Locomotor Evolution in the Earliest Cetaceans: Functional Model, Modem Analogues, and Paleontological Evidence

Paleobiology, 23(4), 1997, pp. 482-490

Locomotor evolution in the earliest cetaceans: functional model, modem analogues, and paleontological evidence

J. G. M. Thewissen and E E. Fish

Abstract.-We discuss a model for the origin of cetacean swimming that is based on hydrodynamic and kinematic data of modem mammalian swimmers. The model suggests that modem otters (Mustelidae: Lutrinae) display several of the locomotor modes that early cetaceans used at different stages in the transition from land to water. We use mustelids and other amphibious mammals to analyze the morphology of the Eocene cetacean Ambulocetus natans, and we conclude that Ambu- locetus may have locomoted by a combination of pelvic paddling and dorsoventral undulations of the tail, and that its locomotor mode in water resembled that of the modem otter Lutra most closely. We also suggest that cetacean locomotion may have resembled that of the freshwater otter Pteronuru at a stage beyond Ambulocetus.

1. G. M. Thewissen. Department qf Anatomy, Northeastern Ohio Universities College of Medicine. Roots- town, Ohio 44242 E E. Fish Department of Biology, West Chester Unrwrsrty. West Chester, Pmnsvlmnia 19380

Accepted. 7 July 1997

Introduction that wen the earliest whales had adopted modem ways of swimming. In spite of noting The locomotor morphology of terrestrial differences in overall morphology and muscle mammals is very different from that of aquatic development, Kellogg (1936) proposed that mammals. The musculoskeletal system of swimming mammals displays conspicuous late Eocene Basilosaurus had a tail fluke and specializations that can be considered as ad- implied that it swam by dorsoventral oscilla- aptations for life in the water. These adapta- tion (p. 286), the locomotor mode of all mod- tions are the result of dramatic morphological em cetaceans. Barnes and Mitchell (1970) re- evolution that has made it impossible for the constructed Basilosaurus with a broad horizon- most specialized secondarily aquatic mam- tal tail fluke, and Uhen (1991) found that ver- mals, cetaceans and sirenians, to locomote on tebral dimensions of Basilosaurus indicated land. that it swam similarly to modem cetaceans. Cetaceans descended from terrestrial mam- Gingerich et al. (1994) concluded that dorso- mals, and their amphibious ancestors must ventral oscillations of the tail were the mode have had morphologies consistent with land of locomotion for the protocetid cetacean Rod- i and aquatic-locomotion. Both ancestors (land hocetus kasrani. mammals) and descendants (marine cetace- One of the few cetaceans for which a differ- ans) of these amphibious taxa must have been ent means of aquatic locomotion has been pro- morphologically very different from these posed is Ambulocetus nntans, an Eocene ceta- amphibious whale intermediates. The terres- cean from Pakistan (Thewissen et al. 1994, trial relatives of cetaceans, mesonychians, did 1996). Most of the limb skeleton of this taxon not resemble early cetaceans in their locomo- is known and is very different in shape from tor morphology. The postcranial skeleton of that of all other known cetaceans. On the basis mesonychians shows that they were cursors to of overall morphology, Thewissen et al. (1994, varying degrees (Zhou et al. 1992; O'Leary 1996) concluded that the power stroke of Am- and Rose 1995). bulocetus during swimming resulted mainly In spite of the locomotor changes that early from strokes of the feet, and not from strokes cetaceans underwent, few authors in the pa- of the tail. Propulsive strokes were probably in leontological literature have analyzed these the dorsoventral plane, similar to modem ce- changes, and it has commonly been assumed taceans.

'I LOCOMOTION IN THE EARLIEST CETACEANS 483

In contrast to the paleontological literature, guild. If the model is valid, members of these the zoological literature has extensively ad- guilds can be used as functional analogues for dressed changing function of aquatic loco- the stages of cetacean swimming. Among the motion in early cetacean evolution. Howell three terminal locomotor stages, the sequence (1930 p. 21) discussed the evolving locomotor leading to cetaceans can best be studied using patterns in aquatic mammals and stated that modem analogues, because a variety of mod- “there seems to be a decided tendency for em mammals of diverse phylogenetic back- aquatic mammals to develop as the primary ground are members of each guild. means of locomotion a single organ or pair of This paper will modify Fish’s (1996) model organs.” Howell (1930) also made a distinc- and develop a morphological context for the tion between (modem) cetaceans with a fusi- part of the model that relates to cetacean ori- form body and those with an anguiliform gins (Fig. 1). We here propose morphological body (such as Basilosmtrtcs), and suggested correlates for the different locomotor stages that the latter type swam differently. SIijper and test these with the use of quantitative and (1946) studied spinal morphology of cetace- qualitative morphological data from the skel- ans and concluded that Eocene cetaceans and eton. Although the morphology of mammali- mysticetes used undulating locomotor pat- an swimmers has been extensively studied terns, whereas odontocetes employed more (e.g., Howell 1930; Slijper 19361, there has been oscillating patterns, but he did not base these very little explicit work relating morphology conclusions on observations of actual swim- to swimming modes. ming cetaceans. The model as discussed here differs from Locomotor patterns in semiaquatic and that of Fish (1996) in two important respects. aquatic mammals have been studied in mus- We do not differentiate between the two pelvic telid carnivores (Tarasoff et al. 1972; Williams paddling modes, as these are not distinct mor- 1983, 1989; Fish 1994), the rodent Ondatrn phologically. On the other hand, we do distin- (Fish 1982a,b, 1984), dideiphid marsupials guish two distinct modes of dorsoventral un- (Fish 1993a), talpid moles (Hickman 1984), dulation, one that is propelled by the feet (pel- and delphinid cetaceans (Slijper 1961; Fish vic undulation, as discussed by Fish, 1996) and Hui 1991; Fish 1993c). These analyses are and one that is driven by the tail (caudal un- based on kinematic data of swimming behav- dulation). ior and on hydrodynamic principles. They Once the morphological context for the commonly did not involve morphological model is established, we test the model by analyses, the focus of this paper. Williams considering the morphology of one early ce- (1989) and Fish (1993b) discussed the kine- tacean, Ambulocefus nntuns. Ambufocefusis the matics and energetics of several swimming most primitive cetacean for which much of the modes of semiaquatic mammals, and Fish postcranial skeleton is known (Thewissen (1996) put the transition of drag-based to lift- 1994). It is also one of the few archaic cetace- based forms of aquatic locomotion in an evo- ans for which most of the limb skeleton is lutionary perspective. Fish’s (1996) model known. If cetacean locomotor evolution summarized the evolution of the three most- passed through the stages identified in Fig. 1 derived modes of aquatic locomotion in mam- and if Ambulocefus is a good model for the an- mals (lateral pelvic oscillation in phocid seals, cestor of modem cetaceans, Ambulocetus must pectoral oscillation in otariid sea lions, and have displayed a locomotor pattern that is dorsoventral caudal oscillation in cetaceans). typical for one of these stages, and a modern According to this model, cetacean locomotion mammal could be a locomotor analogue for went through the following stages: quadru- Ambulucetus. Thewissen et al. (1994) did not pedal paddling, alternate pelvic paddling, si- identify a modem mammal as a locomotor an- multaneous pelvic paddling, dorsoventral un- alogue of Ambulocefus, instead noting that its dulation, and caudal oscillation. Each of these locomotor pattern combined aspects of seals, locomotor modes is used by a number of mod- otters, and cetaceans. Thewissen et al. (1996) em mammals that compose a locomotor proposed that Ambulocefus’s swimming w?s . 484 J. G.M. THEWISSEN AND F. E. FISH

1936; Fish 1982a,b, 1984). Terrestrial moles (Sculopus, Tulpu) swim by paddling with all four feet, but the more aquatic star-nosed mole, Condyluru, abandons this gait for pelvic paddling during fast swimming (Hickman 1984). Lutru cunudensis swims using a variety of paddling modes (including quadrupedal and forelimb-only), but during rectilinear surface FIGURE 1. Model for the locomotor stages that cetace- swimming it usually paddles by alternating ans passed through in their evolution, modified from a strokes of the hind limbs (Fish 1994). Fore- model developed by Fish (1996). Whereas Fish's model distinguished two pelvic paddling stages, we recognize limbs are not involved in locomotion during only one on the basis of morphological attributes. Fish this locomotor pattern. The locomotor cycle specified the "dorsoventral undulation" mode based on for each limb can be divided into a power the locomotor behavior of Enhydra, but we distinguish "pelvic undulation" and "caudal undulation," and pro- stroke and a recovery stroke. During the for- pose that the swimming mode of Enhydro (pelvic un- mer, the hip is extended (distal femur moves dulation) may not have been a necessary intermediate caudally), the knee is extended, and the heel for the origin of cetacean locomotion. Clades of modem mammals with locomotor behaviors in these categories is plantarflexed while the foot is splayed out. are also listed. The opposite motions occur in the recovery stroke, where the surface area of the foot is reduced by half to minimize drag (Fish 1984). A most similar to that of Enhydra. Fish (1996) similar sequence occurs in Chirunectes: the toes used data on Enhydra from Williams (1989) to are abducted and extended throughout the characterize his dorsoventral undulation power stroke and propulsive force is provided stage. However, Enhydra is a pelvic undulator by extension at the hip and by knee extension and we propose below that cetacean ancestors (Fish 1993a). Abducting the toes will greatly may not have gone through a pelvic undula- enlarge the surface area of the feet if the toes tion stage (see Fig. 1). are elongate and webbed. Williams (1989) concluded that the shift Locomotor Model from quadrupedal to pelvic paddling in mus- Quadrupedal Paddlin~.-Mammals that oc- telids minimizes the disruption of the bound- casionally enter the water almost invariably ary layer (between water and body) by the swim by alternating strokes of the four limbs forelimbs. Disruption of the boundary layer (Fig. 1). Examples of well-studied quadrupe- would lead to increased drag (Webb 1988) and dal paddlers are minks (Musfefa uzson) (Wil- slow the animal down. Restricting paddling liams 1983), North American opossums (Di- to one pair of extremities might also limit in- delphis), (Fish 1993a), and domestic dogs (Fish terference between fore- and hind limbs (Fish 1996). Exact patterns of footfalls have not been 1993a) and has been shown to increase aerobic determined for many quadrupedal paddlers, efficiency (Fish 1982, 1993a; Williams 1983). but Williams (1983) and Fish (1993a) inter- There is no sharp functional distinction be- preted the swimming patterns as a modified tween simultaneous pelvic paddling and dor- terrestrial gait. soventral undulation. In the latter, flexion and pelvic Puddling.-More aquatic members of extension of the back provide the main pro- clades that include quadrupedal paddlers pulsive force, but limb movements are similar. commonly swim by alternate beats of their Consistent with this, Fish (1994) suggested hind limbs only (alternating pelvic paddling) that downward motions of the tail in paddling and do not use the forelimbs for aquatic pro- otters provide some thrust during the recov- pulsion. Examples include river otters (Lutru ery stroke of the feet. canuhsis) (Tarasoff et al. 1972; Fish 1994), the Pelvic undulation in the Dorsowntrul Plum.- South American opossum Chironectes (Fish During this mode of locomotion, waves travel 1993a), and the muskrat, Ondatru (Mizelle through the vertebral column and swing a hy- . LOCOMOTION IN THE EARLIEST CETACEANS 485

drofoil at the feet through the watet Undula- dulators propelled by the feet acting as a hy- tion mainly occurs in submerged swimming. drofoil, similar to Enhydra. On the other hand, Williams (1989) described the transition be- it is also possible that cetaceans went from a tween pelvic paddling and undulation in sea pelvic paddling stage, similar to Lutru, direct- otters (Enhydra). She noted that the trunk of ly to a caudal undulating stage by enhancing the animal remains rigid during surface the caudal hydrofoil and reducing the pelvic swimming, but that the subsurface undula- paddling motions. Pteronura could be an ad- tions allow greater speeds in spite of reduced equate model for this type of locomotion. Cau- stroke frequency. The hind limb muscles pro- dal dorsoventral undulation is thus a swim- vide thrust in paddling sea otters, but during ming mode that may have preceded dorsoven- undulation, the hind limbs and tail passively tral oscillation, the cetacean mode of swim- trail the undulations of the vertebral column. ming. We here expand Fish’s (1996) model to Overall, the shift from paddling to undulation include caudal dorsoventral undulation (Fig. improves performance, as indicated by low- 1). ered use of oxygen (Williams 1989). Caudal Oscillation.--All modern cetaceans Paddling relies on drag forces: propulsion is swim by dorsoventral oscillations of their tail provided during the power stroke in paddlers, flukes. These movements are powered by the but not during the recovery stroke (Fish 1984). muscles of the back and abdomen and thrust Undulation, in contrast, is a lift-based propul- is provided on both the upstroke and down- sive mode: it lacks a recovery stroke and thus stroke. Fish (1993~)and Fish and Hui (1991) minimizes accelerations and decelerations of reviewed the kinematics of swimming in the body throughout the propulsive cycle small cetaceans. (Fish 1984). In undulation and oscillation, sinusoidal The tail of Lutru acts as a hydrofoil and pro- waves run through the body of the swimmer. vides some lift, but the main propelling or- In undulation, different sections of the body gans are the feet, which act as paddles (Fish are in different phases of the sinusoid, where- 1994). In Enhydra the hydrofoil has shifted as in oscillation, the entire body is in the same from the tail to the feet. The main forces in- phase of a sinusoid, and the motions resemble voIved in foot movements are thus different; a standing wave. Undulation and oscillation in Lutru they are drag forces, whereas in En- are common swimming modes in fish, and hydra lift is important. there is a clear correlation between these lo- Caudal Undulation in the Dorsmtral Plum.- comotor modes and body shapes (Motani et As in pelvic undulation, the propulsive force al. 1996). These body shapes have been ar- in caudal undulation is provided by flexion ranged in a graded morphological series that and extension of the lumbar and caudal vertebral column. Unlike in pelvic undulation, ranges from anguiliform, to subcarangiform, the hydrofoil in caudal undulation is located to carangiform, and finally thunniform (see, at the tail. Undulation is rare among mam- e.g., reviews by Lighthill 1969; Webb and mals, but Fish (1994) noted that Lutra mainly Blake 1985; Webb 1988). Locomotor modes undulates under water and remarked that this change gradually along this ciine, with un- mode might also be practiced by the Brazilian dulating movements (defined as waves that river otter, Pteronura. In Lutru, the tail, pow- travel through the body and that propel it) de- ered by lumbosacral flexors, can provide creasing, and oscillations increasing. Cetacean thrust during the recovery stroke during pad- morphology is not unlike that of thunniform dling. The tail thus acts as a hydrofoil, but the fish (Webb and Blake 1983). main propulsive force is still provided by the No mammalian group includes living taxa feet. Lufru therefore represents the transition that document the transition from undulatory between pelvic paddling and pelvic dorsoven- to oscillatory modes. Slijper (1946) suggested tral undulation. that mysticetes might swim by undulatory Fish (1996) suggested that cetaceans went movements, but there is no experimental evi- through a stage where they were pelvic un- dence supporting this. . 486 J. G. M. THEWISSEN AND F. E. FISH

Quadrupedal Paddling aptation for generating propulsion with its .I feet during swimming. A similar conclusion 1- 1- Pelvic Paddling 1.o was reached by Thewissen et al. (1994, 1996) based on the morphology of individual limb 0.8O-1 I I n elements. It is unlikely that Arnbulocetus was a caudal oscillator. Modifications of the mammalian llhLll0.4 Muridae Mustelidae body plan associated with caudal oscillation Didelahidae- TalDidae Cete might include the shape of the hydrofoil on __ r -~- . the tail. There are hydrodynamic implications FIGURE2. Comparison of relative metatarsal length (IonRest metacarpal/longest metatarsal) as an indicator of hydrofoil shape in cetaceans (Webb 1988; of foot length in quadrupedal paddlers and pelvic pad- Bose et ai. 19901, but because the flukes do not diers of four modern clades. Ratios for extinct Ambulo- fossilize, these be evaluated in Cetus natans and Parhyaena ossifiaga are also shown cannot (sources respectively Thewissen et ai. 1996 and O'Leary evohtion. During move- and Rose 1965). Genera included are those listed in Fig. 1. ments are concentrated in one part of the body, and vertebral modifications are to be expected in the area of rotation. Crovetto (1991) Testing the Model: Swimming in and Watson and Fordyce (1993) documented Ambulocetus natans changes in vertebral body shape and interver- The fossil record documents the locomotor tebral disc in the region of greatest oscillation morphology of several cetaceans that are very in dolphins. Reduction of tail length can also different from modem forms. Among these, be expected, as a shorter stalk can provide Arnbulocetus is furthest removed from modem greater moments. cetaceans morphologically. The search for a On the basis of length of the caudal verte- locomotor mode that is different from that of brae, Thewissen et al. (1996) inferred that the modem cetaceans is thus most promising in tail of Arnbulocctus was long and robust. A this taxon. long tail implies that a fluke at the distal part It is not likely that Ambulocctus was a qua- of the tail would have a poor lever arm (Thew- drupedal paddler. Most quadrupedal pad- issen et al. 1996) and would not be an efficient dlers are terrestrial animals, and they lack hydrofoil. In contrast to the long vertebrae in specific morphological specializations for the tail of Arnbulocctus, tail vertebrae of the swimming. Pelvic paddlers display elongated modem cetaceans are relatively short. It re- feet when compared with their terrestrial rel- mains possible that the tail of Arnbulocetus atives (Howell 1930; Stein 1981; Fish 1984), provided some propulsion (see below), but it and this is borne out by comparison of relative is unlikely to have matched the propulsive hind foot length (Fig. 2). The hind-limb pad- force generated by a fluke. dler of four pairs of closely related swimmers Arnbulocetus was not a quadrupedal paddler has longer metatarsals than the quadrupedal or caudal oscillator. This implies that it used paddler. This suggests that foot length covar- an intermediate mode of locomotion and was ies with locomotor behavior. However, it is different in this respect from modem cetace- clear that the phylogenetic history of a clade ans. If cetaceans went through the locomotor matters since the metapodial length ratio dif- stages predicted by the proposed model, then fers vastly between the different clades. For in- Ambulocetus's swimming may have included stance, some quadrupedal paddlers (e.& Rat- pelvic paddling, pelvic undulation, or caudal tus) have lower ratios than some bipedal pad- undulation. The morphology of Arnbulocetus dlers (e.g., Lutra). could then be consistent with that of lutrines The ratio for Arnbulocetus is much smaller using these locomotor modes. Comparing lu- than that for its terrestrial relatives, the me- trine morphology with that of ArnbuZocetus can sonychians (Puchyma [Zhou et al. 1994; then be used to evaluate the hypotheses pre- O'Leary and Rose 19951). We interpret the dicted by the model. metatarsal elongation of Arnbulocetus as an ad- The morphology of the propulsor (paddle

. LOCOMOTION IN THE EARLIEST CETACEANS 487

1 longest finger length longest toe length

0.8

0.7

iengm of toe 1 E 0 - length of toe 5 I I I I I I FIGURE 3. Examples of lift-based (AB) and drag-based 0.3 0.4 0.5 0.6 0.7 (C) propulsors, showing the simrlaritv of Ambulocrtus's foot (D) to a drag-based propulsor. Figured are dorsal FIGURE 1. hot bize and shape in mustelids. Symmetrv vtews of right foot of the sea otter Enhydra (A), rrght of the foot (ratio of the lengths of the first and fifth hand of sea lion Zulophus (B), right foot of rlver otter Lu- digits) 1s plotted against the relative length of the lon- tra canadensis (C),and right foot of cetacean Arnbulocetus gest digit (longest finger/longest toe). Ptrronuru and (after Tbewissen et al. 1996) (D). Ambulocetus cannot be plotted because none of the rn- vestigated specimens had a full complement of hand and foot bones. or hydrofoil) is an important indicator of these intermediate locomotor modes. The transition feet (Fig. 3, y-axis), but these are also highly from hind-limb paddling modes of locomo- asymetrical (x-axis), unlike other mustelids, tion to foot- or tail-based undulatory modes whereas those of Lutra are more symmetrical involves the development of a hydrofoil and and function as paddles. the shift from drag-based to lift-based pro- The feet of Ambulocetus had the shape of a pulsion. Optimal propulsor shape for a drag- high triangle (Fig. 3D),and the axis of the foot based paddle is very different from that of a is between third and fourth digit, as in its ter- lift-based hydrofoil (Blake 1981; Webb 1988; restrial relatives, the mesonychians (Mesonyx Webb and Buffrtkd 1990). Drag-based pro- and PachyMna [OLeary and Rose 19951). Am- pulsors consist of a flat plane that is held nor- bubcetus's feet are larger than those of meson- mal to the direction of movement (Webb and ychians and have longer central digits, as in- Blake 1985). The optimal propulsor of this dicated by metatarsal length (Fig. 5). The first type has the shape of a high triangle with a digit of Ambulocetus was absent and the sec- narrow stalk (Webb 1988). The effiaency of a ond and fifth digit are similar in robustness lift-based propukor (hydrofoil) depends on and length of preserved elements. Ambuloce- its aspect ratio: the ratio width/length, where tus's feet also closely match those of drag- width is measured perpendicular to the flow. based mustelids such as Mustela uison and Lu- Effiaent hydrofoils have a high aspect ratio. tra, in that the central digits are longer than This ratio often cannot be calculated in fossils digits I1 and V (Fig. 5). This suggests that the because the position of the hydrofoil during feet of Ambulocetus used a drag-based loco- locomotion is not known. Overall, hydrofoils motor mode in water. This is unlike the lift- tend to be narrow and end in a pointed tip lat- based hydrofoil of Enhydra, in which toe (and erally (Webb 1988) and their shape is highly metatarsal) length diminish from digit V to I asymetrical (Fig. 3A,B). These shape differ- (Figs. 3A, 5). This suggests that Ambulocetus ences can be used to evaluate propulsive forc- was not a pelvic undulator. Aquatic locomotor es generated by the foot. Propulsor shape evolution from mesonychians to Ambulocetus within mustelids can serve to illustrate this. is thus best compared with modem analogues Enhydra not onIy has disproportionately Iong such as Mustela uison, Lutra, and Ptermura. . 488 J. C. M. THEWISSEN AND F. E. FISH

160 Length of caudal 111 ) metatarsals [mml 140

120 3.0

100 2.5 1 80 3-51 ( Lenahof) Ln radius I I I I I I 60 3.5 4.0 4.5 5.0 5.5 FIGURE 6. Length of caudal vertebra 111 in mustelids (open circles), mesonychians, and Arnhulocetus. Enhydra’s tail is short and not used for propulsion, whereas the tail 40 does provide lift in Lutro and Ptrronura. Similarly, the tail vertebrae of mesonychians (Mcronyx [Scott 18883; Padyaena [Zhou et al. 19921) are shorter than rhosr of cetaceans, consistent with the presumed absence of tail- ( Len@0f ) Ln radius based propulsion in mesonychians. In thr fossil taxa, (black circles), caudal vertebra was not alwavs pre- I I I11 served and values for a nearby proximal caudal vertebra 3.5 4.0 4.5 5.0 5.5 were substituted. FiGURL 5. Metatarsal lengths for mustelids and Cete (cetaceans and mesonychians) as plotted against an es- timator of body size (Ln radius length). Nonspecialized size from Mustela vison to Lutru and Ptrrortrirn. mustelids and cetan3 haw metatarsals 11 and \I that arc Body size as well as tail-produced lift also in- not greatly different in length from central digits (111 and I\’, grry), which art*the longest. This morphology is crease along this series. On the other hand, found in taxa that arcs mainly terrestrial (Mustelidae: Enhydra is larger than Pterorruru and has a Mustela itiwn. Martes pmnanti; Cetc: Mesonwx obtusidens shorter third caudal vertebra, consistent with [Scott 18881; Padiyarm ossifrap [O’Leary and Rose 199.51). Longer central digits arc found in groups with the trivial role of the tail in propulsion in En- paddles that lack a pedal hvdrofoil (Lutra canadensis, hydra. This pattern is consistent with that in Ambufocctus nntans). A pedal hvdrofoil is present in En- cetaceans, where Ambulocetus’s tail (as based hydra. which has very large and proportionally unusual feet. on the third caudal vertebra) is much larger and more powerful than that of its terrestrial relatives, Mesonyx and Pachpmu (Fig. 6). It is In addition to being propelled by the feet, likely that the tail provided some propulsion Ambulocetus also had a powerful tail, as in in Ambulocetus, although it was not the main Pteronura and Lutru. The tail of Pteronura is as propulsor. The exact surface areas of tail and long as in Lutru, 125% of the thoracolumbar feet of AmbuZocetus cannot be determined at length, versus 64% in Enhydra. The tail of Pter- this point, and their relative contributions to mum is probably more involved in propulsion propulsion are thus unknown, but it seems than that of Lutra, since the transverse pro- likely that the feet were more important. It is cesses of the proximal caudal vertebrae are clear that, based on present evidence, Ambu- larger for insertion of the flexors and extensors Zocetus is most similar in locomotor morphol- of the tail. In addition, the fleshy part of the ogy to Lutru suggesting that the taxon proba- distal tail of Pteronura is laterally expanded bly used pelvic paddling and caudal undulat- (see Redford and Eisenberg 1992: Plate 8). ing in swimming. The swimming behavior of This is consistent with the morphology of in- Pteronura has not been analyzed in detail, but dividual vertebrae. Figure 6 shows that the this taxon might also be a good analogue for length of the third caudal vertebra increases in Ambulocetus‘s locomotion. , * ;-' d.. 3

LOCOMOTION IN THE EARLIEST CETACEANS 489

The locomotor skeleton of Arnbulocetus dif- tus was ancestral to all later cetaceans; a com- fers from that of mesonychians (Scott 1888; prehensive phylogenetic analysis of early ce- Zhou et al. 1992; O'Leary and Rose 1995) in taceans has yet to be undertaken (Thewissen the presence of much larger feet and a stron- 1994). It also does not imply that all early ce- ger tail. Both are consistent with improved taceans swam similarly. Our study does not aquatic locomotion, pelvic paddling, and cau- encompass locomotor diversity across all Eo- dal undulating. Given that modem cetaceans cene cetaceans, instead focusing on one tran- swim by means of caudal oscillation, it can be sitional taxon for which the osteology is well expected (Fig. 1) that the caudal undulation known. component was increased in those cetaceans On the other hand, locomotor patterns are that are intermediate between Arnbufocetus remarkably stable within higher clades. Mem- and modem whales. Subsequently, caudal os- bers of extant groups are mostly restricted to cillation replaced caudal undulation in ceta- minor variations of one locomotor pattern: all cean evolution. Cetacean skeletons interme- sirenians and cetaceans are caudal oscillators, diate between Arnbulocetus and Neogene all phocids and odobenids are pelvic oscilla- whales are rare, but one of the more complete tors, and all otariids are pectoral oscillators is that of middle Eocene Rodhocrtris kasruni (Fish 1996). This suggests that selection for CR- (Cingerich et al. 1994). Consistent with the re- hanced swimming was severely constrained sults of this paper, Rodliocetus probably had by existing morphologies and that phylogeny, small feet (the feet are not preserved, but the possibly through conservative neuromuscular pelvis and femur are) and a strong tail. Cin- patterns, plays a major part in determining lo- gerich et al. (1994) proposed that Rodhocetus comotor patterns. This further emphasizes the had a fluke, but this remains to be confirmed need to study locomotor morphology in a rig- by recovery of distal caudal vertebrae. orous phylogenetic context. Mustelids do dis- In spite of the similarity between Ambufo- play a remarkable variety of locomotor modes, cetus and lutrines, there also are obvious dif- from quadrupedal paddling to dorsoventral ferences. Otters are agile hunters that display undulation. Nevertheless, all these modes are a variety of swimming modes and commonly part of the same transformational series (Fig. chase prey. Arnbufocetus was probably not a 1). pursuit hunter, but rather an ambush hunter, There is rapid evolution of locomotor relying on sudden bursts of movement to se- modes in most mammalian clades--cetaceans cure prey (Thewissen et al. 1996). Ambufocefus may have evolved caudal oscillation in ap- lacked the agility and sustained speed of lu- proximately four million years (Gingerich et trines and its swimming was probably not re- al. 1994). It is thus surprising that the entire lated to chasing prey, and this may be reflect- range of swimming modes is still represented ed in the morphology of the skeleton. It is pos- among modern lutrines, a subfamily that is sible that ambush hunting in the earliest ce- approximateiy as old as pinnipeds (late Oli- taceans represents a phylogenetic constraint gocene). It is possible that body size was an and was present in the mesonychian ancestors important factor since locomotor specializa- of cetaceans. tions tend to be more pronounced in larger taxa (Webb and Buffrenil 1990). Body sizes Conclusions may have strongly constrained the evolution We conclude that the paleontological evi- of pinnipeds, cetaceans, and sirenians, as all dence is consistent with a model for the origin three of these clades radiated at body sizes of cetacean locomotion as initially proposed much larger than lutrines did. by Fish and modified in Figure 1. Lutrines are the best extant functional models for early ce- Acknowledgments tacean locomotion and that the locomotor Financial support for this work was provid- morphology of Ambulocetus may have been ed by the National Geographic Society (5536- most similar to that of Lutra or (less likely) 95) and the National Science Foundation Ptetunura. This does not imply that Ambufoce- (EAR-9526686) to Thewissen and by the Officg . 490 J. G. M.THEWISSEN AND F. E. FISH

of Naval Research (N00014-95-1-1045)to Fish. Motani, R., H. You, and C. McGowan. 1996. Eel-like swimming in the earliest ichthyosaurs. Nature 382347-348. We thank P. Lemelin, S. Madar, M. Uhen, and O’Leary, M. A., and K. D. Rose. 1995. Postcranial skeleton of T. Williams for reading and commenting on the early Eocene mesonychid Pochycrrnu (Mammalia: Meson- the manuscript. ychia). Journal of Vertebrate Paleontology 15:401-430. Redford, K. H.,and J. E Eisenberg. 1992. Mammals of the Neo- tropics, Vol. 2. University of Chicago Press, Chicago. Literature Cited Scott, W. 6. 1888. On some new and little known creodonts. Journal of the Academy of Natural Sciences Philadelphia (Sec- ond 9:155-185. Barnes, 1. C., and E. D. Mitchell. 1978. Cetacea. Pp. 582-602 in Series) Slijper, J. 1936. Die Cetacem: Vergleichend-anatomisch und V. J. Maglio and H. 6. S. Cook, eds. Evolution of African E. systematixh. Asher and Cie., Amsterdam. mammals. Harvard University Press, Cambridge. - . 1946. Comparative biologic-anatomical investigations Blake, R. W. 1981. Medranics of drag-based mechanisms of pro- on the vertebral column and spinal musculature of mammals. pulsion in aquatic vertebrates. Symposium of the Zoological Verhandelingen der Koninklijke Nederlandsche Akadcmie Society of London 4829-52. van Wetenschappen, Natuurkunde 62:l-128. Bw,N., J. Lien, and J. Ahia. 15-90. Measurements of the bodies -. 1961. Locomotion and locomotory organs in whales and of several cetacean species. Proceedings of the flukes and dolphins (Cetacea). Symposium of the Zoological Society Royal %ety of London B 242:163-173. of London 577-94. Crovetto, A. 1991. Etude osteometrique et anatomo-functionel- Stein, 8. R. 1981, Comparative limb myology of two opossums, le de la colonne vertebrale chez les grands cttaces. Investi- Didelphis and Chironcctrs. Journalof Morphology 169:113-140. gations on Cetacea (Berne, Switzerland) W:7-189. Tarasoff, F. J., A. Bisallon, J. Pierard, and A. P. Whitt. 1972. Lo- Fish, F. 1982a. Aerobic energetics of surface swimming in the comotory patterns and external morphology of the river otter, muskrat Ondotru :ibrthrcus. Physiological Zoology 55:190-189. sea otter, and harp seal (Mammalia). Canadian Journal of 20- - . 1982b. Function of the compressed tail of surfaceswim- ology 50:915-929. ming muskrats (Ondotru :ibcthicus). Journal of Mammalogy Thewissen, J. C.M. 1994. Phylogenetic aspects of cetacean or- 63591- 597. igins: a morphological perspective. Journal of Mammalian - . 19M. Mechanics, power output and efficiency of the Evohtion 2157-184. swimming muskrat (Ondotru tibethicus). Journal of Experi- Thewissen, J. C.M., S. T. Hussain, and M.Arif. 1994. Fossil ev- mental Biology 110:183-201. idence for the origin of aquatic locomotion in archamete -. 19935. Comparison of swimming kinematics between whales. Science 263:210-212. terrestrial and semiaquatic opossums. Journal of Mammalogy Thewissen, J. C. M., S. 1. Madar, and S. T. Hussain. 1996. Am- 74275-284. bulocrfrts notans, an Eocene cetacean (Mammalia) from Paki- -. 1993b. Influence of hydrodynamic design and propul- stan. Courier Forschungsinstitut Senckenberg 191:1-86. sive mode on mammalian swimming energetics Australian Uhen. M. D. 1991. Vertebral proportions as indicators of loco- Journal of Zoology 42:79-101. motor style in mammals. Journal of Vertebrate Paleontology -. 1993~.Power output and propulsiveefficiency ofswim - ll:59A. [Abstract.l ming bottlenose dolphins (Tursropz truncotus). Journal of Ex- Watson, A. C., and R. E. Fordyce. 1993. Skeleton of two minke perimental Biology 185:179-193. whales, BQ/@fY.nOptV~ocutorostruto. stranded on the south-east -. 1994. Association of propulsive swimming mode with coast of New Zealand. New Zealand Natural Sciences 2O:l- 14. behavior in river otters (Lutro cunndmsis). Journal of Mam- Webb, W. 1988. Simple physical principles and vertebrate malogy 75:989-997. P. aquatic locomotion. American Zoologist 28709-725. -. 1996. Transitions from drag-based to lift-based pro- Webb, P. W., and R. W. Blake. 1985. Swimming. Pp.110-128 rn pulsion in mammalian swimming. American Zoologist 36: M. Hildebrand, D. M. Bramble. K. F. Liem, D. 8. Wake, eds. 62R-641. Functional vertebrate morphology. FJeIknap Press, Cam- Fish, F. E., and C. A. Hui. 1991. Dolphin swimming-a review. bridge, Mass. Mammal Review 21:181-195. Webb, P, W.. and V. de Buffrtnil. 15-90. Locomotion in the biol- Cingertch, P. D., 5. M. Raza. M. Arif, M. Anwar, and X. Zhou. ogy of large aquatic vertebrates. Transactions of the American 1994. New whale from the Eocene of Pakistan and the origin Fisheries Society 119629-641. of cetacean swimming. Nature 368644-847. Williams, T. M. 1983. Locomotion in the North American mink, Hickman, C. C. 1984. Swimming ability of talpid moles, with a semi-aquatic mammal. 1. Swimming energetics and body particular reference to the semi-aquatic Condyluro cristutn. drag. Journal of Experimental Biology 103:155-168. Mammalia 48505-513. -, 1989. Swimming by sea otters: adaptations for low en- Howell. A. 8. 1930. Aquatic mammals: their adaptations to life ergetic cost locomotion. Journal of Comparative Physiology A in the water. Thomas, Springfield, Ill. 164:815-824. Kellogg, R. 1936. A review of the Archaeoceti. Carnegie Insti- Zhou, X., W.J. Sanders, and P. D. Gingerich. 1992. Functional tution of Washington Publication 4821-366. and behavioral implications of vertebral structure in F’~?chyur- Lighthill, M. J. 1969. Hydromechanics of aquatic animal pro- no ossifrogo (Mammalia, Mesonychia). Contributions from the pulsion. Annual Review of Fluid Mechanics 1:413-446. Museum of Paleontology, University of Michigan 28:289-319.