp. cf. Ma. colliveri (NMV P199587) is equally unrevealing about symphyseal anatomy. However, an incomplete right mandible (NMV P16417; consisting of the symphyseal region and anterior alveolar portion) (Fig. 33) of an indeterminate toothed mysticete may shed light on the symphysis of Mammalodon. Although the edge of the mandible ventromedial to this concavity is corroded in NMV P (Fig. 32D), in NMV P16417 there is a raised area of rugose bone forming a short

45 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 411 A B C D E Figure 34. Mandibles of a basilosaurid and basal mysticetes in right lateral view. Mandibles are scaled to the same length. Dotted lines represent reconstructed regions of mandibular anatomy. Where necessary, left mandibles have been reflected to aid comparison. A, Zygorhiza kochii (modified from Kellogg, 1936); B, Mammalodon colliveri; C, Chonecetus goedertorum (modified from Barnes et al., 1995); D, Aetiocetus polydentatus (modified from Barnes et al., 1995); E, Eomysticetus whitmorei (modified from Sanders & Barnes, 2002a). (preserved length ~18 mm) mandibular symphysis (Fig. 33C). It is possible that the corroded anteroventral edge of the mandible s apex in NMV P represents the broken surface of a short mandibular symphysis, similar to that of NMV P This hypothesis remains to be falsified through discovery of a more completely preserved mandible that may be referred to Ma. colliveri. The mandibular symphysis of Ma. colliveri was clearly short compared to the elongate symphysis of basilosaurids. Viewed dorsally (Fig. 30B) the mandible is approximately straight, with no medial or lateral bowing. Despite some post-mortem crushing of its medial surface, the body of the mandible appears to have been transversely compressed and suboval in cross-section (Fig. 32A, B). Posterior to the mandibular symphysis, the dorsal and ventral edges of the body are subparallel, with the mandible having a constant height, until the level of the alveolus for p2. Posterior to this point the mandible gradually increases in depth towards the coronoid process. Despite having an incomplete ventral margin posterior to the level of m4, the form of NMV P suggests that the mandible of Ma. colliveri lacked a ventral expansion of the pan bone. Thus, the ventral margin of the mandible was straight along its length as in Basilosauridae (Fig. 34A), and did not have a dorsally arched ventral profile as in the aetiocetid C. goedertorum (Fig. 34C). The medial surface (Fig. 30C) of the body is gently dorsoventrally concave along its length. The mandibular foramen lies at the junction of the body of the mandible and the ramus. In Ma. colliveri the foramen and mandibular canal are large, having a maximum depth exceeding the greatest height of the body (> 65 mm) and a maximum width of about 20 mm. The mandibular foramen opens posteriorly into a broad fossa occupying the entire medial surface of the ramus ventral to the coronoid process, as well as the mandibular neck extending to the posteromedial edge of the mandibular condyle. The enlarged mandibular foramen and canal, as well as the fossa posterior to the mandibular canal, probably housed an intramandibular fat body. The lateral and medial walls of the mandibular foramen, and lateral wall of the posteriorly adjacent fossa, are composed of transversely thin (less than 3 mm thick) and dense bone: the pan bone (Figs 31A, 32B). The pan bone is typical of Basilosauridae and Odontoceti, but is absent in crown group Mysticeti. The lateral surface (Figs 30A, 31A) of the mandible is flat to dorsoventrally convex, with its greatest convexity occurring at the level of the pan bone on the ramus. The most prominent features on the lateral surface of the body are the mental foramina. There are seven primary mental foramina, and one smaller foramen located anterodorsal to the fourth mental foramen from the apex of the mandible. Viewed laterally (Figs 30A, 31A), the mental foramina are arrayed in an arc parallel with the lateral edge of the alveolar margin. Along the length of the arc formed by the mental foramina, the dorsal edge of each foramen is located mm ventral to the lateral edge of the alveolar margin.

46 412 E. M. G. FITZGERALD The lateral surface (Figs 30A, 31A) of the mandibular ramus and neck is vertically convex and featureless, apart from a rather indistinct condyloid crest that extends anteriorly from the mandibular condyle. In generalized placental mammals the condyloid crest forms the posteroventral border of the masseteric fossa. In Ma. colliveri there is no discrete fossa for the masseter. The mandibular condyle has a posterodorsally directed and transversely expanded articular surface, and in posterior view (Fig. 32B) has a pyriform outline. The mandibular condyle is positioned low on the mandible, with the dorsal edge of its articular surface at a level ventral to the alveolar margin of the body. In dorsal view (Fig. 30B), the mandible is transversely inflated posterior to the level of m4, and is widest at the level of the coronoid process and mandibular foramen. Immediately posterior to the alveolus for m4, the mandibular ramus gives rise to a transversely narrow coronoid crest that grades into the anterodorsal edge of the coronoid process. In lateral (Figs 30A, 31A) or medial (Figs 30C, 31B) views, the coronoid process of the mandible has a triangular shape with an almost straight posterior profile. The coronoid process is both dorsoventrally and anteroposteriorly expanded, as in Basilosauridae (Fig. 34A, B). A broad shallow fossa on the lateral surface of the coronoid process presumably represents part of the insertion for the superficial temporalis muscle. Viewed posteriorly (Fig. 32B), there is a medial inflection of the coronoid process at its midheight. A well-developed fossa on the medial surface of the coronoid process (Figs 30C, 32A B), anteroventral to its apex, is interpreted here as part of the insertion for the deep temporalis muscle. Viewed dorsally (Fig. 30B), the alveolar margin rotates laterally from the sagittal plane towards the apex of the mandible. Thus, alveoli posterior to the level of m1 are located on the dorsal edge of the body in line with the dorsal edge of the coronoid process and longitudinal axis of the mandible; alveoli anterior to the level of m1 are progressively more laterally positioned towards the symphysis. There are 12 alveoli in NMV P All of the alveoli are closely spaced along the alveolar margin with no diastemata between them. In Mammalodon, the lateral edge of the alveolar margin is salient and ridge-like; consequently the lower alveoli are distinct, but lie within a widely open alveolar groove. The preserved interalveolar septa between alveoli for i1 c1 are orientated at an angle of about 45 to the medial edge of the alveolar margin (Fig. 32C). The interalveolar septa between alveoli for p1 m4 are orientated perpendicular to the medial edge of the alveolar margin. The incisor-bearing alveoli occupy the anterior 42 mm of the mandible (Fig. 32C). In lateral view (Fig. 30A), it Figure 35. Isolated left upper incisor or right i1 associated with holotype skull of Mammalodon colliveri, NMV P199986, whitened with ammonium chloride: A, labial; and B, lingual views. can be seen that the latter alveoli as well as those for the canine and p1 open on the lateral surface of the mandible, as opposed to the more dorsal position along the alveolar margin of alveoli for p2 m4. The alveoli for i1 and i2 are confluent, forming an elongated alveolus with an ellipsoid outline in dorsal view (Fig. 32C). A medially directed eminence on the lateral wall of this long alveolus represents the vestigial interalveolar septum between the tiny anteriorly directed alveolus for i1 and the larger more dorsally directed alveolus for i2. The alveoli for i3 p1 are oval and have approximately equal maximum diameters. Its rounded alveolus, and lack of a transverse septum arising from its medial wall, suggests that p1 was single-rooted. The alveoli for p2 m3 are rectangular in outline, whereas the alveolus for m4 is oval. The sixth to 11th mandibular teeth (p2 m3) are in situ. DENTITION Perhaps the most noteworthy feature of Ma. colliveri, as a baleen whale, is its well-developed functional dentition (Figs 35 38). The teeth of Ma. colliveri are heterodont, with a dental formula of I1, 2, or 3/3, C1/1, P4/4, M3/4 (based on NMV P199986). Thus, although the upper dental formula is not polydont (i.e. tooth complement greater than the primitive placental dental formula of I3/3, C1/1, P4/4, M3/3), the lower dental formula is, with one additional molar

47 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 413 Figure 36. Mammalodon colliveri, NMV P199986, holotype elements, whitened with ammonium chloride: A, lateral view of preserved alveoli for right upper maxillary teeth (C1 to M2); B, buccal view of preserved left upper maxillary teeth (C1 P4); C, occlusal view of preserved left upper maxillary teeth (C1 P4). (m4). Together, the preserved portions of the left and right maxillae of the Ma. colliveri holotype show that there were eight alveoli in each maxilla (Fig. 36). The premaxillary (i.e. upper incisor) tooth complement is more difficult to determine: there is one alveolus preserved in each premaxilla of Ma. colliveri (Fig. 17). This presents two possible hypotheses for the upper incisor complement in Ma. colliveri: firstly, that there was at least one upper incisor and secondly that there were two or three, albeit vestigial, upper incisors. This uncertainty is reflected in the incisor formula given above. The preserved teeth of Ma. colliveri are interpreted as the permanent dentition, with no evidence of deciduous teeth: Mammalodon is tentatively interpreted as being monophyodont. Note that premolars and molars are differentiated here by their position in the tooth row. All described Basilosauridae, which (if monophyletic) is the sister group of Neoceti, have a tooth complement of I3/3, C1/1, P4/4, M2/3. Given this dental formula (primitive for Neoceti), the first four teeth posterior to C1/c1 are interpreted as P1 4/p1 4 in Ma. colliveri, with all teeth posterior to the fourth premolar considered as molars. The teeth preserved in the skull of NMV P are the root of left C1, and the left P1, P2, P3, and P4 (Fig. 36B, C). Teeth preserved in the holotype mandible of Ma. colliveri include the right p2, p3, p4, m1, m2, and m3 (Fig. 37). Isolated teeth include one incisor, which is

48 414 E. M. G. FITZGERALD Figure 37. Mammalodon colliveri, NMV P199986, right mandibular teeth (p2 to m3), whitened with ammonium chloride: A, buccal; B, occlusal; and C, lingual views. either a left upper incisor or right i1 (Fig. 35), and a left?p4 (NMV P17535) (Fig. 38). The isolated incisor is recurved posteriorly, singlerooted, and absolutely smaller than other preserved teeth, as well as being both absolutely and relatively smaller than the upper incisors of J. hunderi, which has a condylobasal length (CBL) similar to that of Ma. colliveri. In labial (Fig. 35A) or lingual (Fig. 35B) views the incisor has an elongated digitiform outline, with the preserved occlusal apex of the tooth being about twice as wide as the attenuated apex of the root. The lingual surface (Fig. 35B) of the root is more convex than the labial surface (Fig. 35A), which is relatively flat. The root is ovoid in cross-section. All enamel has been worn away from the crown of the incisor, which bears an oblique apical wear facet on its occlusal surface continuous with a shear wear facet extending onto the lingual surface of the root s base (Fig. 35B). Based on its size and orientation of the wear facets, it is equally plausible that this isolated incisor is either a left upper (perhaps second or third) incisor, or the right i1. The left C1 is broken across its single root (Fig. 36B), which shows that the root of C1 had an oval cross-section with a maximum anteroposterior diameter greater than its maximum mediolateral diameter. On the skull, all postcanine (or cheek) teeth (P1 P4) appear to be undistorted and in their original position (Fig. 36B C). These upper cheek teeth are strongly emergent, with crowns standing some

49 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 415 Figure 38. Isolated left?p4, NMV P17535, associated with holotype skull of Mammalodon colliveri, NMV P199986, whitened with ammonium chloride: A, buccal; B, lingual; C, distal; and D, occlusal views mm clear of the alveolus and projecting ventrolaterally. Indeed, the bases of the crowns on the buccal surface of the upper cheek teeth are located at a level about 7 mm exterior to the lateral edge of the maxilla. On the mandible (Fig. 37), p2 m3 are in place and do not appear to be distorted from their original positions within the alveoli. Here again, the cheek teeth are emergent, their crowns standing between seven and 12 mm clear of the alveoli. Interpretation of the crowns of both upper and lower cheek teeth are hampered by heavy occlusal wear that has obliterated most, if not all, details of coronal features. Nonetheless, the general appearance of the upper cheek tooth crowns appears to have been different to that of the lower cheek tooth crowns. The upper cheek teeth seem to have had relatively conical triangular crowns with almost subequal mesiodistal and buccolingual basal diameters. Lower cheek teeth have a more anteroposteriorly elongated, laterally compressed, crown with a mesiodistal basal diameter nearly twice the buccolingual diameter. The relatively complete crowns of m2 m3 show that the lower cheek teeth of Ma. colliveri had triangular outlines, similar to the lower cheek teeth of J. hunderi (Fitzgerald, 2006: 2958). The lower cheek teeth had a triangular apical denticle that was probably larger than the two or more accessory denticles (Fig. 37). Although complete upper tooth crowns of Ma. colliveri are unknown, it is probable that, like the lower cheek teeth, upper cheek teeth possessed a large apical denticle and multiple, smaller, accessory denticles. The distal accessory denticles on p4 m3 are relatively small, triangular (but more conical than laterally compressed), distinct from the rest of the crown, and are carinate. The distal accessory denticles on m3 decrease in size towards the base of the crown. These accessory denticles are arranged anteroposteriorly along the distal keels of the lower cheek teeth. The distal accessory denticles on p3, p4, and m1 are arranged further lingually than those on

50 416 E. M. G. FITZGERALD m2 m3, which are arranged along the midline of the crown. Apical wear is present on the preserved distal accessory denticles of p3, p4, m3, and m2. The upper cheek teeth (Fig. 36B, C) present a large, oblique (heavier on the lingual surfaces of the crown), apical wear facet on the crown that has removed all trace of any denticles. This wear facet is heavier on the distolingual aspect of the crown, and along the upper tooth row wear is heaviest on P1 (where no enamel remains), and decreases posteriorly towards P4. The enamel on the lingual base of the crown is preserved only on P4. The large apical wear facet grades into a lingual shear wear facet that extends onto the lingual surface of the root, just apical to the base of the crown, on P1, P2, and P3 (on which this wear facet is most strongly developed). The lower cheek teeth present a subhorizontal apical wear facet developed on the buccal half of the crown, with wear heaviest towards the mesiobuccal corner of the crown (Figs 37A B, 38A, D). This wear facet is heaviest on p2 where it has removed all enamel from the buccal and mesial surface of the crown. In addition to the apical wear facet on the crowns of lower cheek teeth, there is a narrow, dorsoventrally trending, shear wear facet on the mesiobuccal surface of the crowns of all teeth. This shear facet is continuous posteriorly with the apical wear facet, and extends ventrally onto the mesiobuccal surface of the anterior root of the right p2, p3, m2, and m3 (Fig. 37A). On p2, there is also a shear wear facet on the distobuccal surface of the posterior root. Where the occlusal wear is heaviest, on p2, the apical and buccal shear wear facets are blended. Oblique wear on the dorsolingual surface of the crown of lower cheek teeth (Figs 37B C, 38B, D) is interpreted as a third wear facet, resulting from lingual shear. This lingual wear facet is heaviest on p2 where it has removed all but a triangular wedge of enamel on the distolingual base of the crown. All cheek teeth preserve distinctly ornamented crown enamel that appears to be equally well developed on the buccal (Figs 37A, 38A) and lingual (Figs 37C, 38B) aspects of at least the lower teeth. On the upper cheek teeth (Fig. 36B), where most of the lingual surface of the crowns is worn off, the buccal ornament consists of closely spaced fine and low ridges orientated parallel to the longitudinal axis of the tooth, that arise from the base of the crown. Buccal ornament on the mandibular cheek teeth (Fig. 37A) is similar to that on the upper cheek teeth, although the ridges of enamel on the lower cheek teeth are relatively wider near the crown base, project further from the crown, and become narrower towards the apex. These lower buccal ridges anastomose, and in some instances bifurcate, towards the crown apex. The lower tooth lingual ornament (Fig. 37C) consists of finely fluted enamel at the anterior margin of the crown, grading posteriorly into low, broad-based ridges. A lingually wrapping cingulum arises from the base of the crown on the distal surface of p3 m3 (Figs 37C, 38B C). In distal view (Fig. 38C), this cingulum has an open U-shape, and the basalmost distal accessory denticle arises from its mid-point. On its buccal half (Fig. 38C), the cingulum bears tiny papillae. Lingually (Figs 37C, 38B), three to five conical papillae project from the cingulum, which forms a salient distolingual accessory shelf. On the right p3, p4, m1, m2, and left?p4, the lingual cingular papillae present apical wear. On all the lower cheek teeth (except the right p2 and m3), a dorsoventral median sulcus extends onto the buccal and lingual surfaces of the crown base from the vertical cleft between the two roots. On all cheek teeth, the boundary between enamel and cementum is distinct. In buccal view (Fig. 36B), the enamelocementum boundary on upper teeth has a straight profile. The enamelocementum boundary on the buccal surface of lower cheek teeth has an inverted V shape profile on p3 and p4, and a concave to straight profile on m1, m2, and m3 (Fig. 37A). Of the preserved cheek teeth at least P3, P4, p2, p3, p4, m1, m2, and m3 have multiple roots. In all of these teeth, the roots are fused along their length that is emergent from the alveoli. A break in the external surface of the left maxilla (Figs 6, 7) shows that the double roots of P4 curve posteriorly. A posteromedial swelling of the posterior root of P4 (Fig. 36C) suggests that it has fused with a vestigial lingually positioned third root. The lower cheek teeth possess two elongated roots that have an oval crosssection, are recurved posteriorly and arranged linearly (Fig. 38A, B). The roots of p3 m3 have a slight basal swelling ventral to the crown base. The roots of the isolated left?p4 are fused for over half their length via a transversely narrow isthmus (Fig. 38A, B). The posterior root of p4, m1, and m2 has a greater anteroposterior diameter than that of the anterior root, whereas the anteroposterior diameters of the roots of p2, p3, and m3 are subequal. HYOID APPARATUS Thyrohyoid The thyrohyoid is a paired element and one of the three bones that comprise the cetacean hyoid apparatus, the other two bones being the single, ventrally positioned basihyoid on the midline (from which the thyrohyoid projects posterolaterally), and the paired, dorsally positioned, stylohyoid. The thyrohyoid of Ma. colliveri (Fig. 39C, D; measurements in Table 9) is not fused to the basihyoid, and presumably articulated via a cartilaginous connection with a posterolateral facet on the basihyoid, as in basilosaurids (Uhen,

51 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 417 Figure 39. Mammalodon colliveri, NMV P199986, holotype elements: manubrium of sternum in: A, ventral; and B, dorsal views; right thyrohyoid in: C, anteroventral; and D, posterodorsal views. Table 9. Measurements (in mm) of the holotype right thyrohyoid of Mammalodon colliveri, NMV P Preserved length of thyrohyoid Maximum diameter of proximal end of thyrohyoid ), J. hunderi, and some extant odontocetes (Reidenberg & Laitman, 1994). Unlike that of extant mysticetes (Omura, 1964; Werth, 2007) and odontocetes (Werth, 1992, 2007; Reidenberg & Laitman, 1994; Bloodworth & Marshall, 2007), the thyrohyoid of Ma. colliveri is not dorsoventrally flattened and plate-like: it is massive and tubular, closely resembling the thyrohyoids of Dorudon atrox (Uhen, 2004: 62), Z. kochii (Kellogg, 1936: pl. 16), and Basilosaurus cetoides Owen, 1839 (Kellogg, 1936: 37). The thyrohyoid preserved with NMV P is the right element. In anteroventral (Fig. 39C) or posterodorsal (Fig. 39D) view, the proximal articular head of the thyrohyoid is expanded, with the more distal shaft being approximately triangular in cross-section and decreasing in diameter towards the distal extremity. The distal tip of the thyrohyoid is not preserved. A low ridge on the anteroventral edge of the thyrohyoid extends posterolaterally from the proximal articular end towards the preserved distal apex. This ridge is interpreted as part of the origin of the hyoglossus muscle, which acts to retract and depress the tongue. POSTCRANIAL SKELETON Axis vertebra The axis (cervical vertebra 2) (Fig. 40; measurements in Table 10) is the only vertebra preserved with the Ma. colliveri holotype. The pedicles of the neural arch are broken at their base, as are the apices of the transverse processes. The axis is not ankylosed to the third cervical vertebra (C3), and the posterior epiphysis is completely fused to the body of the axis. The axis has two articular surfaces for contact with the atlas: the ventral surface of the odontoid process, and the anterior articular surfaces. The transverse diameter of the axis, measured across the anterior articular surfaces (98 mm), is almost twice the maximum vertical diameter of the body measured across the epiphysis (53 mm). The preserved bases of the pedicles suggest that the neural arch projected dorsally. The preserved maximum transverse diameter of the neural canal is 41 mm. The odontoid process projects about 18 mm anteriorly from the level of the anterior articular surface, and has a maximum vertical diameter of about 28 mm and maximum transverse diameter of about 36 mm. In overall form, the odontoid process is relatively large and conical with a rounded-off anterior apex. Lateral and dorsolateral to the odontoid process (on both left and right sides) is the broad anterior articular surface. The right anterior articular surface, which is more complete than the left anterior articular surface, has a maximum vertical diameter of about 38 mm and a transverse diameter of 27 mm. Viewed anteriorly (Fig. 40A), the

52 418 E. M. G. FITZGERALD Figure 40. Mammalodon colliveri, NMV P199986, holotype axis vertebra (C2): A, anterior; B, posterior; C, dorsal; D, ventral; E, right lateral; and F, left lateral views. Table 10. Measurements (in mm) of the holotype axis vertebra of Mammalodon colliveri, NMV P Maximum width across the anterior articular surfaces Maximum transverse diameter of the odontoid process Maximum dorsoventral diameter of the odontoid process Maximum preserved width across the transverse processes Maximum length of the vertebra (including the odontoid process and epiphysis) Maximum transverse diameter of the posterior epiphysis Maximum dorsoventral diameter of the posterior epiphysis anterior articular surface has a reniform outline. The anterior articular surface is largely planar, becoming slightly concave towards its medial margin. The axis of Ma. colliveri differs from that of Z. kochii, D. atrox, C. goedertorum, and Micromysticetus rothauseni in having a poorly developed anterior articular surface ventral to the odontoid process The posterolaterally projecting transverse process is formed by fusion of the dorsally placed diapophysis and the ventrally positioned parapophysis. As seen in right lateral view (Fig. 40E), the diapophysis is barely distinguishable as a small eminence immediately dorsal to the massive parapophysis. The transverse foramen (or vertebrarterial canal) does not perforate the transverse processes, although a pit in the posterior surface (Fig. 40B) of the transverse process probably represents the vestige of the transverse foramen. The transverse foramen is patent in the axis of archaeocetes (Kellogg, 1936; Uhen, 2004), but not in that of basal mysticetes (Emlong, 1966; Sanders & Barnes, 2002a, b). The transverse foramen passes the vertebral artery. In posterior view (Fig. 40B), the epiphysis has an approximately pentagonal outline and its surface is flat around the margins, becoming concave towards its centre. Ventrally (Fig. 40D), the transversely broad ventral crest runs along the midline of the body of the axis, with the surface of the body lateral to the ventral crest being excavated. Viewed laterally (Fig. 40E, F), the ventral profile of the body forms an angle of about with the posterior edge of the epiphysis. On the dorsal surface of the body of the axis (Fig. 40C), a sharp dorsal crest occurs along the midline. Three pairs of nutrient

53 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 419 foramina, with one foramen in each pair placed either side of the dorsal crest, open on the dorsal surface. The surface of the body lateral to the dorsal crest is transversely concave. Sternum The manubrium (Fig. 39A, B; measurements in Table 11), the anteriormost element in the sternum, is preserved with NMV P The posterior edges of the manubrium are rugose, suggesting that this surface articulated with intersternebral cartilage present between the manubrium and mesosternal element posterior to it. Thus, the sternum of Ma. colliveri was composed of multiple unfused sternebral elements, as in archaeocetes (see Kellogg, 1936; Uhen & Gingerich, 2001; Uhen, 2004) and the toothed archaic mysticete Aetiocetus cotylalveus (Emlong, 1966: 38). This contrasts with extant mysticetes, wherein the sternum is composed of one element and there is no division into manubrium and sternebrae at any stage of ontogeny (Turner, 1870; Struthers, 1895a; Schulte, 1916; Arvy & Pilleri, 1977; Klima, 1978). Note that Rommel & Lowenstine (2001: 155) identified the single sternebral element in extant mysticetes as the manubrium. In extant odontocetes, the sternum is made up of multiple sternebral elements, which are variably fused (Arvy & Pilleri, 1977; Klima, 1978; Rommel, 1990). The manubrium of Ma. colliveri is morphologically disparate from all hitherto described mysticete sternebrae. It is approximately T-shaped, with its maximum width (measured across its anterior end) being greater than its maximum length. The T-shaped outline of the manubrium of Mammalodon resembles the manubria of protocetid grade archaeocetes (e.g. Uhen, 2001), being unlike the pentagonal manubria of basilosaurids (Kellogg, 1936; Uhen, 2004) and aetiocetids (pers. observ.). It is a relatively massive element, as in basilosaurids and aetiocetids, but is more dorsoventrally compressed than the manubria of those taxa and approaches the plate-like morphology of the sternum in modern mysticetes (Klima, 1978). The lateral edges of the body are concave in ventral (Fig. 39A) and dorsal (Fig. 39B) views. The preserved anterior edge of the manubrium is straight, whereas the posterior edge is triangular, with two roughened surfaces for articulation with the intersternebral cartilage forming an angle of about 160. The body of the manubrium has transversely convex dorsal (Fig. 39B) and ventral (Fig. 39A) surfaces, such that it has an oval crosssection. The dorsal and ventral surfaces of the body are longitudinally flat to slightly convex. The vertical diameter of the manubrium is greatest at its anterior end. Anteriorly, a robust process projects laterally from the anterolateral corner of either side of the body. On the dorsal surface (Fig. 39B), at the junction between the laterally projecting process and the body of the manubrium, there is a salient ovoid area of irregular bone. This area of roughened bone on both the right and left sides of the manubrium may represent the costal notch for articulation with the costal cartilage/sternal rib at the distal end of the first vertebral rib. Note that there is currently no evidence for sternal ribs in basilosaurids (Uhen, 2004), although costal cartilage may have been present. Extant odontocetes possess a sternal rib between vertebral rib 1 and the manubrium (Klima, 1978; Rommel, 1990), and costal cartilage occurs in foetal mysticetes (Schulte, 1916; Klima, 1978). Medial to the costal notches a broad, transversely concave, furrow occurs on the midline of the dorsal surface of the manubrium (Fig. 39B). This median furrow is continuous anteriorly with a transversely broad, anterodorsally directed, facet on the anterior face of the manubrium. This surface is both transversely and longitudinally concave. I interpret this broad depression as the deep origin for the sternohyoideus muscle. The sternohyoideus inserts on the posterior side of the basihyoid and thyrohyoid and acts to depress and retract the hyoid apparatus and tongue. PHYLOGENETIC SYSTEMATICS METHODS Background to analysis Cetacea represents a monophyletic group, either forming a clade nested within a paraphyletic Artiodactyla, or posited as sister group to a monophyletic Artiodactyla (e.g. Boyden & Gemeroy, 1950; Barnes & Table 11. Measurements (in mm) of the holotype manubrium of the sternum of Mammalodon colliveri, NMV P Length of the manubrium along its midline 97 Width of the manubrium across the costal notches 74 Width of the manubrium at its midpoint (anteroposteriorly) 56 Width of the manubrium across the surfaces for articulation with the intersternebral cartilage 76 Proximal depth of the manubrium, measured at the level of the costal notches 25 Distal depth of the manubrium, measured at the level of the anteriormost point on the posterior edge 26 of the articular surface for the intersternebral cartilage

54 420 E. M. G. FITZGERALD Mitchell, 1978; Barnes, 1990; Fordyce & Barnes, 1994; Heyning, 1997; Messenger & McGuire, 1998; Fordyce & Muizon, 2001; Gingerich et al., 2001; Thewissen & Williams, 2002; O Leary et al., 2004; Gingerich, 2005; May-Collado & Agnarsson, 2006; O Leary & Gatesy, 2007; Thewissen et al., 2007). A recent study corroborates a sister group relationship between Cetacea and the extinct Mesonychia, as a clade nested within a paraphyletic Artiodactyla (O Leary & Gatesy, 2007). Nevertheless, the identity of the proximate sister group to Cetacea amongst Artiodactyla (or Cetartiodactyla), remains controversial: Hippopotamidae (e.g. Gatesy et al., 1996; Nikaido et al., 2001a; Geisler & Uhen, 2003; May- Collado & Agnarsson, 2006), the extinct Anthracotheriidae + Hippopotamidae (Boisserie, Lihoreau & Brunet, 2005), extinct Raoellidae (Thewissen et al., 2007), and extinct Mesonychia (Mesonychidae + Hapalodectidae) (O Leary & Gatesy, 2007) have all been suggested as potential candidates. Within Cetacea, three groups have traditionally been recognized at the subordinal level: Archaeoceti, Odontoceti, and Mysticeti (Simpson, 1945; McKenna & Bell, 1997; Rice, 1998; Gingerich, 2005). Of these three taxa, Odontoceti and Mysticeti represent reciprocally monophyletic subclades, whereas Archaeoceti is clearly a paraphyletic cluster of stem taxa to the cetacean crown group clade Neoceti (Odontoceti + Mysticeti) (Fordyce & Muizon, 2001; Fordyce, 2002b). The grade Archaeoceti includes the ancestors of Neoceti, and the taxon Archaeoceti is thus equivalent to stem group Cetacea. Amongst stem Cetacea, the Basilosauridae, and, in particular, the subfamily Dorudontinae, are generally hypothesized to include the common ancestor of Neoceti (thus rendering Basilosauridae paraphyletic: Barnes & Mitchell, 1978; Barnes, 1990; Fordyce & Barnes, 1994; Uhen, 1998, 2004; Fordyce & Muizon, 2001; Uhen & Gingerich, 2001; Fordyce, 2002b, 2003a); or alternatively, if Basilosauridae is monophyletic then it would comprise the sister group to Neoceti (McLeod et al., 1993; Fordyce & Barnes, 1994; Luo & Gingerich, 1999). A monophyletic Basilosauridae has hitherto been recovered twice in a computeraided cladistic analysis that included multiple basilosaurid species (Luo & Marsh, 1996; Luo & Gingerich, 1999). The hypothesis of odontocete paraphyly, whereby sperm whales (Physeteridae) form a clade with monophyletic Mysticeti to the exclusion of other Neoceti (Milinkovitch, Ortí & Meyer, 1993; Milinkovitch, 1995, 1997), has been falsified by morphological and molecular data (e.g. Heyning, 1997; Messenger & McGuire, 1998; Cassens et al., 2000; Nikaido et al., 2001a, b; Geisler & Sanders, 2003; Fitzgerald, 2006; May-Collado & Agnarsson, 2006; Deméré et al., 2008). A stem-based clade Mysticeti is recognized here, consisting of several toothed archaic mysticete taxa placed stemward of an edentulous, baleen-bearing, clade designated Chaeomysticeti by Mitchell (1989), which includes the mysticete crown group. There is consensus that Chaeomysticeti is monophyletic (Kimura & Ozawa, 2002; Geisler & Sanders, 2003; Deméré, Berta & McGowen, 2005; Bouetel & Muizon, 2006; Fitzgerald, 2006; Steeman, 2007; Deméré et al., 2008). Beyond that consensus, phylogenetic relationships within Chaeomysticeti, including the interrelationships of the extant families Balaenidae, Neobalaenidae, Eschrichtiidae, and Balaenopteridae, are contentious and the subject of much recent research with as yet no general agreement (Kimura & Ozawa, 2002; Rychel, Reeder & Berta, 2004; Bouetel, 2005; Deméré et al., 2005; Sasaki et al., 2005; Bouetel & Muizon, 2006; Nikaido et al., 2006; Steeman, 2007; Deméré et al., 2008; Whitmore & Barnes, 2008). The relationships of basal mysticetes stemward of Chaeomysticeti are less controversial, reflecting fewer analyses of relevant fossils. Described species of toothed mysticetes have hitherto been classified in one of four families: Aetiocetidae (Emlong, 1966), Llanocetidae (Mitchell, 1989), Mammalodontidae (Mitchell, 1989), and Janjucetidae (Fitzgerald, 2006), each of the latter three families until now being monotypic. In addition to described taxa, an unnamed clade of archaic toothed mysticetes is recorded from the Oligocene of South Carolina, USA (Barnes & Sanders, 1996; Geisler & Sanders, 2003; Fitzgerald, 2006). As yet, no rigorous cladistic analysis has been performed that includes data from all the stem mysticete taxa listed above, although the works of Fitzgerald (2006) and Deméré et al. (2008) were preliminary steps in that direction. Uncertainty over the interrelationships and pattern of divergence amongst toothed mysticete taxa hinders attempts to understand the tempo and mode of basal mysticete evolution, the transition from archaeocetes to Neoceti, and evolutionary processes involved therein. Pivotal here is the Late Eocene Llanocetus denticrenatus, which, despite being the geologically earliest mysticete, indeed the earliest record of Neoceti, has been included in only one published analysis (Steeman, 2007). Intriguingly, that study posited Llanocetus as the sister taxon to Chaeomysticeti. This hints at a potentially more complex pattern of stem mysticete phylogeny than that set forth in Fitzgerald (2006) and Deméré et al. (2008). The overall aim of the following analysis is to test proposed phylogenetic schemes of toothed archaic Mysticeti. Special emphasis will be placed on evaluating the relationships of Ma. colliveri with other toothed mysticetes. In doing so, this analysis will test the patterns and processes of early mysticete evolution derived from previous studies of mysticete phylogeny.

55 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 421 Furthermore, the phylogenetic context of Mysticeti within Neoceti, and the relationship of Neoceti to stem Cetacea will be elucidated. Consequently, this analysis aims to test specifically the hypotheses that: (1) Basilosauridae is paraphyletic, with Neoceti derived from within Dorudontinae (e.g. Barnes & Mitchell, 1978; Uhen, 1998; Fordyce & Muizon, 2001; Uhen & Gingerich, 2001; Fordyce, 2002b; Bouetel & Muizon, 2006; Fitzgerald, 2006); (2) toothed mysticetes form a paraphyletic pectinate succession of stem taxa to Chaeomysticeti (Geisler & Sanders, 2003; Fitzgerald, 2006; Steeman, 2007; Deméré et al., 2008); (3) L. denticrenatus is a stem mysticete, sister to Chaeomysticeti (Steeman, 2007); (4) Aetiocetidae is monophyletic (Barnes et al., 1995; Geisler & Sanders, 2003; Steeman, 2007; Deméré & Berta, 2008; Deméré et al., 2008); (5) J. hunderi is stemward of Ma. colliveri, Aetiocetidae, and L. denticrenatus (Fitzgerald, 2006; Steeman, 2007; Deméré et al., 2008); (6) undescribed toothed mysticetes from South Carolina represent the most basal-branching clade within Mysticeti (Geisler & Sanders, 2003; Fitzgerald, 2006). Ingroup taxa Ingroup taxa were included with the aims of (1) sampling archaic odontocetes to establish the monophyly of Odontoceti and Neoceti; and (2) sampling the diversity of stem group Mysticeti, with an emphasis on toothed archaic mysticetes. All operational taxonomic units (OTUs) are species, and two OTUs are represented by undescribed specimens (ChM PV4745 and ChM PV5720). The ingroup includes 18 taxa: two species of Odontoceti (both taxa are extinct), and 16 species of Mysticeti (two extant taxa and 14 extinct taxa). Where possible, taxa were coded for character states via examination of original and/or cast specimens, supplemented with data acquired from published literature. Appendix 1 lists the specimens and references used to code the 18 ingroup and four outgroup OTUs. Simocetus rayi Fordyce, 2002a is the type species of the monotypic Simocetidae, and is known from a virtually complete skull with the right periotic in situ, incomplete right mandible, five isolated teeth, partial vertebrae, and several incomplete ribs (USNM , holotype) from the Alsea Formation of Oregon, northwest USA (Early Oligocene: early Rupelian). In his original description of S. rayi, Fordyce (2002a: 189) cites a late Oligocene (23 30 Mya) age for the Alsea Formation: recent work by Prothero et al. (2001) indicates that the Alsea Formation is entirely Early Oligocene, approximately Mya. With this revision to its geological age, S. rayi is now the geologically earliest formally named odontocete. Cladistically, S. rayi has been placed as the sister taxon to crown group Odontoceti (including the extinct families Eurhinodelphinidae, Squalodontidae, and Waipatiidae), being crownward to the basal odontocete Archaeodelphis patrius Allen (1921) (Fordyce, 2002a). Simocetus is diagnosed by the following autapomorphies: dorsoventrally flattened and edentulous rostral apex that is deflected ventrally; relatively short rostrum; ascending process of the maxilla not expanded laterally over the supraorbital process of frontal; deep and narrow optic infundibulum; complex sutures between the alisphenoid and squamosal; posterior upper cheek teeth occlude with diastemata between posterior lower cheek teeth; and i1 is vestigial (Fordyce, 2002a). Waipatia maerewhenua Fordyce, 1994a is the type species of Waipatiidae, and is known from an almost complete skull including periotics and the left tympanic bulla, teeth, both mandibles, atlas vertebra, a natural cast of the anterior axis, and one anterior thoracic vertebra (OU 22095, holotype) from the Otekaike Limestone of North Otago, South Island, New Zealand (Late Oligocene: late Chattian). There are two contrasting hypotheses for the phylogenetic position of Waipatia: (1) Waipatia is a crown odontocete, being one of several extinct taxa, which with the extant Platanista comprise clade Palatanistoidea (Fordyce, 1994a); and alternatively, (2) Waipatia is a proximal stem group to the odontocete crown clade (Geisler & Sanders, 2003). Waipatia can be distinguished from other platanistoid odontocetes (according to Fordyce, 1994a) by the following autapomorphies: relatively short and unfused mandibular symphysis; short and broad nasals; relatively large pterygoid sinus fossa posteromedial to the level of the cranial foramen ovale; bifid falciform process of squamosal, which apparently does not contact the lateral lamina of the pterygoid; relatively large and transversely inflated anterior process of the periotic; anterior process of the periotic is strongly reflected ventrally; anterior process of the periotic has a blunt apex; and the atlas vertebra lacks an elongated hypapophysis (Fordyce, 1994a). Aetiocetus cotylalveus Emlong (1966) is the type species of the family Aetiocetidae, and is known from a virtually complete skull (including the left periotic and tympanic bulla in place on the basicranium), six isolated presumed upper cheek teeth, 39 vertebrae, 23 ribs of varying completeness, and two sternal elements including the manubrium (USNM 25210, holotype) (Emlong, 1966). The holotype of Ae. cotylalveus was collected from the Yaquina Formation of Lincoln County, north-central coastal Oregon, USA (late Early to early Late Oligocene: Rupelian Chattian) (Emlong, 1966; Barnes et al., 1995; Prothero et al., 2001). In his original description, Emlong (1966) cited a late Late Oligocene age for the Yaquina Formation, and Barnes et al. (1995) followed this determination. In a review of North American Pacific Coast Palaeogene chronostratigraphy, Prothero et al. (2001) have indicated that

56 422 E. M. G. FITZGERALD the Yaquina Formation is somewhat older than previous estimates, being deposited between about 28 and 31 Mya and therefore mid-oligocene in age (late Rupelian to earliest Chattian). In addition to Ae. cotylalveus, three other species of Aetiocetus have been formally described: Ae. polydentatus (included here: see below), Ae. tomitai Kimura & Barnes in Barnes et al. (1995), and Ae. weltoni Barnes & Kimura in Barnes et al. (1995) (included here: see below). Aetiocetus cotylalveus is differentiated from other species of Aetiocetus by possession of an anteroposteriorly shortened intertemporal constriction and tuberosities on either side of the mid-line of the supraoccipital (Barnes et al., 1995). Aetiocetus weltoni Barnes & Kimura in Barnes et al. (1995) is known from an almost complete skull including the periotics and tympanic bullae, mandibles, hyoid bones, vertebrae, and ribs (UCMP , holotype) from the Yaquina Formation of Lincoln County, north-central coastal Oregon, USA (late Early to early Late Oligocene: Rupelian Chattian). Aetiocetus weltoni is differentiated from other species of Aetiocetus by having: a deeply concave supraorbital margin of the frontal, rostral bones with their posterior edges at a level within the posterior half of supraorbital process of the frontal, and an anteroposteriorly elongated squamosal fossa (Barnes et al., 1995; Deméré & Berta, 2008). Aetiocetus polydentatus Sawamura in Barnes et al. (1995) is known from a nearly complete skull with both tympanic bullae and periotics in place, both mandibles, several isolated left mandibular teeth, hyoid elements, and the atlas and axis vertebrae (AMP 12, holotype), collected from the Morawan Formation of Ashoro-cho, Hokkaido, Japan (Late Oligocene: Chattian). The postcranial elements of Ae. polydentatus have yet to be described. Deméré & Berta (2008: 344) recorded the following equivocal apomorphies as being diagnostic of Ae. polydentatus: broad nasals, ascending process of the premaxilla terminates at a level posterior to the posterior edge of the maxilla, nuchal crest overhangs the external surface of the braincase, and polydonty. Ichishima (2005) and Fitzgerald (2006) have suggested that its suite of autapomorphies may distinguish Ae. polydentatus from other Aetiocetus spp. at the generic level. This hypothesis has been challenged by Deméré & Berta (2008). Balaenoptera physalus Linnaeus, 1758 (fin whale: no holotype in existence) is a Recent species of Balaenopteridae included in the ingroup as a representative of crown mysticete diversity (see True, 1904; Tomilin, 1967; Gambell, 1985; Rice, 1998; Deméré et al., 2005; Sokolov & Arsen ev, 2006). The enigmatic Chonecetus sookensis Russell, 1968 is known from an incomplete braincase with the left periotic in place, and four partial vertebrae (Canadian Museum of Nature Vertebrate Fossil Collection No , holotype) from the Hesquiat Formation of Vancouver Island, British Columbia, Canada (earlymiddle Late Oligocene: Chattian) (Russell, 1968; Barnes et al., 1995). Barnes et al. (1995) stated that the age of the Hesquiat Formation is early Late Oligocene; Draus & Prothero (2003) suggested that deposition of the Sooke Formation, which overlies the Hesquiat Formation, occurred between 24 and 25 Mya, thus providing a minimum age constraint of about 25 Mya for the Hesquiat Formation. Chonecetus sookensis is characterized by the following features: paired clefts in the parietal on either side of the sagittal suture, relatively small zygomatic process of the squamosal, relatively small occipital condyles, anterior edge of the supraoccipital that is arcuate in outline, transversely expanded anterior process of the periotic, relatively short posterior process of the periotic (Barnes et al., 1995). It remains to be determined whether any of these features represent autapomorphies, although Barnes et al. (1995: 424, 426) imply that possession of paired clefts in the parietal on either side of the sagittal suture is an autapomorphy of this species. Chonecetus goedertorum Barnes & Furusawa in Barnes et al. (1995) is known from a reasonably complete skull with the left periotic in situ, left tympanic bulla, left mandible, and several postcranial elements including amongst others the atlas, axis, and other vertebrae, left and right scapulae, humerus, radius, and ulna (LACM , holotype), from the Pysht Formation of Clallam County, Washington, northwest USA (Late Oligocene: Chattian). The holotype tympanic bulla and postcranial elements (apart from the atlas) were not reported by Barnes et al. (1995), and remain undescribed. A partial skull and skeleton (LACM ) has been designated a paratype specimen (Barnes et al., 1995). Chonecetus goedertorum has been diagnosed by the following autapomorphies: transversely expanded intertemporal region, relatively large and rectangular zygomatic process of squamosal, and well-developed dorsal condyloid fossae (Barnes et al., 1995: 424, 426). Diorocetus hiatus Kellogg, 1968 is known from several specimens (e.g. USNM 16567, USNM 16783, USNM 16871, USNM 23494). The holotype (USNM 16783) consists of an essentially complete skull with the right periotic, both mandibles, cervical vertebrae 2, 6, and 7, first thoracic vertebra, and four ribs, collected from bed 14 of the Calvert Formation of Calvert County, Maryland, eastern coastal USA (Middle Miocene: Langhian) (Kellogg, 1968; Gottfried, Bohaska & Whitmore, 1994). Kellogg (1968: 134) did not provide a differential diagnosis of D. hiatus, and there are as yet no autapomorphies recognized for this species.

57 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 423 Eomysticetus whitmorei Sanders & Barnes, 2002a is known from an incomplete skull including both periotics and tympanic bullae, both mandibles, seven cervical vertebrae, seven thoracic vertebrae, two lumbar vertebrae, a possible caudal vertebra, at least 17 ribs, right scapula, humerus, radius, and ulna (ChM PV4253, holotype) collected from the Chandler Bridge Formation of Dorchester County near Charleston, South Carolina, south-east USA (Late Oligocene: early Chattian). In addition to Eo. whitmorei, the species Eomysticetus carolinensis Sanders & Barnes (2002a) (excluded here) has been described and is distinguished from Eo. whitmorei by numerous features (see Sanders & Barnes, 2002a: 340). The genus Eomysticetus is characterized by the following features: transversely narrow intertemporal constriction, elongated nasal, elongated zygomatic process of squamosal, zygomatic process of squamosal diverges anterolaterally from the sagittal plane, blade-like anterior process of the periotic (Sanders & Barnes, 2002a: 351). Eubalaena glacialis Müller, 1776 (North Atlantic right whale: no holotype in existence) is a Recent species of Balaenidae included in the ingroup as a representative of crown mysticete diversity (see True, 1904; Cummings, 1985; Rice, 1998; Kraus & Rolland, 2007). Janjucetus hunderi Fitzgerald, 2006 is known from a virtually complete skull with periotics and tympanic bullae, left and right mandibles, basihyoid, atlas, axis and third cervical vertebrae, two ribs, both scapulae, and an incomplete radius (NMV P216929, holotype) from the Jan Juc Formation of Jan Juc, south-west of Torquay, central coastal Victoria, south-east Australia (Late Oligocene: Chattian). The following features diagnose this taxon: dental formula is 9-10/10-11; premaxillae adjacent and anterior to the level of P2 overhang the maxillae; posterior edge of the premaxilla is in line with the anterior half of the supraorbital process of frontal; postorbital ridge absent; lateral end of the orbitotemporal crest is on the dorsal surface of the supraorbital process of frontal; sagittal crest present; superior lamina of the pterygoid bears a crest orientated at 45 to the sagittal plane that traverses the pterygoid sinus fossa and links the lateral and medial laminae of the pterygoid; accessory ossicle or homologous region on lip of tympanic bulla fused to the anterior process of the periotic (Fitzgerald, 2006; this study). Llanocetus denticrenatus Mitchell, 1989 was described on the basis of a partial right mandible with two posterior cheek teeth in situ, a fragment of maxilla bearing an incomplete cheek tooth, two isolated fragmentary cheek teeth, bone fragments, and a referred natural endocranial cast (USNM , holotype) from the top of unit Telm 7 of the upper La Meseta Formation of Seymour Island, northern Antarctic Peninsula (late Late Eocene: latest Priabonian) (Mitchell, 1989; Fordyce, 2003a, 2003b). Fordyce (1989b: 274) recorded a large skull and partial skeleton of a possible mysticete from the La Meseta Formation, which he later recognized as part of the holotype of L. denticrenatus (Fordyce, 1992: 373). These additional remains of the L. denticrenatus holotype add substantially to the anatomy of this putative mysticete and include the following elements: relatively complete skull including periotics and the right tympanic bulla, incomplete right mandible, several teeth, sternal elements, cervical vertebrae, numerous ribs, a possible femur, and possible pelvic remains (Fordyce, 2003b). Fordyce (2003a, b) has suggested that some teeth and skull remains, originally described by Keyes (1973) as a proto-squalodontid, probably represent a new (and smaller) species of Llanocetus from the Lower Oligocene Ototara Limestone of New Zealand. Llanocetus denticrenatus is currently characterized by the following features: massive and deep mandible; alveolar margin of the mandible is transversely narrow, forming a longitudinal crest medial to the alveoli; cheek teeth have palmate accessory denticles, with a broadly open notch between the denticles. Mammalodon colliveri Pritchard, 1939 was described in detail in this study. Micromysticetus rothauseni Sanders & Barnes, 2002b is known from an incomplete braincase including both periotics, and the axis vertebra (ChM PV4844, holotype), and a second partial braincase (ChM PV5933, paratype) known from the Ashley Formation of Dorchester County, north of Charleston, South Carolina, southeast USA (late Early Oligocene: latest Rupelian). In addition to Mi. rothauseni, Micromysticetus tobieni has been referred to the genus Micromysticetus (Sanders & Barnes, 2002b). Micromysticetus rothauseni is distinguished from Mi. tobieni by the following features: broad posterior margin of the squamosal fossa, external occipital crest does not extend posteroventrally proximal to the dorsal intercondylar notch, and basioccipital crests have rhomboid outline. Pelocetus calvertensis Kellogg, 1965 is known from several referred specimens (e.g. USNM 11976, USNM 14693, USNM 21306, USNM 23058). The holotype (USNM 11976) consists of an essentially complete skull including the periotics and tympanic bullae, both mandibles, hyoid bones, several vertebrae, scapulae, humeri, radii and left ulna, carpals, metacarpals, and phalanges, and ribs from bed 13 of the Calvert Formation of Calvert County, Maryland, eastern coastal USA (late Early early Middle Miocene: late Burdigalian early Langhian) (Kellogg, 1965; Gottfried et al., 1994). Kellogg (1965: 3) did not provide a differential diagnosis of P. calvertensis, and there are as yet no autapomorphies recognized for this species.

58 424 E. M. G. FITZGERALD The specimen ChM PV4745 comprises an almost complete skull (rostrum disarticulated from cranium), with periotics and tympanic bulla, teeth, atlas vertebra, and other elements from the Ashley Formation of Dorchester County, near Charleston, South Carolina, south-east USA (late Early Oligocene: latest Rupelian). This specimen remains undescribed, although Barnes & Sanders (1996) reported that it represents one of two new species in a new genus and new family of toothed archaic mysticetes. This new family of toothed archaic mysticetes is hypothesized to be the most basal-branching clade of mysticetes (Geisler & Sanders, 2003; Fitzgerald, 2006). These stem mysticetes bear many plesiomorphies in common with basilosaurids, most notably: upper cheek teeth not separated by wide diastemata; relatively large cheek teeth with crowns bearing smooth enamel and closely spaced accessory denticles; embrasure pits; and a sutural mandibular symphysis (Barnes & Sanders, 1996). Indeed, these plesiomorphic features led Sanders, Weems & Lemon (1982: H118) to originally consider these cetaceans as archaeocetes. Unlike basilosaurids, the South Carolina toothed mysticetes have a third upper molar (M3) and a fourth lower molar (m4) (Barnes & Sanders, 1996). Being undescribed, apomorphies for the new taxon represented by ChM PV4745 have yet to be recognized, although Barnes & Sanders (1996) listed the following characteristics: abruptly tapered, broad-based rostrum, closely spaced and imbricated posterior upper cheek teeth, smooth symphyseal surface on the mandible. The specimen ChM PV2778 (not included in this analysis) represents a second species in this genus, and was collected from the Upper Oligocene Chandler Bridge Formation of South Carolina (Geisler & Sanders, 2003). The specimen ChM PV5720 comprises a nearly complete skull including periotics and tympanic bullae, teeth, mandibles, atlas and axis vertebrae, and other elements, collected from the Chandler Bridge Formation of South Carolina, south-east USA (early Late Oligocene: early Chattian). This specimen represents an undescribed new genus and species in the new family reported by Barnes & Sanders (1996), which also includes the new genus represented by ChM PV4745 and ChM PV2778 (Geisler & Sanders, 2003; Fitzgerald, 2006). The genus represented by ChM PV5720 differs from the other new genus in this family by having: a markedly larger skull ( m long), more tubular anterior half of the rostrum formed by elongated tooth-bearing premaxillae, anteroposteriorly shortened nasals with elevated anterior margins (phenetically similar to the up-turned nasals of Aetiocetus polydentatus), and a rugose symphyseal surface on the mandible (Barnes & Sanders, 1996). Note that Zimmer (1998: 207) referred to ChM PV5720 by the generic name Archaeomysticetus, but in the latter instance this name is a nomen nudum. Outgroup taxa Four species of stem Cetacea were included as outgroups: Georgiacetus vogtlensis Hulbert et al. (1998), a nominal protocetid ; and three species of Basilosauridae, the basilosaurine Basilosaurus isis Beadnell in Andrews, 1904, and the dorudontines D. atrox Andrews, 1906 and Z. kochii Reichenbach in Carus et al., Barnes & Mitchell (1978) argued that mysticetes and odontocetes arose from within Dorudontinae, a concept that has persisted in subsequent discussions of the relationship of Odontoceti + Mysticeti to archaeocetes (Fordyce, 1980, 2002b, 2003a; Barnes, Domning & Ray, 1985; Barnes, 1990; Fordyce & Barnes, 1994; Barnes & Sanders, 1996; Geisler & Luo, 1998; Uhen, 1998, 2002; Fordyce & Muizon, 2001; Uhen & Gingerich, 2001; Gingerich, 2005). Most previous phylogenetic analyses have placed one taxon from a paraphyletic Basilosauridae as the sister group to Neoceti, whether that taxon is Saghacetus osiris Dames, 1894 (Uhen, 1998), Chrysocetus healyorum (Uhen & Gingerich, 2001; Uhen, 2004), D. atrox (Bouetel & Muizon, 2006), or Basilosaurus sp. (Geisler & Uhen, 2003; O Leary & Gatesy, 2007). The analyses of Luo & Marsh (1996) and Luo & Gingerich (1999) recovered a monophyletic Basilosauridae (Bas. isis, D. atrox, and Z. kochii) sister to Neoceti. No analysis has hitherto included multiple basilosaurid species and toothed mysticete families simultaneously. Basilosaurus isis, D. atrox, and Z. kochii are well-known basilosaurid taxa represented by abundant, complete, and superbly preserved skeletal material. In addition to Basilosauridae, the stem cetacean G. vogtlensis Hulbert et al., 1998 is a Neoceti outgroup, which lacks synapomorphies that unite basilosaurids with Odontoceti + Mysticeti. Georgiacetus vogtlensis is considered a species within the Protocetidae positioned proximal to Basilosauridae amongst stem Cetacea (Hulbert et al., 1998; Uhen & Gingerich, 2001; Geisler & Sanders, 2003; Geisler & Uhen, 2003; Geisler, Sanders & Luo, 2005; Fitzgerald, 2006). Basilosaurus isis Beadnell in Andrews (1904) is known from several virtually complete skeletons from the Gehannam, Birket Qarun, and Qasr el-sagha formations of Fayum, Egypt (late Eocene: latest Bartonian to early Priabonian) (Gingerich, 1992; Uhen, 1998). The holotype (CGM 10208) is an incomplete mandible. Despite being known from almost the entire skeleton, most elements (including the skull and ear bones) have yet to be described and figured in detail (Uhen, 1998). As a result, it is unclear what autapomorphies differentiate Bas. isis from other species of Basilosaurus, although Gingerich, Smith & Simons (1990: 155) and Uhen (1998: 33) state that Bas. isis has

59 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 425 a lesser total skeletal length than Basilosaurus cetoides Owen, Basilosaurus isis was referred to the genus Basilosaurus on the basis of its possession of anteroposteriorly elongated posterior thoracic, lumbar, and anterior caudal vertebrae (Gingerich et al., 1990; Uhen, 1998). What published information exists on the anatomy of Bas. isis is available in Andrews (1906), Stromer (1908), Kellogg (1936), Gingerich et al. (1990), Uhen (1998), and Luo & Gingerich (1999). The periotic of Bas. isis is illustrated in Figure 27. Dorudon atrox Andrews, 1906 is known from numerous specimens that have enabled in-depth description and analysis of almost every skeletal element (Uhen, 1998, 2004). The holotype (CGM 9319) consists of a virtually complete skull with right mandible. Dorudon atrox has been recovered from the Gehannam and Birket Qarun formations (Gingerich, 1992; Uhen, 1998, 2004), and less certainly the Qasr el-sagha Formation (Uhen, 1998), of Fayum, Egypt (late Eocene: late Bartonian to early Priabonian). No explicit autapomorphies of D. atrox have yet been identified, although Uhen (2004) presents a noncladistic differential diagnosis for this species. Zygorhiza kochii Reichenbach in Carus et al. (1847) was described and illustrated by A. Remington Kellogg (1936) in his landmark monograph. Owing to the well-preserved cranial and postcranial remains of Zygorhiza, this monotypic genus has figured prominently as an outgroup to Neoceti in numerous phylogenetic analyses (e.g. Fordyce, 1994a; Luo & Marsh, 1996; Geisler & Sanders, 2003; Bouetel & Muizon, 2006; Fitzgerald, 2006; Deméré et al., 2008). Despite this wide citation and the key work of Kellogg (1936), Z. kochii deserves a modern redescription and interpretation of its morphology. Zygorhiza is recorded from several formations in the south-east USA, all of which are Late Eocene (Priabonian) (Uhen, 1998). The type specimen (MNB 15324a-b (m.44), a partial cranium) was collected from the Moodys Branch Formation of Alabama (Kellogg, 1936; Uhen, 1998). Uhen (1998) notes that Kellogg (1936) distinguished Z. kochii from other dorudontine basilosaurids by its possession of well-developed crenulated cingula on the mesiolingual and distolingual sides of P2 4. Köhler & Fordyce (1997) implied that P3 4 with a posterior root notably larger than the anterior root is a potential autapomorphy of Zygorhiza. Additionally, Zygorhiza differs from Basilosaurus, Dorudon, Saghacetus osiris, and Ancalecetus simonsi Gingerich & Uhen, 1996 in having a paroccipital process of the exoccipital that is directed anterolaterally (subparallel with the axis of the external acoustic meatus) such that its apex is at a level anterior to the occipital condyles (Luo & Gingerich, 1999: 41). Georgiacetus vogtlensis Hulbert et al., 1998 is one of the more crownward cetaceans placed in the paraphyletic Protocetidae (Uhen & Gingerich, 2001; Geisler et al., 2005) (holotype: GSM 350), and is known from well-preserved, virtually complete, cranial material and much of the postcranial skeleton (holotype: GSM 350), described by Hulbert (1998) and Hulbert et al. (1998). Georgiacetus is derived from a stratigraphical unit informally named the Blue Bluff unit, which has been dated to the late Middle Eocene (late Lutetian to early Bartonian) (Hulbert et al., 1998). The type locality is near Augusta, Burke County, Georgia, USA. Georgiacetus is therefore the geologically oldest taxon included in this analysis. Hulbert et al. (1998) provided a differential diagnosis of G. vogtlensis; however, autapomorphies for this species have yet to be explicitly identified. Analysis protocol The complete data matrix (Appendix 3) consists of 22 taxa (18 ingroup taxa and four outgroup taxa), coded for 123 morphological characters (Appendix 2). Of the 123 characters in the data matrix, seven are original to this study. Appendix 2 provides the sources for all characters in the matrix. The character/taxon data matrix was entered and formatted in MacClade (version 4.05) (Maddison & Maddison, 2002). Phylogenetic analyses were performed using parsimony in PAUP* 4.0b10 (Swofford, 2002), with all characters equally weighted and all characters, except characters 17 and 25, treated as unordered. Polymorphic character codings were treated as uncertainties. Where taxa were coded for gaps, the gap coding was interpreted as an additional state. During searches, branches found with a minimum length of zero were collapsed ( amb- option) (see Kearney & Clark, 2003). Two analyses were performed under different character state optimization regimes: one under accelerated transformation (ACCTRAN), and a second under delayed transformation (DELTRAN) optimization. Unequivocal apomorphies are considered to be those that occur at nodes under both ACCTRAN and DELTRAN character state optimization. The analysis protocol consisted of 5000 tree bisection and reconnection (TBR) heuristic search replicates holding a single tree for each replicate, with branch swapping on the shortest trees. A bootstrap analysis was performed in PAUP* with 1000 bootstrap replicates and ten random stepwise addition heuristic pseudoreplicates per bootstrap replicate. Character optimization and evolution on the most parsimonious tree was traced using MacClade. RESULTS The parsimony analysis of the data set recovered a single most parsimonious tree (MPT) of 451 steps [consistency index (CI) = 0.490; retention index (RI) = 0.636; rescaled consistency index (RC) = 0.311] (Fig. 41). The

60 426 E. M. G. FITZGERALD MPT includes the following monophyletic groups: Basilosauridae, Neoceti (Odontoceti + Mysticeti), Odontoceti, Mysticeti, Llanocetidae, Mammalodontidae, Aetiocetidae, and Chaeomysticeti. The toothed archaic mysticete families form a pectinate succession of stem taxa to the edentulous mysticete clade (Chaeomysticeti). Specific taxa and unnamed clades are discussed in further detail below. Basilosauridae In contrast to nearly all previously published phylogenetic analyses (see discussion above), this analysis found Basilosauridae to be monophyletic and hence the sister group to Neoceti (Fig. 41). The most parsimonious phylogenetic hypothesis presented here suggests that Dorudontinae is paraphyletic, with Bas. isis forming a clade with D. atrox to the exclusion of Zygorhiza. This topology is perhaps spurious, and may be biased by relatively low taxonomic sampling of basilosaurids in this analysis, and exclusion of postcranial characters from the data set. The latter is potentially significant because basilosaurines are diagnosed by having elongated posterior thoracic, lumbar, and anterior caudal vertebrae (Uhen, 1998). The analyses of Uhen (1998, 2004) and Uhen & Gingerich (2001) recovered a monophyletic Basilosaurinae sister to a paraphyletic cluster of dorudontine taxa positioned stemward of Neoceti. Geisler & Sanders (2003: 67) hypothesized that an analysis including multiple basilosaurids and archaic Neoceti may find one or more basilosaurid taxa to be within Neoceti. Hence, some nominal basilosaurids would be more closely related to Odontoceti or Mysticeti than to other Basilosauridae. This analysis does not corroborate such a hypothesis: odontocetes and mysticetes are more closely related to one another than either is to any basilosaurids included here. A monophyletic Basilosauridae as hypothesized in this analysis is diagnosed by the following unequivocal synapomorphies: character 5 (1 2) narrow-based rostrum, character 84 Georgiacetus vogtlensis Zygorhiza kochii Basilosauridae Basilosaurus isis Dorudon atrox Odontoceti 97 Simocetus rayi Waipatia maerewhenua Neoceti ChM PV5720 ChM PV4745 Llanocetus denticrenatus L Mysticeti Mammalodon colliveri Janjucetus hunderi M Chonecetus sookensis Chonecetus goedertorum Aetiocetus polydentatus Aetiocetus cotylalveus Aetiocetus weltoni A Eomysticetus whitmorei 88 Micromysticetus rothauseni Diorocetus hiatus Pelocetus calvertensis Eubalaena glacialis Balaenoptera physalus C Figure 41. Single most parsimonious tree of 451 steps produced by analysing a data matrix of 22 taxa and 123 characters. Numbers at nodes are bootstrap proportions. In the shortest tree there is a monophyletic Basilosauridae, Neoceti, Odontoceti, Mysticeti, Llanocetidae, Mammalodontidae, Aetiocetidae, and Chaeomysticeti. Taxa in bold represent toothed archaic mysticetes. Abbreviations: A, Aetiocetidae; C, Chaeomysticeti; L, Llanocetidae; M, Mammalodontidae. For other abbreviations see Material and methods.

61 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 427 (1 0) anterior process of the periotic is highly elliptical in cross-section, and character 92 (0 2) well developed longitudinally orientated ridge on the anterolateral side of the pars cochlearis of the periotic. Neoceti As in previous phylogenetic analyses (e.g. Messenger & McGuire, 1998; Uhen & Gingerich, 2001; Geisler & Sanders, 2003; Fitzgerald, 2006), Odontoceti + Mysticeti is recovered as the strongly supported (bootstrap = 88%) crown cetacean clade Neoceti. This corroborates the hypothesis that odontocetes and mysticetes share a common ancestor within Archaeoceti (e.g. Van Valen, 1968; Barnes & Mitchell, 1978; Barnes, 1990; Fordyce, 2002b). Five unequivocal synapomorphies diagnose Neoceti: character 7 (1 or 0 4) open mesorostral groove, character 38 (0 2) three or more dorsal infraorbital foramina in the maxilla, character 83 (0 1) reduced articulation between anterior process of the periotic and the squamosal such that the edges of the anterior process are free of the squamosal, character 90 (0 1) rounded anterointernal angle of the pars cochlearis of the periotic, and character 112 (0 1) amastoid skull such that the posterior process of the periotic is not exposed on the external surface of the braincase. Two other characters represent possible neocete synapomorphies: character 2 (0 1) anteroposteriorly elongated rostral portion of maxilla, and character 65 (0 1) apex of the zygomatic process of squamosal contacts or nearly contacts the postorbital process of frontal. Of these characters 7 and 112 were identified by Fordyce (2002b); 38 by Barnes (1990), Fordyce (2002b), Sanders & Barnes (2002a), and Geisler & Sanders (2003); and 65 by Barnes (1990) and Sanders & Barnes (2002a), as potentially synapomorphic for Neoceti. Some previously hypothesized neocete synapomorphies are not recognized in this analysis, such as: (1) concave facial region, (2) posteriormost upper tooth located anterior to the level of the antorbital notch, and (3) mandibular ramus lateral to the mandibular foramen is transversely thin and dense forming the pan bone (Barnes, 1990; Fordyce, 2002b; Sanders & Barnes, 2002a; Geisler & Sanders, 2003). A concave facial region (= facial fossa) (character 37 of this study) is found to be present in basal Neoceti, but also in stem cetacean taxa included within the outgroup. The posteriormost upper tooth is anterior to the level of the antorbital notch (character 122 of this study) in stem odontocetes (e.g. Simocetus rayi; Fordyce, 2002a: and Xenorophus sloanii; Kellogg, 1923) and aetiocetids (Fig. 2). Inclusion of Llanocetus, Janjucetus, and Mammalodon in this analysis, which have a posteriormost upper tooth level with or posterior to the antorbital notch (Fig. 2), results in this character being equivocally optimized at the base of Neoceti. The pan bone on the mandibular ramus is present in Basilosauridae as well as basal odontocetes and stem mysticetes, implying that this character may be a synapomorphy of a more inclusive clade (Basilosauridae + Neoceti). Finally, polydonty (tooth count increased over the generalized placental mammal dental complement; i.e. four or more teeth posterior to the third molar in a quadrant) and monophyodonty have been noted as potential synapomorphies of Neoceti (Barnes & Mitchell, 1978; Barnes, 1990; McLeod et al., 1993; Fordyce & Muizon, 2001; Uhen & Gingerich, 2001; Sanders & Barnes, 2002a; Uhen, 2002). Polydonty is equivocally optimized on the most parsimonious tree produced by this analysis. Whereas basilosaurids are diphyodont (Kellogg, 1936; Uhen, 1998, 2000, 2002, 2004), extant Neoceti are monophyodont (Karlsen, 1962; Fordyce, 1982b; Uhen & Gingerich, 2001; Uhen, 2002). Monophyodonty was not included as a discrete character in this analysis because of the lack of unambiguous evidence on tooth eruption in basal odontocetes and mysticetes. New synapomorphies of Neoceti are identified by this analysis. A reduced articulation between the anterior process of the periotic and the squamosal, whereby the ventrolateral and dorsolateral edges of the anterior process are free of the squamosal (character 83), occurs in the basal odontocete Simocetus and several toothed archaic mysticetes (e.g. Fig. 25; also see fig. 1E in Fitzgerald, 2006). In protocetids (such as Georgiacetus) and basilosaurids, the entire lateral surface and ventrolateral edge of the anterior process of the periotic is tightly appressed to the squamosal, or covered by it (e.g. Hulbert et al., 1998; Luo & Gingerich, 1999). A rounded anteromedial corner (strictly, the anterointernal angle) of the pars cochlearis of the periotic (character 90) is present in each ingroup taxon included in this analysis and scored for this character. Amongst basal Neoceti, a rounded anterointernal angle of the pars cochlearis occurs in Simocetus (Fordyce, 2002a: fig. 14), Mammalodon (Fig. 23A), Janjucetus (Fitzgerald, 2006: fig. 1E), an undescribed?mammalodontid (NMV P48794: Fig. 25), and both Chonecetus species (pers. observ.). Possession of a rounded anterointernal angle of the pars cochlearis results in the neocete pars cochlearis having a hemispherical outline when viewed ventrally. Basilosauridae possess a distinctly angular anteromedial corner of the pars cochlearis, with the anteroventral and ventromedial edges of the pars cochlearis being almost perpendicular to one another: thus, the pars cochlearis of the periotic of Basilosauridae has a trapezoid to rhomboid outline in ventral view (Fig. 27A). A potential new synapomorphy of Neoceti is an anteroposteriorly elongated rostral portion of maxilla that is between 48 and 70% of the condylobasal length (minus premaxilla length) (character 2). In

62 428 E. M. G. FITZGERALD Georgiacetus and basilosaurids the rostral portion of maxilla is relatively short, comprising < 43% of the condylobasal length excluding the premaxilla. The synapomorphic state of Neoceti is presumably related to the decrease in length of the cranium and concomitant increase in the length of the maxilla, and the increase in its proportional contribution to the total rostral length as the incisor-bearing portion of the premaxilla is foreshortened (cf. Basilosauridae and toothed mysticetes in Fig. 2). Mysticeti A moderately well-supported (bootstrap = 67%) monophyletic Mysticeti occurs in the most parsimonious tree, and is diagnosed by five unequivocal synapomorphies: character 10 (0 1) steep face on the anterior edge of the antorbital process of maxilla clearly separating it from the rostral portion of maxilla, character 22 (0 1) no entocingulum on the upper cheek teeth, character 62 (0 1) anteriormost edge of the supraoccipital is at a level anterior to the anterior margin of the squamosal fossa, character 76 (0 3) transversely thickened basioccipital crest, and character 104 (1 0) vestibular foramen is in a common recess (the area cochleae) with the cochlear foramen, a well-developed transverse crest separates the area cochleae from the internal foramen of the facial canal. Note that this is a more robustly supported Mysticeti than recovered in the analyses of Geisler & Sanders (2003) (branch support = 2) and Fitzgerald (2006) (bootstrap 50%; branch support = 1), and is about as well supported as in the work of Deméré et al. (2008) (some ordered characters/morphological data only: bootstrap = 74%; branch support = 2). Mysticeti includes a monophyletic Llanocetidae, Mammalodontidae, Aetiocetidae, and Chaeomysticeti (Fig. 41). Within Mysticeti, the sequence of clade divergence hypothesized here is broadly similar to that proposed by Geisler & Sanders (2003) and Fitzgerald (2006) (Fig. 42A, B, respectively), with the undescribed toothed mysticetes from South Carolina forming a basal clade, and aetiocetids positioned proximal to Chaeomysticeti. Further, resolution of monophyletic Chaeomysticeti crownward to a paraphyletic grouping of toothed mysticetes mirrors previous results (Geisler & Sanders, 2003; Fitzgerald, 2006; Steeman, 2007; Deméré et al., 2008) (Fig. 42). These studies, along with the present results, render the concept of a suprafamilial grouping of toothed mysticetes ( Aetiocetoidea : Llanocetidae + Mammalodontidae + Aetiocetidae; Sanders & Barnes, 2002a) as paraphyletic. Otherwise, the hypothesis of basal mysticete interrelationships presented here is novel, and merits further discussion (see below). Mitchell (1989), Geisler & Sanders (2003), Fitzgerald (2006), and Deméré et al. (2008) showed that most traditionally recognized mysticete synapomorphies are either Basilosauridae + Neoceti and/or Neoceti symplesiomorphies; or, with the inclusion of toothed archaic mysticetes in phylogenetic analyses, are applicable to a less inclusive clade (i.e. Chaeomysticeti). The results of this analysis support the identification by previous studies of two synapomorphies of Mysticeti: character 10 (Geisler & Sanders, 2003; Fitzgerald, 2006), and character 76 (Sanders & Barnes, 2002a; Geisler & Sanders, 2003; Deméré et al., 2005; Bouetel & Muizon, 2006; Fitzgerald, 2006). Several mysticete synapomorphies have been proposed by Bouetel & Muizon (2006: 374): (1) ventral and dorsal surfaces of the anterior half of the rostrum form a < 45 angle in cross-section; (2) posterior edge of the ascending process of premaxilla is level with the posterior edge of the maxilla; (3) open mesorostral groove (i.e. premaxillae separated anterior to the external bony nares); (4) anterior edges of the preorbital processes of frontal level with the nasals; (5) elongated and digitiform postorbital process of frontal; (6) deeply notched supraorbital margin of frontal; (7) viewed laterally, the postglenoid process of squamosal tapers to a point; (8) external foramen ovale perforates the squamosal only; (9) apex of the anterior process of periotic is deflected ventrally; (10) posterior process of the periotic is made of spongy bone; (11) lateral tuberosity present; (12) suprameatal fossa on periotic is concave and the tegmen tympani (superior process) is elevated and thick; (13) prominent caudal tympanic process forms a sharp crest; (14) nonsutural, fibrocartilaginous mandibular symphysis; (15) mandible has a constant height along its entire length; (16) notch for insertion of the internal pterygoid muscle is only present on the medial side of the mandible; (17) opening of the mandibular canal (= mandibular foramen) is approximately level with the apex of the coronoid process; and (18) the mandibular condyle is transversely narrow. The increased taxonomic sampling of this analysis over that in Bouetel & Muizon (2006) shows that of those authors 18 characters (2), (3), (4), (8), and (11) represent symplesiomorphies of Neoceti; and characters (1), (5), (6), (7), (9), (10), and (12) to (18) represent potential synapomorphies of less inclusive clades within Mysticeti. Three novel mysticete synapomorphies are presented here. The upper cheek teeth of L. denticrenatus, J. hunderi, Ae. polydentatus, Ae. cotylalveus, and the South Carolina toothed mysticetes lack a cingulum on the lingual side of the crown base (character 22), whereas this feature is present in Georgiacetus, some basilosaurids, and the odontocete W. maerewhenua. Most mysticetes included in this study possess a supraoccipital shield that is extended forward to a level anterior to the anterior edge of the squamosal fossa (character 62). Although not hitherto explicitly identified as a mysticete synapomorphy,

63 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 429 Sus s. Sus s. Hippopotamus a. Georgiacetus v. Zygorhiza k. Hippopotamus a. Georgiacetus v. Zygorhiza k. Odontoceti Odontoceti ChM PV5720 ChM PV5720 ChM PV2778 ChM PV2778 Chonecetus g. Janjucetus h. Aetiocetus c. Mammalodon c. A Chaeomysticeti B Chonecetus g. Aetiocetus c. Chaeomysticeti Zygorhiza k. Dorudon a. Dorudon a. Aetiocetus c. Chonecetus g. Chonecetus g. Aetiocetus c. Llanocetus d. Chaeomysticeti Chaeomysticeti C D Zygorhiza k. Odontoceti Janjucetus h. Mammalodon c. Chonecetus g. Zygorhiza k. Dorudon a. Odontoceti Janjucetus h. Mammalodon c. Morawanocetus y. Aetiocetus p. Chonecetus g. E Aetiocetus c. Aetiocetus w. Chaeomysticeti F Aetiocetus c. Aetiocetus p. Aetiocetus w. Chaeomysticeti Figure 42. Previous phylogenetic hypotheses of toothed mysticete relationships: A, Geisler & Sanders (2003); B, Fitzgerald (2006); C, Bouetel & Muizon (2006); D, Steeman (2007); E, Deméré et al. (2008); and F, Deméré & Berta (2008). Phylogenetic trees (A), (B), (C), (D) and (F) are based on morphologic data, whereas (E) is based on morphologic and molecular data. Taxa in bold represent toothed archaic mysticetes. For abbreviations see Material and methods. anterior thrust of the supraoccipital represents the primary component of cranial telescoping, proposed by Miller (1923) as diagnostic of Mysticeti (also see Barnes & McLeod, 1984). Mammalodontidae and Chonecetus godertorum lack cranial telescoping, which, according to this analysis represents secondary reversal to the plesiomorphic condition of stem Cetacea. Amongst Cetacea, mysticete periotics exhibit a unique condition of the internal acoustic meatus where the vestibular foramen and cochlear foramen open in the fundus of the area cochleae, which is separated from the internal foramen of the facial canal by a thick, internally projecting, transverse crest (character 104) (Fig. 23C, D). In stem Cetacea and odontocetes the foramina of the internal acoustic meatus (including the internal foramen of the facial canal) open within a common recess, the transverse crest being low (e.g. Bas. isis: Fig. 27C, E).

64 430 E. M. G. FITZGERALD South Carolina toothed mysticetes Previous analyses have posited the undescribed toothed mysticetes from South Carolina (hereafter abbreviated to SCTM) as the most basal-branching clade within Mysticeti (Geisler & Sanders, 2003; Fitzgerald, 2006). Despite further addition of toothed mysticete taxa in this analysis (including the geologically earliest mysticete L. denticrenatus), the SCTM are recovered as a strongly supported (bootstrap = 95%) basalmost clade within Mysticeti, supporting the findings of earlier studies. The SCTM clade forms the sister to a clade consisting of all described toothed mysticetes + Chaeomysticeti (see below). Hence, this implies that the SCTM indeed represent the most archaic known mysticetes, and are an ancient lineage, diverging from all other mysticetes in the Late Eocene (> 34 Mya). Interestingly, especially in light of their basal position within Mysticeti, the SCTM are highly disparate from both basilosaurids and more crownward toothed mysticetes. Indeed, it remains possible that future phylogenetic analyses may find that the SCTM lie outside Mysticeti and perhaps outside Neoceti. In that instance, the apparent mysticete characters of the SCTM would be homoplasies. Unfortunately, the SCTM remain undescribed and unnamed. Future description and careful analysis of the SCTM clade will be essential to our understanding of the odontocete/mysticete divergence, and the earliest evolution of Neoceti. Monophyly of described toothed mysticetes + Chaeomysticeti The three described families of toothed archaic mysticetes (Llanocetidae, Mammalodontidae, and Aetiocetidae) and edentulous baleen-bearing mysticetes (Chaeomysticeti) form a weakly supported (bootstrap 50%) clade. A clade consisting of Llanocetidae + Mammalodontidae, and Aetiocetidae are positioned as successive sister groups to Chaeomysticeti (Fig. 41). The described toothed archaic mysticetes + Chaeomysticeti clade is diagnosed by three unequivocal synapomorphies: character 4 (0 1) dorsoventrally thin lateral margin of the maxilla, character 6 (0 1) premaxilla adjacent and anterior to the nasal opening widens anteriorly, and character 18 (0 1) wide diastemata between posterior cheek teeth. A dorsoventrally thin lateral margin of the maxilla has previously been submitted as a synapomorphy of Mysticeti (Barnes & McLeod, 1984; McLeod et al., 1993; Sanders & Barnes, 2002a; Deméré et al., 2005). As noted by Geisler & Sanders (2003: 69) the SCTM (ChM PV5720 and ChM PV4745) have maxillae with thick lateral margins, suggesting that character 4 is a synapomorphy of a clade within Mysticeti. This hypothesis was supported in the analyses of Fitzgerald (2006) and Deméré et al. (2008), and is further substantiated by the results of this study. Note that Fitzgerald (2006) hypothesized character 4 to be a synapomorphy of aetiocetids + Chaeomysticeti, whereas the analysis of Deméré et al. (2008) optimized this character as synapomorphic for Ma. colliveri and all more crownward mysticetes. Here, Llanocetus, Mammalodon, and Janjucetus, in addition to Aetiocetidae + Chaeomysticeti, are interpreted as possessing maxillae with dorsoventrally thin lateral margins. A nonsutured, fibrocartilaginous, mandibular symphysis was proposed by Fitzgerald (2006) and Deméré et al. (2008) as a synapomorphy of a clade equivalent to the Llanocetidae + Mammalodontidae, Aetiocetidae, and Chaeomysticeti group recovered in this analysis. Previous coding of the mandibular symphysis as being nonsutured in Janjucetus and Mammalodon was based on inaccurate interpretation of anatomy by Fitzgerald (2006). It is now clear that the mandibular symphysis is not preserved in the holotype of J. hunderi, and as demonstrated here (see Morphological description ) the condition of the mandibular symphysis in Ma. colliveri is ambiguous. As a result of uncertainty over the symphyseal state in Llanocetus, Mammalodon, and Janjucetus, the synapomorphic condition of the mandibular symphysis for this clade is equivocally optimized. This uncertainty notwithstanding, a nonsutured mandibular symphysis is optimized as present in the common ancestor of Aetiocetidae + Chaeomysticeti, although this character is not interpreted as synapomorphic for the latter groups. The two remaining synapomorphies of this clade are new. On the rostra of Basilosauridae (Fig. 2), basal odontocetes (e.g. S. rayi and W. maerewhenua: Fordyce, 1994a, 2002a), and the SCTM, the premaxilla adjacent and anterior to the level of the nasal opening tapers or has a consistent width towards the apex of the rostrum. Janjucetus, Mammalodon, aetiocetids (Fig. 2), and basal Chaeomysticeti, possess premaxillae that are transversely expanded towards the rostral apex (character 6). Basilosauridae and the SCTM possess posterior upper cheek teeth (P3 M2 and P3 M3, respectively) that are closely spaced along the tooth row, lacking a diastema between each tooth (Fig. 2; Kellogg, 1936; Uhen, 2004). Llanocetidae, Mammalodontidae, and Aetiocetidae have broad diastemata between the posterior upper cheek teeth (character 18). Monophyly of southern hemisphere toothed mysticetes A novel result of this analysis is the recovery of a clade consisting of Llanocetidae (Llanocetus denticrenatus) + Mammalodontidae (Mammalodon colliveri + Janjucetus hunderi), posited as the sister group to Aetiocetidae + Chaeomysticeti (Fig. 41). Monophyly of the southern

65 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 431 hemisphere toothed mysticetes received a low bootstrap value (< 50% to 50%), but is supported by three unequivocal synapomorphies: character 5 (1 0) broad-based rostrum (rostral width at base is > 92% of the width across orbits); character 35 (2 3) medially positioned anteriormost point on the posterior edge of the supraorbital process of frontal; and character 76 (3 2) ventrolateral edge of the basioccipital crest forms a bulbous prominence projecting laterally in an axis parallel with the external acoustic meatus. See below for discussion of Mammalodontidae. The relationships of Llanocetus within Mysticeti have been unclear for some time. This is a reflection of Mitchell s (1989) original description of fragmentary material, and lack of description of the remainder of the holotype, which includes a relatively complete skull (Fordyce, 2003a, b). A complete description of L. denticrenatus is still awaited. Geisler & Sanders (2003: 69) implied that L. denticrenatus represents the same grade of mysticetes as the SCTM, based on the supposed similarity of their relatively large basilosauridlike teeth. This analysis refutes their hypothesis, instead suggesting that Llanocetus is a relatively specialized clade, and occupies a more crownward position within Mysticeti than the SCTM. Llanocetus denticrenatus has hitherto been included in one cladistic analysis: Steeman s (2007) hypothesis posited L. denticrenatus as the sister group of Chaeomysticeti, with a monophyletic Aetiocetidae stemward of L. denticrenatus (Fig. 44D). Steeman (2007: 879) cited two characters in support of a sister group relationship between Llanocetus and Chaeomysticeti: (1) viewed laterally, the premaxillae just anterior to the nasals have a straight or convex dorsal profile (character 47, state 1, of this study); and (2) occipital condyles are level with the posterior edge of the cranium and not projecting posteriorly on a swelling (this character was not included in the present analysis, but it appears to be correlated, if not equivalent, with character 78, state 1 or 2, of this study). State 1 of character 47 is present in Llanocetus, Janjucetus, Eomysticetus, and more crownward chaeomysticetes, and is equivocally optimized at the base of the described toothed mysticetes + Chaeomysticeti clade. State 1 and/or 2 of character 78 is present in SCTM, Llanocetus, Aetiocetus spp., and Chaeomysticeti, and according to this analysis evolved independently in each of these clades (under both ACCTRAN and DELTRAN character state optimization). ConstrainingLlanocetus to be the sister group of Chaeomysticeti adds four extra steps to the length of the MPT. Mammalodontidae Mammalodontidae has thus far been considered monophyletic and monotypic since Mitchell (1989: 2231) erected the family. Two published cladistic analyses including Ma. colliveri vindicated this concept (Fitzgerald, 2006; Deméré et al., 2008). Nonetheless, Fitzgerald (2006) noted morphological similarities between Mammalodon and J. hunderi. Janjucetus hunderi was placed in a distinct monophyletic and monotypic family Janjucetidae (Fitzgerald, 2006), which the cladistic analyses of Fitzgerald (2006) and Deméré et al. (2008) supported. The present study identified numerous resemblances in skull morphology between Mammalodon and Janjucetus that the present cladistic analysis has shown to be synapomorphies. A clade including Mammalodon and Janjucetus, positioned as the sister group to Llanocetus, was recovered in the most parsimonious tree of this analysis. The Mammalodon + Janjucetus clade is moderately well supported (bootstrap = 60%) and is diagnosed by eight unequivocal synapomorphies (Fig. 43): character 2 (1 0) short rostral portion of the maxilla, which is < 43% of condylobasal length minus the premaxillae; character 33 (0 1) preorbital process of the frontal has a linguiform outline with a roundedoff anterior edge; character 45 (0 1) a triangular wedge of frontal separates the posteromedial edge of the ascending process of maxilla from the posterolateral margin of the nasal; character 55 (0 1) viewed laterally, the dorsal edge of the braincase is low to flat, its dorsal profile at an angle of < 10 to the lateral edge of the rostrum; character 60 (0 1) V-shaped frontoparietal suture on dorsal surface of skull; character 62 (1 0) skull lacks cranial telescoping, with the anterior edge of the supraoccipital at a level posterior to the anterior edge of the squamosal fossa; character 68 (1 0) anterior edge of the supraoccipital has a semicircular outline; and character 123 (1 2) posterior upper cheek teeth are double rooted, with the roots being joined by an isthmus for part or all of their length. The distinctive suite of synapomorphies in Mammalodon and Janjucetus supporting their sister group relationship (Fig. 43), as well as other morphological similarities (see Comparisons below), warrants placement of J. hunderi within the family Mammalodontidae. This action makes Janjucetidae a junior synonym of Mammalodontidae, therefore expanding Mammalodontidae beyond Mitchell s (1989) original definition. Synapomorphies of Mammalodontidae sensu lato, and Janjucetidae, identified by Fitzgerald (2006) represent features diagnostic at the genus level (see below). Enforcing a topological constraint of mammalodontid paraphyly, with Mammalodon and Janjucetus as successive sister groups to Aetiocetidae (as proposed by Fitzgerald, 2006 and Deméré et al., 2008) (Fig. 42B, E, respectively), adds eight extra steps to the length of the MPT. Within Mammalodontidae, Ma. colliveri is diagnosed by five unequivocal autapomorphies: character

66 432 E. M. G. FITZGERALD A V-shaped frontoparietal suture (60) triangular wedge of frontal separates the posteromedial edge of maxilla from the posterolateral edge of nasal (45) linguiform preorbital process of frontal (33) B V-shaped frontoparietal suture (60) triangular wedge of frontal separates the posteromedial edge of maxilla from the posterolateral edge of nasal (45) linguiform preorbital process of frontal (33) short rostral portion of maxilla (2) semicircular supraoccipital (68) short rostral portion of maxilla (2) semicircular supraoccipital (68) supraoccipital not telescoped (62) supraoccipital not telescoped (62) C low-profile braincase (55) D low-profile braincase (55) Figure 43. Skulls of Mammalodon colliveri (A, C) and Janjucetus hunderi (B, D) in dorsal (top row) and right lateral (bottom row) views, illustrating synapomorphies of Mammalodontidae. Numbers in parentheses designate synapomorphic characters from the character list (Appendix 2). Skulls are scaled to the same length. Dashed lines represent reconstructed regions of cranial anatomy. 35 (3 1) laterally positioned anteriormost point on the posterior edge of the supraorbital process of frontal; character 43 (1 2) ascending process of the maxilla is slender and linguiform; character 44 (0 1) posterior edge of the ascending process of maxilla in transverse line with the posterior edge of nasal; character 73 (1 0) alisphenoid extensively exposed in the pterygoid sinus fossa, with the superior lamina of the pterygoid limited to the anteromedial corner of the sinus fossa; and character 117 (1 0) involucrum of the tympanic bulla bears a prominent transverse groove on its dorsal surface that divides it into a thicker posterior part and thinner anterior part. Deméré & Berta (2008: 343) suggested that a narrow intertemporal region (< 70% width across the occipital condyles) is an autapomorphy of Ma. colliveri. Janjucetus hunderi is diagnosed by eight unequivocal autapomorphies: character 36 (0 1) frontal lacks a postorbital ridge; character 38 (2 1) two dorsal infraorbital foramina; character 41 (0 1) posterior end of the ascending process of premaxilla is in line with the anterior half of the supraorbital process of frontal; character 50 (0 1) posterior end of the ascending process of premaxilla faces anteriorly; character 52 (2 1) lateral edges of nasals converge anteriorly; character 58 (0 1) lateral end of the orbitotemporal crest lies on the dorsal surface of the supraorbital process of frontal; character 79 (0 1) apex of the anterior process of periotic bears a tubercle; and character 114 (0 1) the accessory ossicle or its homologue on the outer lip of the tympanic bulla is fused to the anterior process of the periotic. Chaeomysticeti + Aetiocetidae Prior phylogenetic analyses have proposed monophyletic Aetiocetidae as the sister group to Chaeomysticeti (Geisler & Sanders, 2003; Deméré et al., 2008) (Fig. 42A, E) or have positioned the aetiocetids Aetio-

67 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 433 cetus and Chonecetus as successive sister taxa to Chaeomysticeti (Bouetel & Muizon, 2006; Fitzgerald, 2006) (Fig. 42B, C), Aetiocetidae being paraphyletic in the latter instance. A third hypothesis, formulated by Steeman (2007) (Fig. 42D), sees a monophyletic Aetiocetidae sister to a Llanocetus + Chaeomysticeti clade (see discussion above). The results of this cladistic analysis substantiate: (1) a sister group relationship between Aetiocetidae and Chaeomysticeti; and (2) the monophyly of Aetiocetidae (see below for further discussion of aetiocetids). The clade Chaeomysticeti + Aetiocetidae is comparatively weakly supported (bootstrap 50%), and is diagnosed by four unequivocal synapomorphies: character 14 (0 1) lateral nutrient foramina with associated sulci on the palatal surface of maxilla; character 109 (1 2) short posterior process of the periotic; character 116 (0 1) median furrow of tympanic bulla reduced to a notch on the posterior edge of the bulla; and character 120 (0 1) alveolar margin on the mandible is parallel to the ventral margin of the mandible for its entire length (Fig. 34C E). Character 116 was proposed by Fitzgerald (2006) as a synapomorphy of aetiocetids + Chaeomysticeti. It is noteworthy that this clade also received relatively weak support in earlier analyses: bootstrap = 57% and branch support = 2 in Fitzgerald (2006); and bootstrap = 50% and branch support = 1 (some ordered characters/morphological data only) in Deméré et al. (2008). Chaeomysticeti represents a strongly supported clade, with a bootstrap value of 88%, and is diagnosed by four unequivocal synapomorphies: character 17 (1 0) edentulous rostrum; character 43 (1 0) maxilla lacks a distinct ascending process; character 56 (0 1) frontals are at the same height as the nasals; and character 58 (0 2) entire length of the orbitotemporal crest is on the dorsal surface of the supraorbital process of frontal. Character 14 (lateral nutrient foramina with associated sulci on the palatal surface of maxilla) requires further comment. This character state is correlated with the possession of baleen in crown group Mysticeti, and it is inferred that fossil mysticetes possessing lateral nutrient foramina that open into sulci thus possessed baleen in some form (Fitzgerald, 2006; Deméré et al., 2008). This assumes that the lateral nutrient foramina and sulci possessed by fossil mysticetes are homologous with those in Recent mysticetes: Deméré et al. (2008) have argued that this is indeed the case. Amongst aetiocetids, Deméré et al. (2008) identified character 14 in Ae. cotylalveus, Ae. weltoni, and C. goedertorum. I concur with these authors that character 14 is present in Ae. weltoni, and in Ae. cotylalveus. Note that Fordyce (1982b: 424) also recognized longitudinal grooves on the palate of Ae. cotylalveus (USNM 25210, holotype), which he suggested were homologous with the sulci on the palate of crown Mysticeti. Despite the claims of Berta, Sumich & Kovacs (2006: 62) and Deméré et al. (2008: 21 22), I have been unable to identify unequivocal lateral nutrient foramina/sulci on the palatal surface of maxilla in the holotype specimen (LACM ) of C. goedertorum. This does not rule out C. goedertorum possessing character 14, but the holotype LACM appears to lack this character or, at best, its preservation in this specimen is equivocal. I agree with Deméré et al. (2008: 22) that the supposedly vascular grooves on the palate of L. denticrenatus, as identified by Fordyce (2003b), are uncertainly homologous with the lateral nutrient foramina/sulci of Aetiocetus and Chaeomysticeti. Llanocetus is therefore coded as uncertain (?) for character 14. Pending the positive identification of character 14 in Llanocetus, this analysis optimizes lateral nutrient foramina and sulci on the palate as evolving once in the common ancestor of Aetiocetidae and Chaeomysticeti. Fitzgerald (2006) suggested two potential synapomorphies of a clade equivalent to Chaeomysticeti + Aetiocetidae, which in this study are not optimized as synapomorphic: (1) nonsutured, fibrocartilaginous mandibular symphysis; and (2) posterior edge of premaxilla in line with the posterior half of the supraorbital process of the frontal. A nonsutured fibrocartilaginous mandibular symphysis, marked by a longitudinal groove on the medial side of the mandible near its apex, occurs in all Chaeomysticeti, and is preserved in C. goedertorum, Ae. polydentatus, and Ae. weltoni. Although being optimized by this analysis as present in the common ancestor of Aetiocetidae and Chaeomysticeti, this character is unknown in Llanocetus and equivocally preserved in Mammalodontidae. As a result, it remains possible that a nonsutured mandibular symphysis is either a synapomorphy of Chaeomysticeti + Aetiocetidae, or (Llanocetus + Mammalodontidae) + (Chaeomysticeti + Aetiocetidae). A posterior position for the premaxilla s termination is optimized on the MPT as being present in the common ancestor of aetiocetids and edentulous mysticetes. This character is also present in basal odontocetes and SCTM (ChM PV5720), but absent in Llanocetus and Mammalodontidae, and is therefore equivocally synapomorphic for Chaeomysticeti + Aetiocetidae. Monophyly of Aetiocetidae Monophyly of Aetiocetidae is endorsed, albeit weakly (bootstrap 50%), by this analysis in agreement with the cladistic analyses of Geisler & Sanders (2003), Steeman (2007), and Deméré et al. (2008). Aetiocetidae received low support in one recent analysis (branch support = 2: Geisler & Sanders, 2003), and stronger support in another (bootstrap = 84%; branch support = 3: Deméré et al., 2008). Aetiocetidae is diagnosed by two unequivocal synapomorphies: character

68 434 E. M. G. FITZGERALD 49 (0 1) premaxilla adjacent to and at posterior edge of the nasal opening overhangs maxilla, and character 76 (3 1) basioccipital crest is transversely thickened, with its ventrolateral edge forming a longitudinal keel. Character 49 was proposed by Barnes et al. (1995) as a diagnostic feature of Aetiocetidae, and recognized as an aetiocetid synapomorphy in the analysis of Geisler & Sanders (2003). Character 76 is a new synapomorphy of Aetiocetidae. A recent analysis of aetiocetid phylogeny (Deméré & Berta, 2008) found a well-supported (bootstrap = 70%) clade Aetiocetidae diagnosed by three unequivocal synapomorphies: (1) zygomatic process of the squamosal expanded near its anterior margin and at its posterior end but narrow in the middle (this character is essentially equivalent to character 64 of this study); (2) coronoid process of mandible is well developed and has a concave posterior margin; and (3) short overlap of jugal with zygomatic process of squamosal. In this analysis, character 1 is a synapomorphy of Aetiocetus spp. (see below). The definition and scoring of character 2 (character 27 of Deméré & Berta, 2008) is based on misinterpretation of anatomy. Deméré & Berta (2008) considered the posterior margin of the coronoid process of mandible to be concave (viewed laterally or medially) in Ae. weltoni, Ae. polydentatus, Morawanocetus yabukii, and C. goedertorum. These taxa actually possess a coronoid process with a convex to straight posterior margin. Deméré & Berta (2008) appear to be confusing the profile of the posterior margin of the coronoid process with another anatomical feature, the mandibular notch (sensu Schaller, 1992; see Fig. 31B of this study). Barnes et al. (1995) proposed a tripartite division of Aetiocetidae into the subfamilies Chonecetinae, Aetiocetinae, and Morawanocetinae. Apparent clarity of relationships within Aetiocetidae, suggested by these subfamilial divisions, belies the problematic status of these taxa and their included genera. The Morawanocetinae, which includes Mo. yabukii, was not included in this analysis because of the difficulty of coding this taxon on the basis of its published description. Morawanocetus yabukii appears to be a distinct taxon. Ashorocetus eguchii was placed by Barnes et al. (1995) in the Chonecetinae with Chonecetus, but as depicted by those authors in a cladogram (426), As. eguchii is more closely related to Morawanocetus than it is to Chonecetus spp., with Chonecetinae illustrated as paraphyletic. Whether Barnes et al. (1995) consider Chonecetinae as a clade or grade of aetiocetids is ambiguous. Regardless, the poorly preserved holotype specimen of As. eguchii (AMP 3: a partial braincase without periotics and tympanic bullae) is debatably informative enough to diagnose a distinct genus and this taxon may be a nomen dubium. A third aetiocetid taxon described by Barnes et al. (1995), Aetiocetus tomitai, was not included in this analysis. As implied by its referral to Aetiocetus, Ae. tomitai closely resembles Ae. cotylalveus (Barnes et al., 1995). Indeed, Berta & Deméré (2005) suggested that Ae. tomitai (with Ae. weltoni) is a junior synonym of Ae. cotylalveus. The phylogenetic analysis of Deméré et al. (2008) failed to resolve Aetiocetidae into the subfamilies proposed by Barnes et al. (1995), the taxa C. goedertorum, Aetiocetus polydentatus, Ae. cotylalveus, and Ae. weltoni forming a polytomy in the former authors strict consensus tree (Fig. 42E). More recently, an analysis recovered a monophyletic Morawanocetinae, Chonecetinae, and Aetiocetinae within Aetiocetidae (Deméré & Berta, 2008) (Fig. 42F). The latter study hypothesized that Mo. yabukii is the most archaic aetiocetid. The analysis of this study resulted in a reciprocally monophyletic C. sookensis + C. goedertorum group and an Ae. polydentatus + (Ae. cotylalveus + Ae. weltoni) clade (Fig. 41). This division within Aetiocetidae is equivalent to Barnes et al. s (1995) proposed Chonecetinae and Aetiocetinae, and suggests that these taxa represent clades. Nevertheless, the taxonomy and phylogenetics of Aetiocetidae requires substantial revision, and several described and undescribed taxa/specimens merit further detailed analysis. Relationships within Aetiocetidae are likely to change with inclusion of more taxa and more characters in cladistic analyses (e.g. Deméré & Berta, 2008). With these caveats in mind, the relationships amongst aetiocetids proposed here must be considered preliminary hypotheses. The clade consisting of C. sookensis and C. goedertorum received low bootstrap support (51%) and is diagnosed by three unequivocal synapomorphies: character 55 (0 1) viewed laterally, the dorsal edge of the braincase is low to flat, its dorsal profile at an angle of < 10 to the lateral edge of the rostrum; character 60 (0 1) V-shaped frontoparietal suture on dorsal surface of skull; and character 69 (1 2) very low nuchal crest of the supraoccipital that does not project laterally or dorsolaterally. Barnes et al. (1995) considered character 69 to be diagnostic of the genus Chonecetus. Deméré & Berta (2008: 343) identified character 60 (of this study) as an equivocal apomorphy of Chonecetus. According to this analysis, characters 55 and 60 evolved independently in Chonecetus and Mammalodontidae. Although Barnes et al. (1995: 398) stated that C. sookensis is the most primitive aetiocetid, its phylogenetic position in this analysis shows that it is not more basal-branching relative to other aetiocetids. The clade including Ae. polydentatus and Ae. cotylalveus + Ae. weltoni is weakly supported (boot-

69 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 435 strap 50%) and is diagnosed by three unequivocal synapomorphies: character 7 (4 3) mesorostral groove is not completely open for its entire length, being partially roofed by the premaxillae; character 46 (1 2) anterior edge of the nasal is at a level anterior to the anterior edge of the supraorbital process of frontal; and character 64 (0 1) maximum dorsoventral thickness of the zygomatic process of squamosal is within the anterior half of its length. Within this clade, a subclade of Ae. cotylalveus + Ae. weltoni is moderately well supported (bootstrap = 61%), and is diagnosed by four unequivocal synapomorphies: character 5 (1 0) broad-based rostrum, width across rostral base is > 92% of the width across the middle of orbits; character 44 (0 1) posterior edge of the ascending process of maxilla is in line with, or just posterior to, the posterior edge of the nasal; character 77 (1 0) viewed ventrally, the basioccipital crests are parallel to one another with no angle formed between them; and character 78 (0 1) posterior edge of the paroccipital process of the exoccipital is in transverse line with the posterior edge of the occipital condyle. Five unequivocal autapomorphies diagnose Ae. polydentatus: character 17 (2 3) ten or 11 teeth in the maxilla; character 25 (5 6) 14 to 15 teeth in the mandible; character 52 (2 0) nasals anterior to the level of the preorbital process of frontal have parallel lateral edges; character 58 (0 2) orbitotemporal crest is on the dorsal surface of the supraorbital process of frontal; and character 66 (1 0) transversely narrow exposure of squamosal lateral to exoccipital in posterior view. In their original description of Ae. polydentatus, Barnes et al. (1995) noted that this taxon was the most specialized species referred to the genus Aetiocetus, and that its combination of features clearly distinguished it from Ae. cotylalveus, Ae. weltoni, and Ae. tomitai. Ichishima (2005) and Fitzgerald (2006) suggested that the morphological differences between Ae. polydentatus and other Aetiocetus species were sufficient to distinguish Ae. polydentatus at the level of genus. This cladistic analysis finds that Ae. cotylalveus and Ae. weltoni are more closely related to one another than either is to Ae. polydentatus. In contrast, the phylogenetic hypothesis of Deméré & Berta (2008) places Ae. polydentatus as sister to Ae. weltoni on the basis of two equivocal synapomorphies: a narrow (< 70% width across occipital condyles) intertemporal region and a reduced sagittal crest. Pending a thorough redescription and review of Ae. polydentatus, both its taxonomy and phylogenetic position remain debatable. COMPARISONS The results of this morphological description and phylogenetic analysis demonstrate that Ma. colliveri is an archaic stem mysticete, which nevertheless displays a number of specialized characters. Some of these derived characters are shared with basal odontocetes, other toothed archaic mysticetes, and also more crownward mysticetes, whereas others are unique to Mammalodon. Conversely, Mammalodon retains numerous primitive features typical of stem Cetacea. Morphological comparisons between Mammalodon and specific cetacean taxa are made below. Mammalodon is a mysticete Mammalodon colliveri possesses features of its dentition and cranial osteology that are synapomorphies of Mysticeti, including upper cheek teeth without an entocingulum, a steep anterior edge on the laterally projecting antorbital process of maxilla, transversely thickened basioccipital crests, and a periotic with the internal foramen of the facial canal separated from the other foramina of the internal acoustic meatus by a salient transverse crest. On the rostrum, the premaxillae increase in width towards their apex, and the lateral margin of the maxilla is dorsoventrally thin, both derived features of mysticetes more crownward than the SCTM. In odontocetes and archaeocetes the premaxillae have a constant width or taper towards the rostral apex, and the maxilla is thick-edged. As in other Mysticeti, the maxilla of Ma. colliveri bears an infraorbital process that extends posteriorly, ventral to the supraorbital process of the frontal. Unlike other mysticetes, Ma. colliveri has an infraorbital process of the maxilla that is tooth bearing. Fordyce (1982a: 47) was the first worker to suggest that Ma. colliveri was an archaic mysticete, and cited the following features in support of mysticete affinities: (1) loosely sutured rostral bones, (2) relatively broad and flat palate, and (3) lack of a sutured mandibular symphysis. Moreover, he highlighted the lack of a key odontocete synapomorphy (ascending process of the maxilla that is broadened posteriorly and laterally to overlap the dorsal surface of the supraorbital process of frontal), as evidence against odontocete affinities. In addition to the characters of Mammalodon listed above, Fordyce (1984: 942) noted the (4) thin-edged maxilla, and (5) dorsally placed attachment of the temporalis muscle on the supraorbital process of frontal, as features shared with mysticetes, but not with archaeocetes or odontocetes. This study supports the hypothesis that Mammalodon is an archaic mysticete. Nonetheless, it indicates that of Fordyce s characters, only character (4) is an unequivocal synapomorphy of Mysticeti, or at least a subclade of mysticetes. Characters (1) and (2) are not demonstrated mysticete synapomorphies, and character (3) is ambiguously preserved in the holotype of Ma. colliveri. Although character (5) as described by Fordyce (1984: 942) is present in Ma. colliveri, the

70 436 E. M. G. FITZGERALD orbitotemporal crest is located on the posterior edge of the supraorbital process, not on its dorsal surface. At this point, it is pertinent to consider briefly the comments on Ma. colliveri by Milinkovitch (1997: ). It seems that Milinkovitch was attempting to argue that the mysticete affinities of Mammalodon are equivocal. Milinkovitch (1997: 331) implied that Fordyce (1984) based his interpretation of Ma. colliveri as a mysticete solely on the basis of its apparent lack of odontocete and archaeocete characters. The preceding paragraph shows that Fordyce identified features of Mammalodon that were potential mysticete synapomorphies. Contra Milinkovitch (1997), Fordyce (1984: 942) noted that Mammalodon possesses some features of both odontocetes and archaeocetes, but that those features represent symplesiomorphies (see below). Comparisons of Mammalodon and Basilosauridae For much of the last 70 years since its description in 1939, Ma. colliveri was considered an archaeocete whale (Anonymous, 1939; Camp et al., 1949; Romer, 1966; Fordyce, 1978, 1980; Sanders et al., 1982). This evinces the brevity of Pritchard s (1939) original description, but also the generally primitive morphology of Mammalodon, and relative subtlety of its mysticete characters. As shown by Fordyce (1984: 942), the anatomy of Mammalodon is similar to that of basilosaurid archaeocetes in the following respects: well-developed and functional heterodont teeth, bony external nares open at a level well anterior to the antorbital notch, elongated and dorsoventrally thin nasals, parietals extensively exposed on the dorsal surface of the cranium in an elongated intertemporal constriction, braincase not inflated, supraoccipital is not thrust forward anteriorly (i.e. no cranial telescoping), high and long coronoid process on the mandible, and relatively small skull size. Other morphological similarities (symplesiomorphies) between Mammalodon and basilosaurids include: a short rostral portion of the maxilla (as a proportion of condylobasal length); the maxilla is not dorsoventrally flattened; the premaxillae are abruptly depressed just anterior to the nasals; the posteriormost upper tooth is at a level posterior to the antorbital notch; the ascending process of the premaxilla terminates at a level anterior to the antorbital notch; the optic infundibulum is anteroposteriorly elongated, with the optic foramen and orbital fissure being widely separated; the orbitotemporal crest is located along the posterior margin of the supraorbital process of frontal and continues posteriorly onto the dorsolateral edge of the intertemporal constriction; a semicircular anterior edge of the supraoccipital; the alisphenoid is extensively exposed in the pterygoid sinus fossa; a short anterior process of the periotic; a well-developed median furrow on the ventral surface of the tympanic bulla; the posterior half of the alveolar margin of the mandible is at an angle to the ventral margin of the mandible; a large mandibular foramen and associated pan bone forming its lateral wall; and a posteriorly directed mandibular condyle. Other than the derived mysticete and neocete characters considered above and below, there are some salient differences in the morphology of Ma. colliveri from that of basilosaurid archaeocetes. Mammalodon colliveri is more specialized in having: one additional tooth in each quadrant, resulting in three/four molars (basilosaurids have two/three molars); a rostrum and mandible that lack embrasure pits for accommodation of the crowns of the lower and upper teeth, respectively; a rostrum that is not tubular adjacent and anterior to the level of the nasal opening; nasals that increase in width towards their anterior edge; parietals that meet on the midline of the cranium at a smoothly rounded over sagittal suture, and not an elevated sagittal crest; a periotic with a gently convex suprameatal region and not a deeply excavated suprameatal fossa; and an axis vertebra that lacks a transverse foramen (vertebrarterial canal) perforating the transverse process. Comparisons of Mammalodon and Odontoceti The phylogenetic analysis of this study supports a sister group relationship between mysticetes and odontocetes as the clade Neoceti, and as a stem mysticete Ma. colliveri shares several derived characters with archaic Odontoceti that are absent in archaeocetes. Most of the detailed comparisons made here are between Ma. colliveri and S. rayi as a representative of Odontoceti. Simocetus is the most well-known stem odontocete thanks to the detailed description provided by Fordyce (2002a), although it is not the most archaic: Archaeodelphis patrius Allen (1921) has been posited as the most basal-branching odontocete taxon (Fordyce, 1994a, 2002a), or a stemward member of a basal-branching clade including Xenorophus (Geisler & Sanders, 2003). The mesorostral groove on the rostrum of Mammalodon and early odontocetes is open for its entire length: in archaeocetes it is closed anteriorly as a result of the premaxillae contacting, or closely approximating, one another on the midline via an elongated interincisive suture. Mammalodon colliveri and Simocetus rayi possess wide diastemata between their posterior upper cheek teeth. Both Mammalodon and basal odontocetes have a well-developed, basin-like, facial fossa lateral and posterolateral to the level of the anterior edge of the nasals. This suggests well-developed maxillonasolabialis muscles in both stem odontocetes and mysticetes, although the origin of these nasofacial muscles

71 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 437 in Mammalodon seems to be limited to an anterior position, medial and anteromedial to the antorbital notch. In the odontocetes Simocetus, Agorophius, and Xenorophus, the nasofacial (including maxillonasolabialis) muscles have a broad posterior origin on the ascending process of maxilla, dorsal to the supraorbital process of the frontal, and hence appear to have been hypertrophied. Fordyce (2002a: 215) has already drawn a direct comparison between the multiple dorsal infraorbital foramina of S. rayi and Ma. colliveri. Moreover, a cluster of four or more infraorbital foramina open in the facial fossa anterior to the level of the antorbital notch, with one or two dorsal infraorbital foramina occurring posterior to this cluster, in Ma. colliveri, S. rayi, and Agorophius pygmaeus Müller, 1849 (True, 1907; Fordyce, 1981). In Ma. colliveri the single more posteriorly positioned dorsal infraorbital foramen lies just anterior to the level of the antorbital notch: in S. rayi and Ag. pygmaeus the posterior dorsal infraorbital foramina (sensu Fordyce, 1981, 2002a) open on the ascending process of maxilla where it overlaps the supraorbital process of frontal, and therefore at a level posterior to the antorbital notch. On the basicranium, the alisphenoid forms much of the roof of the pterygoid sinus fossa in Mammalodon and Simocetus. The periotics of Mammalodon and Simocetus share a number of derived features: in transverse section, the anterior process of the periotic has an oval outline; a distinct and salient lateral tuberosity occurs at the base of the anterior process, posteriorly adjacent to the anteroexternal sulcus; and the pars cochlearis has a smoothly rounded anterointernal angle. The periotic of Mammalodon is distinguished from that of Simocetus by having: an anterior process with a transversely convex ventral edge and no distinct anterior bullar facet; a saddle-shaped (anteroposteriorly) notch anterior to the mallear fossa, but no distinct fovea epitubaria for reception of the accessory ossicle; and a pars cochlearis that is only inflated on its ventromedial surface in a region medial to the fenestra ovalis, and anteromedial to the fenestra rotunda. The mandibular symphyses of Ma. colliveri and S. rayi are anteroposteriorly shortened, compared to the elongated mandibular symphysis of basilosaurid archaeocetes. As noted above, Ma. colliveri and basilosaurids possess a pan bone forming the wall of the mandibular foramen, which is a feature also shared by S. rayi. Simocetus rayi differs from Mammalodon by having a mandible that is ventrally inflated at the level of the mandibular foramen. Pledge & Rothausen (1977) implied that Ma. colliveri is a squalodontid odontocete, and Ma. colliveri does share a number of features with archaic odontocetes (such as Simocetus) that are not present in extant Mysticeti. Importantly, these morphological similarities between Mammalodon and archaic odontocetes represent shared primitive characters in common with archaeocetes, and shared derived characters of Mysticeti + Odontoceti as a whole. Mammalodon lacks critical derived characters of odontocetes (see Fordyce, 2002a; Geisler & Sanders, 2003), but possesses derived features unique to mysticetes. Thus, it is clear from the foregoing that Mammalodon is not an odontocete. Mammalodon is a mysticete, albeit an archaic one. Comparisons of Mammalodon and Chaeomysticeti Mammalodon bears few resemblances with edentulous baleen-bearing mysticetes in the clade Chaeomysticeti. The most notable similarities between Ma. colliveri and chaeomysticetes are synapomorphies of Mysticeti as a whole, and are discussed above. Other features of the rostrum and periotic are shared between Mammalodon and chaeomysticetes more crownward than Eomysticetus. The rostral and cranial bones of Mammalodon interdigitate via the articulation between the ascending process of the maxilla and frontal. This pattern of contact between the rostrum and cranium appears to be absent in all mysticetes more crownward than Eomysticetus apart from the families Balaenopteridae, Cetotheriidae sensu stricto, and Eschrichtiidae (Bouetel, 2005; Bouetel & Muizon, 2006; Steeman, 2007). Eomysticetus whitmorei, Neobalaenidae, Balaenidae, and numerous Miocene chaeomysticetes (including Diorocetus and Pelocetus), possess a subrectilinear articulation between the rostral and cranial bones (Bouetel, 2005). It is likely that the presence of interdigitated rostral and cranial bones is plesiomorphic for Mysticeti (Bouetel, 2005), if not Neoceti, as basilosaurids, stem odontocetes, most aetiocetids, and J. hunderi all exhibit interdigitation between the rostrum and cranium. Elongated and transversely narrow ascending processes of the maxillae that do not taper to a sharp point or contact one another on the midline of the skull are derived characters shared between Ma. colliveri and Balaenopteridae. Two surprisingly specialized features of the periotic of Ma. colliveri are its internally elongated pars cochlearis, and tubular fundus of the internal acoustic meatus. These are derived features of balaenopterid periotics (e.g. Geisler & Luo, 1996; Bisconti, 2001). The sternum of Ma. colliveri is composed of more than one sternebral element and is therefore essentially different to that of Recent Mysticeti, which have a single (and indeterminate) element in the sternum throughout ontogeny (True, 1904; Klima, 1978). Kellogg (1965) described multiple sternal elements from the holotype of the Miocene chaeomysticete Pelocetus calvertensis.

72 438 E. M. G. FITZGERALD Comparisons of Mammalodon and the South Carolina toothed mysticetes The phylogenetic analyses of Geisler & Sanders (2003), Fitzgerald (2006), and this study consistently support the hypothesis that the basic dichotomy within Mysticeti lies between the SCTM and all described toothed mysticete clades + Chaeomysticeti. Pending the description of these most archaic of known mysticetes, we must rely on a published abstract (Barnes & Sanders, 1996) for comparative information on the SCTM. Shared features of the SCTM and Ma. colliveri include three upper and four lower molars, and an orbitotemporal crest on the posterior edge of a broad supraorbital process of frontal. Otherwise, the SCTM are rather disparate from Mammalodon, as implied by their relative positions in mysticete phylogeny (Fig. 41). Mammalodon colliveri is more advanced than the SCTM in having relatively small upper cheek teeth that lack anteroposteriorly elongated triangular crowns, diastemata between posterior upper cheek teeth, no embrasure pits, a short and broad rostrum with a blunt apex, and relatively large and salient occipital condyles. Comparisons of Mammalodon and Llanocetidae Llanocetus denticrenatus Mitchell (1989) is the type, and thus far only, genus and species in the toothed mysticete family Llanocetidae. Although this taxon has been formally named and described in the literature, it is based on rather fragmentary material that offers little in the way of comparative anatomical data. Future description of the fairly complete holotype skull and other elements will rectify this (Fordyce, 1989b, 1992, 2003a, b). In the mean time, a portion of the body of the right mandible with two lower cheek teeth (?premolars) in place (figs 1 3 in Mitchell, 1989), an isolated left lower?premolar (fig. 4A in Fordyce, 1989b), a synopsis of cranial morphology in a published abstract (Fordyce, 2003a), and two characters mentioned by Steeman (2007: 879) represent the extent of material available for comparison with Ma. colliveri. Mammalodon and L. denticrenatus have a broadbased rostrum, which according to the cladistic analysis presented here is a synapomorphy uniting Llanocetidae and Mammalodontidae in a clade. Other features of the skull, mandible, and dentition of Mammalodon probably represent plesiomorphic similarities with L. denticrenatus. Shared primitive characters include a maxilla with a dorsoventrally thin lateral edge, large coronoid process of the mandible, large mandibular foramen, wide diastemata between posterior cheek teeth, relatively small heterodont cheek teeth, enamel on the buccal and lingual surfaces of cheek tooth crowns that bear heavy vertical ridges, lower posterior cheek teeth have two roots, and multiple sternebral elements. Mammalodon is more derived than Llanocetus in some of its characters. Mammalodon colliveri has posterior lower cheek teeth with two roots joined below the crown base by a transversely narrow isthmus, whereas the posterior lower cheek teeth of Llanocetus have two yoked roots that are widely separate immediately below the crown. The lower cheek tooth crowns of Mammalodon are more anteroposteriorly compressed than the elongate crowns of Llanocetus. Whereas Mammalodon is polydont (it has a fourth lower molar), Llanocetus appears to have not possessed additional teeth beyond the generalized placental dental formula (Fordyce, 2003a). Mammalodon has lower premolars that are closely spaced together along the tooth row, without elongated intervening diastemata. The lower premolars of Llanocetus are widely separated from each other by long diastemata, which is the primitive state exhibited in Basilosauridae (e.g. Kellogg, 1936) The mandible of Mammalodon has a salient lateral edge to the alveolar margin, such that the lower cheek teeth are implanted within an alveolar groove (Figs 32, 39). Llanocetus, as in Basilosauridae, lacks a distinct lateral edge to the alveolar margin, with the medial edge of the lower alveoli being higher than their lateral edge. Llanocetus exhibits some apparently more derived features than Mammalodon. The palatal surface of the maxilla of Llanocetus features abundant fine grooves around the alveoli apparently correlated with vascular supply to the soft tissue of the palate (Fordyce, 2003a: 50A). The palatal surface of the maxilla of Mammalodon lacks evidence of enhanced vascular supply proximal to the alveoli, instead bearing irregularly scattered tiny nutrient foramina. Anterior to the anterior margin of the nasal, the premaxilla of Llanocetus has a straight dorsal profile when viewed laterally (Steeman, 2007). Mammalodon possesses the primitive state wherein the premaxillae are abruptly depressed anterior to the nasals, having a concave dorsal profile. In Llanocetus the posterior edge of the occipital condyle lies at a level anterior to the posterior edge of the paroccipital process of exoccipital (Steeman, 2007). Mammalodon and basilosaurids present the plesiomorphic state, where the posterior edge of the occipital condyle is posterior to or in line with the paroccipital process. The mandible of Llanocetus is relatively deeper and more massively built than that of Mammalodon or Basilosauridae. In Llanocetus, the medial edge of the alveolar margin of the mandible is transversely narrow, forming a longitudinal ridge. This is in contrast to the condition in Mammalodon (and basilosaurids) where the medial edge of the alveoli is relatively low and does not form

73 a distinct ridge. Lastly, the skull of Llanocetus is relatively large (estimated CBL > 1800 mm) compared to that of Mammalodon and other toothed mysticetes, implying a total body length of 9 m or more (Fordyce, 2003a). Mitchell (1989: 2229) concluded that Ma. colliveri was not a close relative of Llanocetus. In disagreement, this study suggests a sister group relationship between Mammalodontidae and Llanocetidae. This hypothesized clade of southern hemisphere toothed mysticetes is characterized by high morphological disparity between mammalodontids and L. denticrenatus. Enforcing a topological constraint of southern hemisphere toothed mysticete paraphyly requires the addition of only one extra step to the length of the most parsimonious tree of this analysis. Interpretation of the relationship between Mammalodontidae and Llanocetus and its implications should therefore proceed with caution. THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 439 Comparisons of Mammalodon and Willungacetus Willungacetus aldingensis Pledge, 2005 from the Lower Oligocene Port Willunga Formation of South Australia is an enigmatic fossil cetacean described on the basis of a partial braincase (without periotics or tympanic bullae) (Fig. 44) with a possible fragment of maxilla or mandible (SAM P40034), and a referred vertebral body (SAM P40044). Despite its apparent lack of synapomorphies, Willungacetus was tentatively assigned by Pledge (2005: 124) to the Aetiocetidae. Willungacetus aldingensis was briefly compared by Pledge (2005: 130) to Ma. colliveri. Pledge noted that the supraoccipital of Willungacetus is fundamentally different to that of Mammalodon: the supraoccipital of Willungacetus has a sharply triangular outline, its nuchal crests forming an angle of almost 90 at the apex of the supraoccipital, whereas in Mammalodon the supraoccipital has a semicircular outline. Further to Pledge s observations, the supraoccipital of Willungacetus is thrust forward, its apex at a level anterior to the anterior edge of the squamosal fossa (Fig. 44A). Thus, the skull of Willungacetus exhibits occipital telescoping. Mammalodon (and Janjucetus) have secondarily reduced/absent occipital telescoping, the anterior edge of the supraoccipital at a level posterior to the anterior edge of the squamosal fossa. Mammalodon further differs from Willungacetus by having a prominent orbitotemporal crest that extends posteriorly onto the dorsolateral surface of the intertemporal constriction (Figs 6, 7). In Willungacetus there is no evidence of an orbitotemporal crest on the intertemporal constriction. The basicranium of Willungacetus is poorly preserved (Fig. 44B) and lacks anatomical detail that would facilitate a meaningful comparison with the corresponding region in Mammalodon. Figure 44. Willungacetus aldingensis, SAMP40034, holotype skull: A, dorsal; and B, ventral views. Pledge (2005: 129) admitted in his description of Willungacetus that there are not many characters preserved to enable subordinal assignment. Nevertheless, W. aldingensis preserves one unequivocal mysticete synapomorphy, as identified in this analysis: the anteriormost edge of the supraoccipital is at a level anterior to the anterior margin of the squamosal fossa (i.e. the cranium exhibits occipital telescoping). Similarities between Willungacetus and odontocetes, and archaeocetes, represent symplesiomorphies (Pledge, 2005). These data corroborate Pledge s hypothesis that Willungacetus is a mysticete. The phylogenetic position of Willungacetus within Mysticeti remains problematic. Contrary to Pledge s (2005: 130) suggestion that there is a slight possibility that, considering its even

74 440 E. M. G. FITZGERALD more primitive looking cranium, Willungacetus is ancestral to the rather aberrant Mammalodon, there is no evidence supporting a close relationship (sister group, ancestor descendant, or otherwise) between these genera. Furthermore, Mammalodon clearly has a more plesiomorphic grade of cranial morphology than Willungacetus. From what is preserved of the skull of W. aldingensis, it lacks two unequivocal mammalodontid synapomorphies: (1) a cranium without occipital telescoping, and (2) a semicircular supraoccipital shield (Fig. 43). Willungacetus was referred to the Aetiocetidae by Pledge (2005) seemingly on the basis of shared primitive characters, and a few debatable similarities between it and species of Chonecetus (C. sookensis and C. goedertorum). Pledge (2005: 130) compared the shape of the cranium, extent of the parietals, conical intertemporal constriction (presumably referring to the intertemporal region having a conical shape in cross-section), and triangular shape of the supraoccipital in Willungacetus with those characters of C. sookensis, and to a lesser extent C. goedertorum. Although Willungacetus and Chonecetus may share these characteristics, they do not seem particularly revealing with respect to phylogenetic position as they are probably plesiomorphic for basal Mysticeti. None of the synapomorphies of Aetiocetidae or taxa within Aetiocetidae are unambiguously preserved in the holotype of Willungacetus. On the basis of available data, it is likely that Willungacetus represents a basal mysticete, but beyond this there is no evidence supporting relationships with any particular mysticete clade. Mammalodon is clearly distinct from Willungacetus, and the latter genus is probably not a mammalodontid. Indeed, those features preserved in the holotype skull of W. aldingensis suggest more derived morphology than that of Mammalodontidae. Additionally, and contra to the conclusions of Pledge (2005), Willungacetus lacks synapomorphies suggesting a relationship with aetiocetids. Future discovery of more complete cranial material, which can be confidently referred to Willungacetus, may elucidate a close relationship between this taxon and Aetiocetidae. Until then, W. aldingensis is most appropriately classified as Mysticeti incertae sedis. Comparisons of Mammalodon and Aetiocetidae The Aetiocetidae are a relatively well-known family of toothed mysticetes and have been considered in some detail within the literature (Emlong, 1966; Barnes et al., 1995; Fordyce & Muizon, 2001; Deméré & Berta, 2008; Deméré et al., 2008). The relationship of Ma. colliveri to Aetiocetidae has until recently been unknown. Fordyce (1982a: 47) suggested that Mammalodon is a relict mysticete even more primitive than Aetiocetus cotylalveus, and that its morphology is intermediate between that of dorudontine archaeocetes... and described pre-cetothere grade mysticetes, such as Aetiocetus cotylalveus (Fordyce, 1984: ). Subsequent authors have stressed uncertainty over possible relationships of Ma. colliveri to aetiocetids (Fordyce, 1992; McLeod et al., 1993; Fordyce & Barnes, 1994; Barnes et al., 1995; Fordyce & Muizon, 2001). Mammalodon colliveri shares numerous plesiomorphic characters of the skull, mandible, and dentition with Aetiocetidae, including: a small skull size, bony external nares that open at a level anterior to the antorbital notch, elongated nasal bones, a broad supraorbital process of the frontal, a ventrolaterally elongated postorbital process of the frontal, an intertemporal constriction with parietals exposed on its dorsal surface, straight mandibles, a large mandibular foramen, a well-developed coronoid process of the mandible, functional heterodont teeth in the rostrum and mandible, and diastemata between upper cheek teeth. Mammalodon and specific aetiocetid taxa share some derived features of skull morphology, but on the basis of this cladistic analysis these resemblances are interpreted as homoplasy. These homoplasious derived characters include: a relatively broad-based rostrum (Aetiocetus spp.); nasals that increase in width anteriorly towards the nasal opening (C. goedertorum, Ae. cotylalveus, Ae. weltoni, and Ae. polydentatus); the posterior edge of the ascending process of maxilla in transverse line with, or posterior to, the posterior edge of the nasal (Aetiocetus spp.); a low-profile cranium (Chonecetus spp.); a V-shaped frontoparietal suture on the dorsal surface of the cranium (Chonecetus spp.); a cranium without occipital telescoping (C. goedertorum); a supraoccipital with a semicircular outline (C. sookensis); roof of the pterygoid sinus fossa formed primarily by alisphenoid (C. sookensis); and a posteriorly directed mandibular condyle (Ae. weltoni). Many of the homoplasious characters shared by Mammalodon and aetiocetids apparently represent independent reversals to a plesiomorphic character state. Mammalodon colliveri is more derived than Aetiocetidae in some features. Its rostrum is short relative to condylobasal length, has a bluntly rounded apex, and a gently convex lateral profile in dorsal view (versus the straight lateral profile of the rostrum in aetiocetids) (Fig. 2). The premaxillae of Mammalodon do not contact or closely approximate one another at the midline of the rostrum, whereas those of all aetiocetids (with the possible exception of C. goedertorum) do. Furthermore, the premaxillae of Mammalodon are vestigial, dorsoventrally flattened, and have a short suture with the nasal terminating at a level just posterior to the bony external nares. Aetiocetids

75 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 441 possess elongated premaxillae that are dorsally convex in transverse section, and border the lateral edge of the nasal along its entire length. In the facial fossa on the maxilla, Mammalodon possesses five dorsal infraorbital foramina, whereas aetiocetids have two or three. The maxilla of Mammalodon extends posteriorly onto the cranium as a slender linguiform ascending process. In Aetiocetidae the ascending process of the maxilla is relatively broad and triangular. The preorbital process of the frontal has a rounded-off anterior edge, whereas in aetiocetids the anterior edge of the preorbital process is squared-off or slightly concave. The crown enamel on the cheek teeth of Mammalodon has closely spaced salient longitudinal ridges on its buccal and lingual surfaces. The cheek teeth of aetiocetids exhibit the plesiomorphic state of having distinct longitudinal ridges limited to the lingual surface of the crown. Aetiocetidae possess more derived morphology than Ma. colliveri in having both longirostral and platyrostral skulls, a rostrum with a sharply triangular outline, a suture between the maxilla and premaxilla on the rostrum that is marked by a deep groove, the posteriormost upper tooth is implanted in the maxilla at a level anterior to the antorbital notch (although Morawanocetus yabukii appears to have a cheek tooth at a level posterior to the antorbital notch: Barnes et al., 1995: fig. 13), lateral nutrient foramina and associated sulci on the palatal surface of the maxilla (in some taxa), an edentulous infraorbital process of the maxilla, premaxillae adjacent to and at posterior edge of the nasal opening overhang the maxillae, an ascending process of the premaxilla that terminates in line with the posterior half of the supraorbital process of frontal, a shorter and broader intertemporal constriction, occipital cranial telescoping (except in C. goedertorum), a supraoccipital with a triangular anterior edge (except in C. sookensis), basioccipital crests with keeled ventrolateral edges that do not project laterally, and an alveolar margin of the mandible that is parallel to the ventral margin of the mandible when viewed laterally. These comparisons show that morphological similarities between Ma. colliveri and aetiocetids represent shared primitive characters (symplesiomorphies), or have evolved independently in Ma. colliveri and aetiocetid taxa and thus constitute homoplasy. Mammalodon possesses a range of specialized characters such that it is morphologically disparate from Aetiocetidae (and other mysticetes). This disparity is emphasized by the morphology of aetiocetids, which in many respects is more derived than that of Mammalodon. The overall morphology of Mammalodon is intermediate in grade between that of basilosaurid archaeocetes and Aetiocetidae (Fordyce, 1984). The Aetiocetidae exhibit a grade of morphological evolution intermediate between that of Mammalodon and Chaeomysticeti. Aetiocetids are more closely related to edentulous baleen whales, with which they share a more recent common ancestry, than they are to Mammalodon. The phylogenetic analyses of Fitzgerald (2006), Deméré et al. (2008), and this study, agree with Fordyce s (1982a, 1984) hypotheses that Mammalodon represents: (1) a toothed mysticete placed more stemward in mysticete phylogeny than Aetiocetidae, and (2) a more primitive grade of morphological evolution than Aetiocetidae. Comparisons of Mammalodon and Janjucetus Mammalodon colliveri and J. hunderi together manifest a generally similar grade of morphological evolution, as implied by their positions at adjacent nodes, stemward of aetiocetids, in the phylogenies of Fitzgerald (2006) and Deméré et al. (2008) (Fig. 42B, E). As shown in the preceding section (see Results ), this study supports a cladistic relationship between Mammalodon and Janjucetus: Mammalodon and Janjucetus are more closely related to one another than either taxon is to other mysticetes. Mammalodon shares several derived characters with J. hunderi, which in combination are absent in all other toothed mysticetes (Fig. 43). Janjucetus hunderi is therefore placed in the monophyletic family Mammalodontidae. Based on this phylogenetic analysis, Mammalodon and Janjucetus possess synapomorphies that unite them in the family Mammalodontidae, and which differentiate this family of toothed mysticetes from the Aetiocetidae and Llanocetidae. These shared derived characters include: a short rostral portion of the maxilla; a triangular wedge of frontal separates the posteromedial edge of the ascending process of maxilla from the posterolateral margin of the nasal; a linguiform preorbital process of the frontal; a lowprofile braincase; a V-shaped frontoparietal suture on the dorsal surface of the cranium; a cranium with secondarily reduced occipital telescoping; a secondarily semicircular supraoccipital shield; lower teeth implanted within an alveolar groove that has a salient, ridge-like, lateral edge; and posterior upper and lower cheek teeth with two roots joined by an isthmus for part or all of their length. Despite their generally similar grade of evolution, Ma. colliveri is more derived than J. hunderi in many morphological features. Mammalodon has a rostrum with a bluntly rounded apex and gently convex lateral profile in dorsal view (versus the straight lateral profile of the rostrum in Janjucetus). On the premaxilla, the alveoli for the upper incisors are coalesced. Janjucetus retains the primitive state by having upper incisor alveoli separated from one another by an interalveolar septum. The body of the premaxilla is delicate, foreshortened, dorsoventrally flattened,

76 442 E. M. G. FITZGERALD and does not project dorsally above the maxilla. In contrast, the premaxilla of Janjucetus is robust, elongate, has a transversely convex dorsal edge, and projects above the dorsal surface of the maxilla on the rostrum. The nasal of Mammalodon broadens towards its anterior end, whereas that of Janjucetus narrows anteriorly. Whereas Mammalodon possesses five dorsal infraorbital foramina, Janjucetus has two. Mammalodon has a maxilla with a transversely narrow linguiform ascending process, whereas the ascending process of maxilla in J. hunderi is broader and has a triangular outline. The orbit of Mammalodon is more anteriorly and anterodorsally directed than that of Janjucetus. In Mammalodon the parietals meet at a low sagittal suture on the dorsal surface of the cranium, such that in transverse section the braincase has a gently convex dorsal profile. Janjucetus has a salient sagittal crest formed by the parietals on the midline of the braincase. The nuchal crest of Ma. colliveri projects anterodorsally and anterolaterally, whereas the nuchal crest of Janjucetus is more dorsally salient and flares posterolaterally. The mandible of Mammalodon is straight, whereas the mandible of Janjucetus is bowed medially. Lastly, Mammalodon possesses one additional tooth in each quadrant above the molar count of basilosaurids, such that the molar formula is 3/4, whereas Janjucetus possesses the primitive basilosaurid molar formula of 2/3. The morphology of J. hunderi is more derived than that of Ma. colliveri in having: an edentulous infraorbital process of the maxilla; a frontal that lacks a postorbital ridge; an ascending process of the premaxilla that terminates in line with the supraorbital process of the frontal; premaxillae anterior to the nasals with a convex dorsal profile in lateral view; the lateral end of the orbitotemporal crest is on the dorsal surface of the supraorbital process of frontal; a superior lamina of the pterygoid covering most of the ventral exposure of the alisphenoid in the pterygoid sinus fossa; a tubercle on the apex of the anterior process of the periotic; and an accessory ossicle, or its homologue on the outer lip of the tympanic bulla, fused to the anterior process of the periotic. The foregoing comparisons show that despite being reciprocal sister taxa, Mammalodon and Janjucetus are rather disparate from each other in their morphology, each genus possessing numerous autapomorphies. This morphological disparity strongly supports taxonomic distinction between Mammalodon and Janjucetus at the generic level, as proposed by Fitzgerald (2006). Mammalodon appears to be a more highly specialized toothed mysticete than Janjucetus. The majority of autapomorphies that distinguish Mammalodon from Janjucetus are related to its aberrant rostral morphology and features of its feeding apparatus. The possible functional and evolutionary implications of the latter are discussed below. DISCUSSION FUNCTIONAL MORPHOLOGY Morphological description of Ma. colliveri provides evidence on cranial functional complexes and their potential implications for interpreting palaeobiology. Until now, most of our understanding of the palaeobiology of archaic mysticetes has been derived from species of Aetiocetidae (Barnes et al., 1995; Fordyce & Muizon, 2001; Fitzgerald, 2006; Deméré et al., 2008). The significance of aetiocetids notwithstanding, it is clear that they are relatively specialized and apart from their possession of functional teeth are rather similar in morphology to baleen-bearing mysticetes. This was recognized by Emlong (1966) and Van Valen (1968), and has been emphasized by more recent research (Fitzgerald, 2006; Deméré et al., 2008). In consequence, the aetiocetids do not shed light on the earlier evolutionary history of mysticetes, and evolutionary transition from archaeocetes to edentulous baleen whales (Fitzgerald, 2006). As verified in the phylogenetic analysis of this study, mammalodontids represent a more archaic grade of mysticete evolution than aetiocetids. Thus, careful analysis of form and function in Mammalodon, within a phylogenetic framework, may yield insights into broader aspects of evolution. The preservation of the holotype of Ma. colliveri permits critical analysis of a major functional complex: the feeding apparatus. Feeding apparatus and feeding ecology Morphology of the dentition, rostrum, cranium, mandible, hyoid bones, and sternum reflect the function of the feeding apparatus, and hence feeding ecology. The morphology of the feeding apparatus in Mammalodon is different to that of archaeocete outgroup taxa, Chaeomysticeti, and even other toothed mysticetes. Indeed, many of the autapomorphies of Ma. colliveri relate to the specialized morphology of the feeding apparatus. The teeth of Ma. colliveri (and J. hunderi) are relatively and absolutely small compared to those of basilosaurids such as D. atrox (e.g. Uhen, 2004). Although the majority of the crown is worn off in both upper and lower teeth, it is clear that the crowns were less mesiodistally elongated than those of basilosaurids. The upper cheek tooth crowns of Mammalodon are relatively broader in a buccolingual direction than those of basilosaurids. Thus, the tooth crown shape was more conical than the elongated, laterally compressed crowns of Basilosauridae. Despite this more conical shape, the teeth are not slender as in strictly

77 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 443 piscivorous/teuthophagous pelagic odontocetes (Slijper, 1979; Massare, 1987; Barnes et al., 1995; Werth, 2000). Diastemata occur between the upper cheek teeth, whereas the lower teeth are more closely spaced along the tooth row without long intervening diastemata. In the rostrum the teeth are implanted along the entire lateral edge of the maxilla, with the ultimate tooth alveolus posterior to the level of the antorbital notch. The maxillary teeth are implanted obliquely, projecting ventrolaterally and slightly anteriorly from the maxilla at an angle of about 70 to the sagittal plane. The lower cheek teeth are implanted almost squarely in the alveolar margin of the mandible, and are reclined posteriorly in the alveoli. Notably, the upper and lower teeth have strongly emergent crowns, with the roots extensively exposed above the alveolar margin of the rostrum and mandible. This implies well-developed gingival tissue. Although the tooth crowns are, for the most part, abraded off, it is this heavy wear pattern that provides evidence on occlusal relationships. The holotype of Ma. colliveri is an adult individual, and the extreme dental wear could be interpreted as a result of elderly ontogenetic age of the individual. This may be the case. Nevertheless, there are peculiarities of the dental wear pattern that suggest a novel form of dental occlusion in Ma. colliveri, regardless of ontogenetic stage. Both the upper and lower dentition of Mammalodon present three types of wear: apical wear, surficial wear, and cingular erosion (Uhen, 2004). Five aspects of dental wear immediately command attention: (1) apices of the upper and lower tooth crowns are worn down along one horizontal plane (Mitchell, 1989: 2229); (2) there are no vertical surficial wear facets on the buccal surface of the crowns of the posterior lower cheek teeth (p4 m3) but there is a surficial wear facet on the lingual surface of the crown of P2 P4, which in P2 P3 extends basally onto the lingual surface of the root; (3) on the occlusal surfaces of the lower cheek teeth there are two distinct wear facets a more buccal longitudinal (parallel to the long axis of the crown) facet that is continuous with a cingular erosion facet on the mesiobuccal edge of the crown, and a longitudinal facet on the lingual half of the crown; (4) wear on the upper cheek teeth is heaviest on the apex and distolingual side of the crown; and (5) the heaviest wear occurs on P1 P3, and p2 p3. These patterns of dental wear, twinned with the lack of embrasure pits in the palate and mandible, indicate that the cheek teeth of Mammalodon did not form a vertical shearing mechanism as hypothesized for basilosaurids (Uhen, 2004). In Basilosauridae, the laterally compressed and elongated triangular cheek teeth bear vertical surficial wear facets on the lingual surfaces of the upper crowns and buccal surfaces of the lower crowns (Uhen, 2004). In these stem cetaceans, the buccal surficial wear facets on p4 m3 continue from the crown basally towards the cingulum. This is not the case in Mammalodon, where the buccal surface of the p4 m3 crowns is not differentially worn compared to their occlusal or lingual surfaces. Rather than the buccal surfaces of the lower cheek teeth vertically shearing past the lingual surfaces of the upper cheek teeth during occlusion (as in basilosaurids), the apices of the lower cheek teeth of Mammalodon occluded directly with the apex of their upper antagonist in a more anteroposterior motion. By analogy with basilosaurids and J. hunderi, p2 occluded with P1, p3 with P2, p4 with P3, m1 with P4, and so on. I hypothesize that during occlusion, the mesiobuccal half of the lower crown s apex initially contacted the mesial half of the upper crown, and then as the mandible was adducted posterodorsally, the buccal half of the lower crown was drawn backwards and lingually against the apex of the upper crown. Thus, the mesiobuccal side of the lower crown would come to meet the distolingual side of its upper antagonist at an angle. With the teeth in complete occlusion, the upper tooth row appears to have been located slightly anterolateral to the lower tooth row. Within Cetacea, this pattern of dental occlusion appears to be most nearly analogous to that on the anterior teeth of the Recent beluga whale Delphinapterus leucas Pallas, 1776 (Odontoceti). In the beluga, lower teeth occlude directly with their upper antagonists at an angle, such that the apices of lower teeth wear against the distolingual side of the upper teeth (Struthers, 1895b; Brodie, 1989). On the anterior four pairs of teeth, an oblique wear facet develops on the lingual half of the occlusal surface of the upper crown, and on the buccal half of the occlusal surface of the lower crown (Struthers, 1895b). As noted by Mitchell (1989: 2229), heavy planar wear, phenetically similar to that of Mammalodon, is also observed on the teeth of elderly killer whales (Orcinus orca Linnaeus, 1758). However, this supposed similarity in dental wear between Mammalodon and Orcinus is superficial: the teeth of the killer whale interlock and are equally worn on the buccal and lingual halves of the crown s occlusal surface, as well as their mesial and distal edges. Moreover, the worn crowns of Orcinus are rounded and polished (Caldwell & Brown, 1964; Massare, 1987). The teeth of Mammalodon display differential crown wear, as in the beluga. Mitchell (1989) implied that the heavy wear on the teeth of Mammalodon may be the result of a diet including prey with hard skeletal parts. It is not clear whether Mitchell (1989) was implying that Mammalodon was durophagous: if so, this seems unlikely given its rather gracile teeth with long emergent roots, delicate premaxillae, nonbuttressed rostral

78 444 E. M. G. FITZGERALD edges, mandibles that are relatively gracile, and lack of a strong mandibular symphysis. This does not rule out Mammalodon preying on fish, which is perhaps likely given that piscivory is primitive for Neoceti (Barnes et al., 1995; Werth, 2000). Mammalodon colliveri differs from Delphinapterus (and other described Cetacea) in its possession of a distinct oblique lingual wear facet on the occlusal surface of the lower teeth. This lingual wear facet on the lower teeth is unlikely to be caused by wear against the upper teeth, because it is the mesiobuccal edge of the lower teeth crown that occludes with the distolingual side of upper teeth. The lingual papillae on the distal accessory shelf of p3 m2 have an oblique apical wear facet, which is the result of abrasion against food objects as no part of the upper tooth crown would contact this region of the lower tooth during occlusion. Notably, the lingual half of the occlusal surface of right p3 m2 and the left?p4 is more heavily pitted than their buccal half. Additionally, the isolated left?p4, and right p3, m1, and m2, have longitudinal grooves in the lingual wear facet on their occlusal surfaces. Given that the lingual component of wear on the lower cheek teeth is not caused by direct occlusion with the crown of the upper teeth, it is most parsimonious to interpret the lingual abrasion on p3 m2 as a result of tooth-to-food contact. An alternative interpretation is suggested by the transversely oblique orientation of the lingual wear facets, and the longitudinal grooves present in their occlusal surface. It is possible that abrasive material, such as quartz grains in benthic sediment, was ingested during feeding, as suggested by Mitchell (1989: 2229). The extant walrus Odobenus rosmarus Linnaeus, 1758 has distinct lingual wear facets on both its lower and upper teeth caused by ingestion of sediment when suction feeding from the seafloor (Cobb, 1933; Fay, 1982; Gordon, 1984). Odobenus produces powerful intraoral negative pressure through piston-like retraction of its tongue, the suction currents generated being used to dislodge invertebrates from unconsolidated substrate and sucking flesh from the shells of molluscs (Fay, 1982; Kastelein, Dubbeldam & de Bakker, 1997). The posterior movement of the tongue draws abrasive grains across the lingual surfaces of the teeth, resulting in the walrus characteristic dental wear pattern, which includes longitudinal grooves (Fay, 1982; Gordon, 1984; Kastelein & Gerrits, 1990). It is noteworthy that the beluga whale feeds on benthic-dwelling fish, cephalopods, crustaceans, molluscs, and polychaete annelids, and is reported to generate powerful suction currents used in prey dislodgement and seizure (Scammon, 1874; Ray, 1966; Slijper, 1979; Brodie, 1989; Stewart & Stewart, 1989). Sand, gravel, and stones have been recovered from the stomach contents of beluga whales (Brodie, 1989; Stewart & Stewart, 1989). Suction feeding, and its relevance to Ma. colliveri, will be returned to below. It is important to note that the lingual occlusal wear facets on the lower cheek teeth of Ma. colliveri differ from those of the walrus in not being smoothly polished. This may indicate that the lingual wear facet on the teeth of Mammalodon was not caused by lingual abrasion, or that lingual abrasion did produce the wear facet, but wear had not progressed to the point seen in the cheek teeth of adult walrus. Fordyce (1984: 943) proposed that the denticulate cheek teeth of Ma. colliveri formed an effective sieve when in occlusion, as do the elaborately lobate postcanine teeth of the Recent crabeater seal Lobodon carcinophaga Hombron & Jacquinot in Dumont d Urville, Fordyce (1984) argued that Mammalodon used its teeth in filter feeding, supplemented with proto-baleen. He specifically hypothesized that Mammalodon employed a specialized gulping (= lunge or ram feeding) type of filter feeding as utilized by modern balaenopterids (Lambertsen, Ulrich & Straley, 1995; Goldbogen, Pyenson & Shadwick, 2007). Following Fordyce (1984), several authors have continued to postulate that Mammalodon used its teeth to filter feed, supposedly like crabeater seals (Fordyce, 1989b, 1994b; Fordyce & Barnes, 1994; Barnes et al., 1995; Bryden, Marsh & Shaughnessy, 1998; Fordyce & Muizon, 2001; McNamara & Long, 2007). Exceptions are Fordyce (1987), Milinkovitch (1997), Ichishima (2005), and Fitzgerald (2006). Three points require comment here: (1) the basis for comparing the teeth of Mammalodon with those of L. carcinophaga, (2) evidence for proto-baleen, baleen, or a soft tissue filtering structure of any kind in Mammalodon, and (3) inference of lunge filter feeding in Mammalodon. The question of whether Lobodon uses its teeth to filter feed was addressed by Fitzgerald (2006). Almost the entirety of the tooth crowns of Mammalodon are worn away, leaving very little morphology that may be compared with the crowns of Lobodon carcinophaga. The second and third right lower molars of Mammalodon possess relatively low and conical distal accessory denticles, with a broadly open notch between each denticle. This is very unlike the denticles on the posterior lower cheek teeth of Lobodon, which are anteroposteriorly compressed, slender, have lobate apices, and are separated from one another by a deep and narrow notch. Furthermore, whereas the distal accessory denticles on the teeth of Mammalodon exhibit apical wear, those of Lobodon do not. Indeed, the teeth of Lobodon contrast markedly from those of Mammalodon in being unworn. The teeth of Lobodon would cease to be functional if the elaborate denticles were worn down. The foregoing comparison demonstrates that there is

79 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 445 no basis for comparing the morphology of the teeth in Mammalodon with the dentition of L. carcinophaga. As noted in the morphological description and discussion of phylogenetic systematics, the palatal surface of the rostrum of Mammalodon lacks unequivocal osteological correlates of baleen or a homologous soft tissue structure that may be implicated in filter feeding. Such soft tissue may have been present, but the key osteological correlates of enhanced vascular supply to baleen (or its ontogenetic/ phylogenetic precursor), lateral nutrient foramina, and associated sulci on the palatal surface of the maxilla (Deméré et al., 2008), are not in evidence. It is therefore most parsimonious to hypothesize that Ma. colliveri lacked baleen or its more archaic homologue. Fordyce (1984) hypothesized that Mammalodon was a lunge filter feeder, and this taxon has been restored with balaenopterid-like ventral groove blubber located on the ventral surface of its throat (Schouten in Fordyce, 1987; Long in Long & McNamara, 1991; Schouten in Fordyce, 1994b; Jantulik in Bryden et al., 1998; Musser in Long et al., 2002). Ventral groove blubber is unique to Balaenopteridae. Lunge feeding, or intermittent ram filter feeding (see Sanderson & Wassersug, 1993), is a highly specialized and biomechanically complex type of suspension feeding that is specific to Balaenopteridae (rorquals) amongst mysticetes (Lambertsen et al., 1995; Goldbogen et al., 2007). Arnold et al. (2005) suggested that lunge feeding is the most highly derived form of filter feeding in Mysticeti. Indeed, lunge feeding is a putative key innovation of Balaenopteridae (Lambertsen et al., 1995; Goldbogen et al., 2007). Specialized morphological features are functionally implicated in lunge feeding, including: large head and body size; baleen; massive, elongated, laterally bowed, and unfused mandibles capable of rotation about their longitudinal axis; lateral deflection of the coronoid process of the mandible; maxillomandibular cam articulation between the infraorbital process of the maxilla and the mandible; frontomandibular stay; and a nonsynovial temporomandibular joint within a meniscus (Lambertsen et al., 1995; Lambertsen & Hintz, 2004; Arnold et al., 2005; Goldbogen et al., 2007). Mammalodon does not possess any of these features or their osteological correlates. It is therefore unlikely that Mammalodon was a filter feeder, let alone a lunge feeder. Mammalodon colliveri possesses aberrant rostral morphology unique amongst mysticetes. This observation has been made by several workers (e.g. Fordyce & Barnes, 1994; Barnes, 2002; Fraser & Dooley, 2002; Berta et al., 2006; Fitzgerald, 2006; Uhen, 2007a). The key specialized features of the rostrum of Mammalodon are its extremely short length, bluntly rounded apex, reduced contribution of the premaxillae to the rostral apex, and apparently vestigial upper incisors. Such novel morphology implies a novel mode of feeding, as noted by Fitzgerald (2006) and Uhen (2007a). On the cranium, the voluminous temporal fossa accommodated a strongly developed temporalis muscle, which is further suggested by the high and long coronoid process of the mandible. The temporalis muscle therefore appears to have been the primary jaw adductor, as in all Neoceti. The posterior origin of the temporalis muscle on the anterior edge of the nuchal crest, and the length of the intertemporal constriction, indicates a relatively long distance from the point of origin of the temporalis on the cranium to its insertion on the coronoid process. Thus the temporalis had a long lever arm, and was capable of adducting the mandible with force, but relatively slowly. Mammalodon lacks an elongated narrow rostrum and mandibles, or pincers jaws, with many closely spaced slender conical teeth ( cage jaws ) as in extant pelagic delphinids such as Stenella (Norris & Møhl, 1983; Norris et al., 1994). This indicates that Mammalodon did not employ a rapid snapping method of raptorial macrophagy. The anterior upper teeth (incisors) of Mammalodon are vestigial and not implanted in alignment with the cheek teeth. This differs from raptorial odontocetes (and basilosaurid archaeocetes: Uhen, 2004), in which strongly developed conical anterior teeth are implanted in line with posterior teeth and used to seize fast-swimming fish and/or cephalopods (Norris & Møhl, 1983; Werth, 2000). Points of origin and insertion of other masticatory muscles are either not preserved, or ambiguous, limiting interpretation of the relative size and functional significance of such muscle groups as the masseter and pterygoideus. The mandible of Mammalodon is relatively deep, but not robust as in J. hunderi. The mandibular symphysis is short, and the mandibular condyle is directed posteriorly, its articular surface facing posterodorsally. The articular surface of the mandibular condyle in the Mammalodon holotype is corroded and does not preserve surface detail. A right mandible referred to Mammalodon sp. cf. Ma. colliveri (NMV P199587) has a well-preserved condylar articular surface bearing numerous fine grooves. This suggests that the articular surface of the mandibular condyle was capped with hyaline articular cartilage, and that the temporomandibular joint was synovial. A synovial temporomandibular joint is present in basilosaurid archaeocetes (Uhen, 2004) and extant balaenids (Lambertsen et al., 1995), but absent in balaenopterids (Lambertsen et al., 1995). From this consideration of the morphology of the feeding apparatus, we may conclude that: (1) Ma. colliveri did not employ filter feeding in prey capture (as in Chaeomysticeti and perhaps aetiocetid toothed

80 446 E. M. G. FITZGERALD mysticetes); and (2) Ma. colliveri did not possess a raptorial feeding apparatus. Furthermore, Ma. colliveri probably did not feed on large vertebrate prey. Such sarcophagous secondarily aquatic tetrapods (e.g. the killer whale) typically have the following features: upper and lower teeth that interlock when in occlusion; relatively large, robust, and deeply rooted caniniform teeth, which are often broken obliquely via contact with the large bones of their prey; the broken edges of the teeth are usually rounded and polished; a robust rostral apex and robust mandibles; and a dorsoventrally thick lateral edge of the rostrum (Caldwell & Brown, 1964; Massare, 1987; Taylor, 1987; Dahlheim & Heyning, 1999; Werth, 2000, 2006a). This leaves suction feeding as a possible strategy for prey capture and ingestion. Suction feeding has evolved independently in virtually all aquatic vertebrate groups, including marine mammals (Sanderson & Wassersug, 1993; Werth, 2000). Those extant marine mammals known to suction feed include mysticetes (grey whale Eschrichtius robustus Lilljeborg, 1861), odontocetes (Physeteridae, Kogiidae, Ziphiidae, Monodontidae, Phocoenidae, and Delphinidae), and pinnipeds (Otariidae, Odobenidae, and Phocidae) (Werth, 2000). Suction feeding in marine mammals occurs by jaw abduction associated with rapid, pistonlike retraction and depression of the tongue, which results in an abrupt increase in size of the intraoral space and concomitant negative (less than surrounding seawater) pressure within the oral cavity, into which water and prey are drawn (Werth, 1992; Heyning & Mead, 1996; Bloodworth & Marshall, 2005). Water engulfed during this process is expelled prior to deglutition (Werth, 2007). Rostral shape is a fundamental variable in suction feeding and capacity for suction generation in odontocetes (Werth, 1992, 2006b). Odontocete species with blunter and wider rostra and mandibles, and smaller, rounded mouth openings are capable of producing greater negative pressures in the oral cavity than those species with elongated rostra (Bloodworth & Marshall, 2005; Werth, 2006a, b). Blunter rostra with a less laterally open gape improve suction by producing, and channelling, unidirectional water flow into the front of the mouth (Werth, 2006b). Elongated and narrow rostra decrease the efficiency of suction as water flow is into the front of the oral opening, but also into the sides of the mouth via the long laterally open gape. The rostrum of Ma. colliveri is qualitatively similar to that of odontocetes that are specialized suction feeders: it is relatively small, short, broad, and has a bluntly rounded apex. Werth (2006a) introduced the term amblygnathy to describe this condition of shortened and rounded jaws, and proposed a simple dimensionless ratio to describe jaw bluntness, the mandibular bluntness index (MBI): W L = MBI where W is the transverse distance between the lateral edges of the mandibular condyles and L is the length of the mandible measured in a straight line from the anterior tip of the mandible to the posterior edge of the mandibular condyle. A higher MBI (i.e. closer to a value of 1.0) signifies blunter jaws. MBI was calculated for Ma. colliveri to gain a quantitative measure of its jaw bluntness. However, only the right mandible is preserved with the holotype specimen, and thus W cannot be directly measured. Nonetheless, an estimate of MBI can be made by measuring a reconstruction of the left and right mandibles in articulation (Fig. 31C). This yields an estimated MBI of for Ma. colliveri. This value is comparable to the MBI calculated for killer whale O. orca (0.625), Dall s porpoise Phocoenoides dalli True, 1885 (0.625), and beluga whale Delphinapterus leucas (0.651) (Werth, 1992, 2006a). The latter two species are considered exemplary suction feeders (Ray, 1966; Brodie, 1989; Werth, 1992, 2006a). Werth (1992, 2006a) concluded that higher MBI correlated with increased specialization and capacity for suction feeding. This correlation has been corroborated by the results of Bloodworth & Marshall (2005, 2007). Werth (2006a: 585) suggested that amblygnathy in the killer whale is related to its sarcophagous raptorial feeding strategy rather than suction feeding. Although an estimate, the high MBI calculated for Mammalodon supports the qualitative observation that it has short and blunt jaws, comparable with odontocete species that are specialized suction feeders. In Odontoceti reduced dentition correlates with amblygnathy and suction feeding (Norris & Møhl, 1983; Werth, 1992, 2000, 2006a). The premaxillae and incisors of Mammalodon are reduced and gracile, suggesting limited function in prey capture and ingestion: this is a specialized condition (c.f. J. hunderi: Fitzgerald, 2006) consistent with suction feeding. In suction feeding odontocetes the hyoid and sternal bones are dorsoventrally flattened and relatively large, presenting large surface areas for the attachment of hyolingual musculature that retracts the tongue and hyoid apparatus itself (Werth, 1992, 2000, 2007; Heyning & Mead, 1996; Bloodworth & Marshall, 2007). The thyrohyoid of Ma. colliveri differs from those of odontocetes in being massive and tubular, which is the primitive morphology possessed by Basilosauridae. The area of origin for the hyoglossus muscle, which retracts the tongue, is small compared to that of odontocetes. Thus, the hyoid apparatus of Mammalodon does not seem to be specialized for rapid and powerful retraction of the tongue. In contrast, the manubrium of the sternum has a derived morphology of being large and massive

81 THE TOOTHED MYSTICETE MAMMALODON COLLIVERI 447 relative to skull size, dorsoventrally flattened, and presenting a large surface area for the origin of the sternohyoideus muscle, which inserts on the basihyoid and thyrohyoid. The sternohyoideus muscle depresses and retracts the hyoid apparatus. In suction feeding odontocetes the sternohyoideus is massive and, with the hyoglossus, is one of the major muscles involved in tongue/hyoid retraction (Werth, 1992, 2000, 2007). Although the evidence from the hyoid is equivocal, the sternum of Mammalodon presents morphology consistent with that of a suction feeding cetacean. Other aspects of the morphology of Mammalodon may be related to feeding mode. Within the facial fossa on the external surface of the maxilla there are five dorsal infraorbital foramina with relatively wide diameters. The anterior four of these infraorbital foramina open into anterolaterally directed sulci. In basilosaurids, J. hunderi, aetiocetids, and extant mysticetes, two or more dorsal infraorbital foramina may be present, but they do not open into distinct sulci as in Mammalodon. This morphology of dorsal infraorbital foramina implies enhanced trigeminal (infraorbital) sensory innervation, and/or enhanced vascular supply via the infraorbital artery, to facial soft tissues. Based on the direction the foramina open, and the orientation of the associated sulci, the branches of the infraorbital nerve and artery may have supplied relatively well-developed upper labial musculature. Correspondingly, the numerous and relatively large mental foramina on the external surface of the mandible imply extensive vascular supply via the mental artery, and enhanced sensory innervation via the mental branch of the trigeminal nerve, to the lower labial musculature. Perhaps Mammalodon possessed well-developed and mobile lips that were an important tactile sense in prey detection, as well as capable of modifying the shape and gape of the mouth, thereby optimizing suction generation during feeding. The relatively large orbits of Mammalodon are directed more anterolaterally and anterodorsally than in other described toothed mysticetes. This implies that Mammalodon possessed relatively good anterodorsal and anterolateral binocular vision. Binocular vision in an anterodorsal direction is a characteristic of benthic feeding marine mammals (e.g. the walrus and the extinct odontocete Odobenocetops), whereby the animal is capable of looking forward while the rostrum and body are orientated vertically or obliquely to the seabed (Fay, 1982; Werth, 2000; Muizon & Domning, 2002; Muizon, Domning & Ketten, 2002). The occipital condyles of Ma. colliveri are almost hemispherical in outline when viewed laterally and are salient, their posterior edge at a level well posterior to the paroccipital process of the exoccipital. The posterior projection and convexity of the occipital condyles implies that the head of Mammalodon was quite mobile relative to the vertebral column. If Mammalodon foraged on the seafloor, the ability to substantially flex and extend the head would be advantageous in directing the mouth towards the benthos while the longitudinal axis of the body remained at an oblique angle to the seabed. A high amplitude of dorsoventral head movement has been hypothesized for the probable benthic feeder Odobenocetops peruvianus (Muizon & Domning, 2002). This assessment of the functional morphology of the feeding apparatus finds little evidence in support of Mammalodon being a filter feeder. Although facultative raptorial sarcophagy cannot yet be ruled out, the novel morphology of Mammalodon is consistent with specialization for suction feeding. Independent evidence from the dentition, rostrum, cranium, mandible, and sternum suggests that this may be the most parsimonious interpretation of the feeding apparatus. There is some evidence to suggest that Mammalodon was a benthic suction feeder (Fig. 45), although not as specialized for this feeding mode as the walrus and Odobenocetops. An arguably more appropriate analogue is the extant beluga whale Delphinapterus leucas, which captures pelagic prey as well as benthic organisms using suction (Brodie, 1989; Stewart & Stewart, 1989). Analysis of dental microwear represents an opportunity to test this hypothesis, although such a project is beyond the scope of this paper. EVOLUTION Analysis of the morphology, systematics, and feeding apparatus of Ma. colliveri provides insights into general problems of mysticete evolution. Many of these aspects of mysticete evolution cannot be addressed in detail here: and nor is it currently possible to do so. Crucial morphologic, phylogenetic, and functional detail is lacking for most described, as well as many undescribed, toothed archaic mysticetes. Hence, this consideration of evolutionary patterns and processes must be preliminary. However preliminary, the phylogenetic framework of this study provides a test of previous hypotheses and speculations regarding the origin and early evolutionary history of Mysticeti. Evolution of feeding Recent mysticetes possess cranial morphology that is amongst the most specialized of all mammals. Most of these morphological distinctions are inherent to the unique filter feeding system of Mysticeti. Consider the following features, which are typical of the feeding apparatus of most mammals: (1) lack of kinetic joints within the skull; (2) mastication, or intraoral processing, of food; and (3) complex and functional

82 448 E. M. G. FITZGERALD Figure 45. Artistic restoration of Mammalodon colliveri in the Late Oligocene seas off south-east Australia. The three fish depicted on the left side of the figure are an extinct species of whiting (Sillago pliocaenica Stinton, 1952), known from otoliths discovered in the Jan Juc Formation (Stinton, 1958; Fitzgerald, 2004) (painting by Brian Choo, Museum Victoria, Melbourne, Australia). postcanine teeth (Smith, 1993). Crown group mysticetes contrast with most mammals in all of these criteria: they possess kinetic sutures between the rostrum and cranium, do not masticate food, and lack functional teeth of any kind. How this specialized cranial morphology and prey capture method evolved is central to understanding and interpreting mysticete evolution in general. Fitzgerald (2006) and Deméré et al. (2008) have proposed a stepwise pattern of mysticete evolution, with a gradual accumulation of crown mysticete specializations (with emphasis on the feeding complex) across phylogeny. Fitzgerald (2006) emphasized the likelihood that supposed key innovations of Mysticeti, for the most part linked to bulk filter feeding, evolved later in the history of this clade and were therefore not implicit in the divergence and radiation of mysticetes. This hypothesis contrasts with the concept that the origin of bulk filter feeding (and attendant specializations of morphology, physiology, behaviour, and ecology) occurred virtually simultaneously with the origin of Mysticeti, or was even the impetus for it (Fordyce, 1977, 1980, 1989b, 1992). Note that Fordyce s hypothesis has never been explicitly tested through cladistic analysis. This study, in general agreement with Fitzgerald (2006) (and to a lesser extent Deméré et al., 2008), rejects the hypothesis that the divergence of mysticetes from odontocetes is linked to the evolution of filter feeding. Several morphological features that have been functionally related to bulk filter feeding (Sanderson & Wassersug, 1993; Fitzgerald, 2006; Deméré et al., 2008) evolved in mysticetes prior to its unequivocal development, suggesting that they are exaptive for filter feeding (Fig. 46). These features include: large body size, a platyrostral skull, a broad-based and elongated rostrum, thin-edged maxillae, a tubular mandible, and a nonsutured mandibular symphysis (Fig. 46). According to this study, the most stemward mysticetes (SCTM, L. denticrenatus, Ma. colliveri, and J. hunderi) were not bulk filter feeders. Yet, they possess many of those features interpreted here as exaptations for filter feeding. Ergo,