process, reduced zygomatic arch, long nar- row muzzle, broad supraorbital process, and teeth that resemble those of other archaeo- cetes, the paraphyletic stem group of ceta- ceans. Archaic archaeocete whales have been found in Africa (16, 17) and North America (1 8, 19), but are best known from Pakistan and India (2, 20-22), and it is likely that cetaceans originated near the subcontinent. Thus, the skeletal morphol- oev of Ambulocetus is critical to our under- u, standing of locomotion in early cetaceans. The closest terrestrial relatives of whales, mesonychids, were running mammals (1 0, 23) that, when swimming, probably paddled by flexing and extending their hindlimbs al- ternately as in extant land mammals (24). At some point, cetaceans switched from unilat- eral paddling to bilateral (spinal) undulating (25) and from using the feet as a primary propulsive surface to having a tail fluke. Am- bulocetw shows that spinal undulation evolved before the tail fluke. These fossils also test several hypotheses concerning early whale locomotion. The greatly expanded feet imply that forelimb propulsion is not primitive for cetaceans even though it mav occur in other u protocetids (12). It corroborates that ceta- ceans have gone through a stage that com- bined hindlimb paddling and spinal undula- tion, resembling the aquatic locomotion of fast swimming otters (25, 26). Unlike modem cetaceans, Ambtllocetus certainly was able to walk on land, probably in a way similar to modern sea lions or fur seals. In water, it combined asvects of the locomotion of modern seals, otters, and cetaceans: Like modern cetaceans it swam by moving its spine up and down, but like seals, the main propulsive surface was provided by its feet. As such, Ambulocetus represents a critical intermediate between land mam- mals and marine cetaceans. REFERENCES AND NOTES 1. R. M. West, J. Paleontol. 54, 508 (1980). 2. J. G M. Thewissen and S T Hussain, Nature361, 444 (1993) 3 Order Cetacea, Suborder Archaeoceti, Family Protocetidae Ambulocetus natans, n. gen., n. sp. Holofype. Howard Geological Survey of Pakistan 18507. skull with left P4I-MI-21 and lacking the rostrum, but preserving both tympan~cs, bas~hyo~d, right posterlor mandible with PI4, MI2-3, and left ramus with alveoli for CI1-PI3, half atlas, fragments of three other cervical vertebrae, two complete and several fragmentary thorac~c and one lumbar verte- bra, three complete ribs, and a sternebra. Forel~mb fragments Include part of the glenoid, left and r~ght radius and ulna, all left carpals except the tri- quetrum, all metacarpals, and at least four prox~mal, two Intermediate phalanges, and a fragment of one d~stal phalanx Hindlimb elements Include complete femur, proximal t~bia, distal fibula, trochlea of as- tragalus, metatarsals Il-V and at least two prox~mal, three intermed~ate, and three d~stal phalanges. The holotype was found at locality HGSP 9209 In the upper Kuldana Formation,Punjab, 3.7 km northwest of Ganda Kas (72"12'2O"E, 33"39'N). Referred ma- terial. PI2 (HGSP 18473), Pi3 (HGSP 18497), caudal vertebra (HGSP 18472), and d~stal femur (HGSP 18476), all from locality 9207, about 5 m above the type locality in the section. All mater~alwill be housed at the Geological Survey of Pak~stanin Islamabad. Known d~stribution. Kala Ch~tta Hills of Pakistan, upper Kuldana Formation and trans~tional beds to the Kohat Formation. Lower to m~ddle Eocene. Differentialdiagnosis of genus. P41 of Am- bulocetus has a single h ~ g h lab~alcusp, unlike Protocetus(it?), but lacks a protocone. M l I and M21 of Ambulocetus are sim~lar in morphology, bear~ng a high and connate para- and metacone, and a pro- tocone that IS on a broad and low lingual shelf, unlike remingtonocetids where this shelf is narrow. The protocone of Ambulocetus is weaker than that of Pakicetus (1) and lchthyolestes [R. Dehm and T, zu Oettingen-Spielberg, Bayer. Akad. Wiss. Math. Naturwiss. KI. Abh. 91, 1 (1958)], while the cusp is absent In Protocetus. This cusp IS well set off from the labial cusps, unlike Indocetus. Crests are poorly developed on the teeth of Ambulocetus, as in their mesonychid ancestors [F. S. Szalay, Evolution 23, 703 (1969)l. The lower canine is large and single rooted, whereas the four lower premolars have two roots, unlike remingtonocet~ds. PI4 consists of a single high cusp, unlike Pakicetus. The trigonid of MI2 and MI3 is much higher than the talon~d, unlike Gandakasia and Pappocetus (15). It lacks a meta- conid, unl~ke Pakicetus Ambulocetus lacks the tu- bercles that occur rostra1 to the protoconid of Pap- pocetus. Few details of the talonid remain, but there seems to have been a single cusp and no basin, as in Pakicetus. Unlike most other archaeocetes, the pterygoid processes are enormous and their dorso- ventral extent matches that of the braincase Ety- mology The genus name IS a combination of ambulare (to walk) and cetus (whale), In recog- nition of a characterist~c mode of locomotion in th~s cetacean. The species indication, natans (swimming), describes another aspect of ~ t s locomotor reperto~re. 4. J. G. M Thewissen, Natl. Geogr. Res. Expl. 9, 125 (1993). Figure 12 shows the type locality for Ambulocetus natans. Its captlon IS wrong. 5. L. Van Valen, Bull. Am. Mus Nat Hist. 132, 1 (1 966) 6. A. Wyss, Nature 347, 428 (1990). 7 D A Pabst, J Zool 230, 159 (1993). 8. E. J Slijper, Cap~ta 2001. (The Hague) 6 7 , 1 11 9361 \ ---, 9. Pachyaena has 15 caudal vertebrae (22), Basilo- saurus has 21 (it?), z~ph~ids have 19, and physe- terids have 24 10. W. D. Matthew and W Granger, Bull. Am. Mus. Nat. Hist. 23, 1 (1 915) 11 F. E. Fish, Mammal Rev 21, 181 (1991). 12. D. P. Aleshire, J. Vertebr Paleontol. Suppl 13, 24A (1 993) 13. F. J Tarasoff, A. Bisaillon, J P~erard, A. P. Wh~tt, Can J. 2001. 50, 915 (1972). 14. A. W English, J. Zool. 178, 341 (1976). 15. M. P. Beentjes, Zool J. L~nn. Soc. 98, 307 (1990) 16. C. W Andrews, Proc 2001. Soc. London 1919, 309 (1919). 17. R. Kellogg, Carnegie Inst. Wash~ngton. Publ. 482, 1 11936). 18. R 'c. ~Llbert Jr, and R. M. Petkewich, J. Vertebr. Paleontol. Suppl. 11, 36A (1991). 19. R. C. Hulbert Jr., ib~d. 13, 42A (1993) 20. K. Kumar and A. Sahn~, J. Vertebr. Paleontol. 6, 326 (1986). 21. P. D. Gingerich, S. M. Raza, M Ar~f, M. Anwar, X. Zhou, Contrib. Mus. Paleontol. Univ. Mich 28, 393 (1 993) 22. P. D. Ginger~ch, N. A. Wells, D. E. Russell, S. M. I. Shah, Science 220, 403 (1983). 23. X. Zhou, W. J. Sanders, P. D Ginger~ch, Contrib. Mus. Paleontol. Univ. Mich. 28, 289 (1 992) 24. F. E. Fish, J. Mammal. 74, 275 (1993) 25. , in Mammalian Energetics Interdiscipli- nary V~ews of Metabolism and Reproducbon, T. E. Tomasi and T. H Horton, Eds. (Cornell Un~v. Press, Ithaca, NY, 1992), chap 3 26. A. B. Howell, Aquatic Mammals (Thomas, Spring- field, IL, 1930). 27. We thankA. Aslan, D. P. Domning, F. E Fish, R. E. Fordyce, P. D. Gingerich, R C Hulbert, S. I Madar, S. M. Raza, K. D. Rose, A. R. Wyss, and the Geological Survey of Pakistan. F~eld work was supported by the National Geographic Society. 28 October 1993, accepted 3 December 1993 An Inverted Double Seismic Zone in Chile: Evidence of Phase Transformation in the Subducted Slab Diana Comte and Gerardo Suarez Data from two microseismic field experiments in northern Chile revealed an elongated cluster of earthquakes in the 'ubducted Nazca plate at a depth of about 100 kilometers in which down-dip tensional events were consistently shallower than a family of compres- sional earthquakes. This double seismic zone shows a distribution of stresses of opposite polarity relative to that observed in other double seismic zones in the world. The distribution of stresses in northern Chile supports the notion that at depths of between 90 to 150 kilometers, the basalt to eclogite transformation of the subducting oceanic crust induces tensional deformation in the upper part of the subducted slab and compressional defor- mation in the underlying mantle. Since the advent of plate tectonics, inter- mediate and deep earthquakes have been interpreted as evidence of cold lithosphere penetrating into the mantle. At intermedi- ate depths, most subducted lithospheres exhibit down-dip tensional faulting, which has been generally interpreted as resulting from the gravitational pull of the slab (1- 6). The presence of a more complex state of stress in the subducted slab was observed first in Tohoku, Japan (7). There, a sheet of compressional earthquakes lies above a sheet of down-dip tensional events. These two seismic planes are separated by -40 km at a depth of -60 km, and they merge at a depth of 200 km. Similar double-planed seismic zones have subsequently been re- ported in other subduction zones (7, 8). 21 2 SCIENCE VOL. 263 14 JANUARY 1994