Phylogenetic analyses of molecular data on extant animals strongly support the notion that hippopotamids are the closest relatives of cetaceans (whales, dolphins and porpoises)1, 2, 3. In spite of this, it is unlikely that the two groups are closely related when extant and extinct artiodactyls are analysed, for the simple reason that cetaceans originated about 50 million years (Myr) ago in south Asia, whereas the family Hippopotamidae is only 15 Myr old, and the first hippopotamids to be recorded in Asia are only 6 Myr old4. However, analyses of fossil clades have not resolved the issue of cetacean relations. Proposed sister groups ranged from the entire artiodactyl order5, 6, to the extinct early ungulates mesonychians7, to an anthracotheroid clade8 (which included hippopotamids), to weakly supporting hippopotamids (to the exclusion of anthracotheres9, 10).
The middle Eocene artiodactyl family Raoellidae11, 12, 13, 14 is broadly coeval with the earliest cetaceans, and both are endemic to south Asia. Raoellids, as a composite consisting of several genera, have been added to some phylogenetic analyses5, 10, but no close relation to whales was found because raoellid fossils were essentially limited to dental material11, 12, 13, 14. We studied new dental, cranial and postcranial material for Indohyus, a middle Eocene raoellid artiodactyl from Kashmir, India (Fig. 1). All fossils of Indohyus were collected at a middle Eocene bone bed extending for about 50 m at the locality Sindkhatudi in the Kalakot region of Kashmir on the Indian side of the Line of Control. Our analysis identifies raoellids as the sister group to cetaceans and bridges the morphological divide that separated early cetaceans from artiodacyls. This has profound implications for the character transformations near the origin of cetaceans and the cladistic definition of Cetacea, and identifies the habitat in which whales originated. Taken together, our findings lead us to propose a new hypothesis for the origin of whales.
Figure 1: Osteology of Indohyus and cross-sections of long bones of Eocene cetartiodactyls.
a, b, Oblique lateral view of skull RR 208 (a) and ventral view of skull RR 207 (b). c–h, Posterior views of humerus (RR 149, c) and femur (RR 101, d), plantar views of metacarpal (RR 138, e) and proximal manual phalanx (RR 19, f), dorsal view of astragalus (RR 224, g), and posterior view of metatarsal (RR 139, h). i–l, Histological mid-shaft sections for humerus of the pakicetid Ichthyolestes (H-GSP 96227, i), humerus of Indohyus (RR 157, j), femur of Indohyus (RR 42, k) and femur of the artiodactyl Cainotherium (IVAU unnum, l). Both scale bars are 1 cm; the scale bar near d goes with a–h, and that near l goes with i–l.
High resolution image and legend (282K)Cetaceans and raoellids are sister groups
To investigate the importance of raoellids in cetacean phylogeny, we excluded raoellids from our initial phylogenetic analysis of artiodactyls plus cetaceans. Our data set differed from previous analyses10 by the addition of several archaic anthracotheres, and some corrected scores for pakicetid cetaceans. This analysis found stronger support for hippopotamid–cetacean sister-group relations than the previous analysis10, consistent with molecular studies1, 2, 3. However, the base of the artiodactyl cladogram is poorly resolved (see Supplementary Information for details on phylogenetic runs). In a second cladistic analysis (Fig. 2), we added the raoellids Khirtharia and Indohyus as well as several archaic ungulate groups (condylarths) and found that raoellids and cetaceans are sister groups and that they are the basal node in the Cetacea/Artiodactyla clade, consistent with some previous analyses that used different character sets5, 6. Our analysis is the first to show that raoellids are the sister group to cetaceans, resolving the biogeographic conundrum and closing the temporal gap between cetaceans and their sister. Relations between most artiodactyl families higher in the tree are poorly resolved, and our data lack implications for the relations between these families. Our analysis strongly argues that raoellids and cetaceans are more closely related to each other than either is to hippopotamids.
Figure 2: Phylogeny of artiodactyls, cetaceans and archaic ungulates.
The figure shows a consensus cladogram produced by heuristic searches with PAUP (random addition sequence, 1,000 repetitions), using a published data set10. See Supplementary Information for further details.
High resolution image and legend (178K)Indohyus shares with cetaceans several synapomorphies that are not present in other artiodactyls. Most significantly, Indohyus has a thickened medial lip of its auditory bulla, the involucrum (Figs 1 and 3), a feature previously thought to be present exclusively in cetaceans. Involucrum size varies among cetaceans, but the relative thickness of medial and lateral walls of the tympanic of Indohyus is clearly within the range of that of cetaceans and is well outside the range of other cetartiodactyls (Fig. 3). Other significant derived similarities between Indohyus and cetaceans include the anteroposterior arrangement of incisors in the jaw, and the high crowns in the posterior premolars.
Figure 3: Plot of the ratio of the thickness of the medial tympanic wall to that of the lateral tympanic wall against the natural logarithm of the width across occipital condyles, showing that the ratio in Indohyus is similar to that in cetaceans.
In cetaceans (open squares), the medial tympanic wall is inflated and called the involucrum, and the lateral tympanic wall is thinned and called the tympanic plate. In artiodactyls (open triangles), the medial and lateral tympanic walls are more similar in thickness, causing values on the y axis to be closer to 1. See Supplementary Information for further details.
High resolution image and legend (59K)Characterizing Cetacea
Until now, the involucrum was the only character occurring in all fossil and recent cetaceans but in no other mammals5, 15, 16. Identification of the involucrum in Indohyus calls into question what it is to be a cetacean: it requires either that the concept of Cetacea be expanded to include Indohyus or that the involucrum cease to characterize cetaceans. We argue that the content of Cetacea should remain stable and include Pakicetidae, Ambulocetidae, Remingtonocetidae, Protocetidae, Basilosauridae, Mysticeti and Odontoceti6, 7, 9, 10, 17. Thus, Cetacea remains a monophyletic group, whereas Artiodactyla remains a paraphyletic group (because Raoellidae are included but Cetacea are excluded). An alternative classification would render both Cetacea and Artiodactyla monophyletic by including Raoellidae in Cetacea and by limiting Artiodactyla to those clades one node above the raoellid/cetacean node (Fig. 2). We do not prefer this classification because it causes instability by significantly altering the traditional content of both Artiodactyla and Cetacea.
Characters identified as synapomorphies for Cetacea in some of our most parsimonious trees include: long external auditory meatus, double-rooted P3/, lack of P4/ protocone, M1-2/ metacones present but small, and lack of M1-2/ hypocone. None of these features characterize all modern and extinct cetaceans; the dental characters, for instance, cannot be scored in toothless mysticetes. In addition, all of these characters are found in some mammals unrelated to cetaceans.
We attach particular importance to two character complexes that characterize basal cetaceans, constitute synapomorphic suites for the order, and are of great functional importance. All fossil and recent cetaceans differ from most other mammals in the reduction of crushing basins on their teeth: there are no trigonid and talonid basins in the lower molars, and the trigon basin of the upper molars is very small (for example in pakicetids and ambulocetids) or absent. Crushing basins are large in raoellids (Fig. 1a, b) and other basal ungulates. This implies that a major change in dental function occurred at the origin of cetaceans, probably related to dietary change at the origin18. Reduced crushing basins also occur in mesonychids, archaic ungulates long thought to be closely related to cetaceans. However, mesonychian molars have wear facets very unlike those of cetaceans7, 18, whereas wear facets in raoellids are more similar to wear facets in early cetaceans14.
The second character complex that identifies cetaceans is the shape of the postorbital and temporal region of the skull. In early cetaceans, this region is long and narrow19. This affects the sense organs: the olfactory peduncle is long and narrow and the orbits are set close together near the roof of the skull. It also affects oral function, the nasopharyngeal duct is narrow, and the out-lever of the masticatory muscles is long, increasing the closing speed of the jaws. We speculate that the importance of different sense organs was related to these changes, or that changing diet led to a change in food-processing organs.
Indohyus was aquatic
Behaviourally, the earliest whales (pakicetids) were aquatic waders5, 20, 21, 22, 23. This led us to investigate whether Indohyus was aquatic too. Cortical bone thickness in secondarily aquatic tetrapods is commonly increased at the expense of the medullary cavity, a pattern called osteosclerosis24. Osteosclerosis occurs in early whales20, 21, 25, manatees26, 27, sea otters28, Hippopotamus29, beavers29, pinnipeds29 and Mesozoic marine reptiles26, 30. Osteosclerosis provides ballast that allows some aquatic taxa to be bottom walkers (hippopotamids) and others to maintain neutral buoyancy in water (manatees)30. Histological sections indicate that the limb bones of Indohyus are also osteosclerotic (Fig. 1i–l), in a similar manner to those of pakicetid cetaceans. Our survey of cortical bone thickness in the limb bones of terrestrial artiodactyls shows that this pattern is unusual for that order: in mid-bone cross-sections of the femur, the medullary cavity makes up between 0.60 and 0.75 of the width of the bone, whereas in aquatic mammals the values are lower (Hippopotamus, 0.55; pakicetids and ambulocetids, 0.25–0.57). In Indohyus this ratio is 0.42, suggesting that Indohyus was osteosclerotic and thus aquatic. We interpret the limb osteosclerosis of Indohyus to be related to bottom walking and not to slow swimming, because the limbs are gracile and not modified into paddles.
To investigate further the hypothesis that Indohyus was aquatic, we studied the stable isotopes of its enamel, a tissue relatively resistant to preburial and postburial alteration of isotopic composition31. Enamel 
18O values are influenced by the oxygen isotope composition of the food and water ingested by an animal as well as by certain physiological processes (such as sweating, panting and respiration)32. For aquatic species, the flux of environmental water by means of direct ingestion and transcutaneous exchange overwhelms all other oxygen sources33 and can cause the enamel 18O values of freshwater taxa (for example Hippopotamus) to be 2–3
lower than those for terrestrial mammals33, 34. Mean 18O values for four individuals of Indohyus are at least 2 lower than those for our comparative sample of Eocene terrestrial and semi-aquatic mammals from formations of India and Pakistan of similar or slightly older age (Fig. 4)22, 23. Although not representative of the specific deposits from which Indohyus was collected, oxygen isotope values for each ecological type from these sites (namely terrestrial, 24–28; semi-aquatic, 23 or less) are surprisingly consistent regardless of age or location (Fig. 4). This suggests that temporal and spatial variation in environmental isotope values was relatively minor and was most probably insufficient to account for the extremely low 18O we have reported for Indohyus. We did not recover tooth material of other mammals at the Indohyus site; until such material can be analysed, the most consistent interpretation is that these low values are a result of the aquatic habits of this species.
Figure 4: Bivariate plot of 
18O and 13C values for enamel samples of early and middle Eocene mammals from India and Pakistan.
![Figure 4 : Bivariate plot of |[dgr]|18O and |[dgr]|13C values for enamel samples of early and middle Eocene mammals from India and Pakistan. Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.com](/nature/journal/v450/n7173/images/nature06343-f4.0.jpg)
Results are shown as means 
s.d. for the sample population. Filled triangle, Indohyus; open triangle, Khirtharia; open squares, terrestrial mammals; filled squares, brackish-water anthracobunids; filled circles, freshwater/brackish-water archaeocetes. See Supplementary Information for details.
Supporting evidence for a semi-aquatic life of Indohyus comes from examination of its inter-individual variation in 18O values. The overwhelming influx and mitigating influence of isotopically homogeneous environmental water causes the variation in individual 18O values for semi-aquatic and aquatic species (s.d. < 0.5) to be much lower than that of terrestrial species (s.d. > 1.0)33. This line of evidence is especially relevant in our study because it does not require an approximation of the mean environmental isotope values for a site through analysis of the associated fauna. Variation in 18O values for Indohyus (s.d. = 0.4) is extremely low and when compared with species of sufficient sample size (n > 3) it is similar to that of the semi-aquatic archaeocete Pakicetus. Given that the influence of physiological and environmental factors on body water 18O values is more strongly felt at smaller body sizes35, this low level of variation is particularly compelling for Indohyus, with a body mass of less than 50 kg.
To explore the diet of Indohyus we studied carbon isotopes. Enamel 13C values are defined by the carbon isotope composition of an animal's diet and can be used to identify the food webs and resources used by an animal36. The 13C values of primary producers at the base of aquatic and terrestrial food webs overlap, but values for freshwater phytoplankton are typically depleted in 13C relative to freshwater macrophytes37, and both types of aquatic producer are depleted in 13C relative to terrestrial C3 plants37, 38, 39. Consumers foraging within food webs fuelled by freshwater phytoplankton (for example freshwater and brackish-water foraging Eocene whales) typically have lower 13C values than species foraging on aquatic macrophytes37 or on terrestrial resources33 (Fig. 4). Enamel 13C values for Indohyus are higher than those for most early cetaceans and are most similar to the 13C values in enamel for terrestrial mammals from early and middle Eocene deposits in India and Pakistan. Indohyus could have been feeding on land or in water, but it was clearly eating something different from archaeocetes such as Pakicetus and Ambulocetus. If the large crushing basins in the molars of Indohyus were used for processing vegetation, these 13C values in enamel could come from the ingestion of terrestrial plants or aquatic macrophytes. Alternatively, a more ominivorous diet would suggest that Indohyus might have foraged on benthic, aquatic invertebrates in freshwater systems. Although we cannot exclude the possibility of aquatic foraging by Indohyus, 13C values in enamel do suggest that the diet of Indohyus differed significantly from that of Eocene whales. A more refined interpretation of the dietary preferences of Indohyus will require a study of tooth wear and tooth morphology.
Evolutionary hypothesis for whale origins
Indohyus was a small, stocky artiodactyl, roughly the size of the raccoon Procyon lotor (Fig. 5). It was not an adept swimmer; instead it waded in shallow water, with its heavy bones providing ballast to keep its feet anchored. Indohyus may have fed on land, although a specialized aquatic diet is also possible.
Figure 5: Skeletal reconstruction of Indohyus.
Hatched elements are reconstructed on the basis of related taxa.
High resolution image and legend (62K)The modern artiodactyl morphologically most similar to Indohyus is probably the African mousedeer Hyemoschus aquaticus. Hyemoschus lives near streams and feeds on land, but flees into the water when danger occurs40. Indohyus had more pronounced aquatic specializations than Hyemoschus does, and it probably spent a considerably greater amount of time in the water either for protection or when feeding. As indicated by the evidence from stable isotopes, Indohyus spent most of its time in the water and either came onshore to feed on vegetation (as the modern Hippopotamus does) or foraged on invertebrates or aquatic vegetation in the same way that the modern muskrat Ondatra does.
Raoellids are the sister group to cetaceans, and this implies that aquatic habitats originated before the Order Cetacea. The great evolutionary change that occurred at the origin of cetaceans is thus not the adoption of an aquatic lifestyle. Here we propose that dietary change was the event that defined cetacean origins; this is consistent with the cranial and dental synapomorphies identified. Molars of Indohyus are markedly different from those of pakicetids, and it is widely assumed that pakicetids ate aquatic prey18, 23.
Our working hypothesis for the origin of whales is that raoellid ancestors, although herbivores or omnivores on land, took to fresh water in times of danger. Aquatic habits were increased in Indohyus (as suggested by osteosclerosis and oxygen isotopes), although it did not necessarily have an aquatic diet (as suggested by carbon isotopes). Cetaceans originated from an Indohyus-like ancestor and switched to a diet of aquatic prey. Significant changes in the morphology of the teeth, the oral skeleton and the sense organs made cetaceans different from their ancestors and unique among mammals.
Methods Summary
We chose an existing character matrix10 as the basis for our phylogenetic analysis. We corrected scores of some of the taxa, and made some changes in the taxa included. Details on these taxa, the rationale for using them, and their scores are given in Methods and in Supplementary Table 1.
Tympanic wall thickness was investigated to address the presence of the involucrum quantitatively. We calculated the ratio of medial tympanic wall thickness divided by lateral tympanic wall thickness (see Methods and Supplementary Table 2). Bone histology was studied to investigate the presence of osteosclerosis. It was quantified as the ratio of medullary cavity width divided by bone width in the mediolateral plane (see Methods and Supplementary Table 3).
For analysis of stable isotopes, we prepared powders by following published methods (see Methods and Supplementary Tables 4 and 5). Multiple samples were collected for each species to provide an estimate of population means for carbon and oxygen isotope values33. About 5 mg of enamel powder was collected from each specimen (tooth) for study of carbon and oxygen isotope values.
Most fossils of Indohyus were collected by the late Indian geologist A. Ranga Rao, who discovered the locality about 25 years ago (acronym RR); additional fossils were collected by S.B. and B.N.T. at the same locality (acronym IITR-SB-Kal-S).
Full methods accompany this paper.
m by using a precision grinder with 600, 800 and 1,200 grit paper
(Rsample/Rstandard - 1), where Rsample is the observed isotope ratio of the sample (13C/12C or 18O/16O) and Rstandard is the accepted ratio for an appropriate international standard (Vienna Pee Dee belemnite for 
).
