Genes Genet. Syst. (1999) 74, p. 211–217 Model dependence of the phylogenetic inference: Relationship among Carnivores, Perissodactyls and Cetartiodactyls as inferred from mitochondrial genome sequences

Ying Cao1, Kyung Seok Kim2, Ji Hong Ha2, and Masami Hasegawa1 1The Institute of Statistical Mathematics, 4-6-7 Minami-Azabu, Minato-ku, Tokyo 106-8569, Japan 2Department of Genetic Engineering, Kyungpook National University, Taegu City 702-701, Korea

(Recieved 24 June 1999, accepted 12 October 1999)

Some previous analyses of mitochondrial proteins strongly support the / Perissodactyla grouping excluding Cetartiodactyla (Artiodactyla + ) as an outgroup, but the support of the hypothesis remains equivocal from the analysis of several nuclear-encoded proteins. In to evaluate the strength of the support by mitochondrial proteins, phylogenetic relationship among Carnivora, Perissodactyla, and Cetartiodactyla was estimated with the ML method by using the updated data set of the 12 mitochondrial proteins with several alternative models. The analyses demonstrate that the phylogenetic inference depends on the model used in the ML analysis; i.e., whether the site-heterogeneity is taken into account and whether the rate parameters are estimated for each individual pro- teins or for the concatenated sequences. Although the analysis of concatenated sequences strongly supports the Carnivora/Perissodactyla grouping, the total evalu- ation of the separate analyses of individual proteins, which approximates the data better than the concatenated analysis, gives only ambiguous results, and therefore it is concluded that more data are needed to resolve this trichotomy.

Cetacea into Ferungulata but included , INTRODUCTION which is Proboscidea (elephant) + Hyracoidea (hyrax) + Molecular phylogenetics has become a powerful tool in Sirenia (manatee and dugong), as well as Tubulidentata clarifying evolutionary history of placental , but (aardvark). It is now evident from the molecular data there still remain a lot of controversies concerning the that Paenungulata and Tubulidentata form a clade called deep branchings of this group of (e.g., Waddell et together with elephant shrew and golden mole al., 1999a, 1999b). Probably one of the most robust con- (and tenrec), and that they are outside the clade clusions of molecular phylogenetics concerning inter-ordi- formed by Cetartiodactyla, Perissodactyla and Carnivora nal relationships of placental mammals would be the (Stanhope et al., 1996, 1998a, 1998b; Springer et al., 1997, existence of the clade formed by Cetartiodactyla (Cetacea 1999; de Jong, 1998). Since Ferungulata as defined by + Artiodactyla; e.g., Cao et al., 1994; Stanhope et al., 1996; Simpson (1945) includes Paenungulata, Waddell et al. Gatesy et al., 1996, 1999; Shimamura et al., 1997; Graur (1999b) proposed a superordinal taxon Fereuungulata et al., 1997; Ursing and Arnason, 1998), Perissodactyla, (Carnivora + the true excluding Paenungulata) and Carnivora (e.g., Krettek et al., 1995; de Jong, 1998; which consists of Cetartiodactyla, Perissodactyla, Car- Cao et al., 1998). This clade is often called Ferungulata nivora, and Pholidota (). Chiroptera () are (e.g., Xu et al., 1996b; Arnason et al., 1997; Cao et al., probably a sister taxon to Fereuungulata (Pumo et al., 1998). 1998; Waddell et al., 1999a). On the other hand, recent analyses of morphological Analysis of complete mitochondrial DNA (mtDNA) data do not indicate such a clade (Novacek, 1992; sequences is becoming a new standard for studying deep McKenna and Bell, 1997), although the association of Car- branchings of the placental mammals (Penny and nivora, Perissodactyla, and Artiodactyla was already rec- Hasegawa, 1997), but data are still not available from ognized by Simpson (1945). Simpson did not include Pholidota, and therefore we do not discuss on Pholidota in this paper. The existence of Fereuungulata is robust * Corresponding author. from the molecular evidence, but the relationship among 212 Y. CAO et al.

Carnivora, Perissodactyla, and Cetartiodactyla could not Phylogenetic analyses of mitochondrial proteins be resolved by the previous analyses even though whole (mt-proteins) sometimes give strong support for the mtDNA data were used (Krettek et al., 1995; Adachi and Carnivora/Perissodactyla clade (Tree-1 in Fig. 1). For Hasegawa, 1996a; D’Erchia et al., 1996; Kim et al., 1998) example, Xu et al.’s (1996b) analysis of mt-protein genes and a lot of nuclear genes in addition to mitochonrial ones gave 82.9, 98.0, and 90.8% bootstrap supports for this were used (Graur et al.,1997). clade, respectively, by the MP, NJ, and ML of amino acids,

Fig. 1. Three possible trees among Carnivora, Perissodactyla, and Cetartiodactyla. Model dependence of phylogenetic inference 213 although the authors noted that the more traditional view X63726), Halichoerus grypus (grey seal; X72004), Felis of the Perissodactyla/Cetartiodactyla clade (Tree-2 in Fig. catus (cat; Lopez et al., 1996 ; U20753), Canis familiaris 1) could not be rejected at the 5% level and the support is (dog; Kim et al., 1998 ; U96639), Equus caballus (horse; less significant by the nucleotide sequence analysis. A X79547), Equus asinus (donkey; Xu et al., 1996a; X97337) high bootstrap support for the clade was also given by Rhinoceros unicornis (Indian rhinoceros; Xu et al. 1996b; Reyes et al.’s (1998) parsimony analysis of first and sec- X97336), Ceratotherium simum (white rhinoceros; Xu and ond codon positions of mt-protein genes (92%). Further- Arnason, 1997; Y07726), Artibeus jamaicensis (Jamaican more, amino acid LogDet NJ analysis gave 96 and 93% fruit-eating bat; Pumo et al., 1998; AF061340), Dasypus bootstrap support, respectively, with all sites included and novemcinctus (nine-banded armadillo; Arnason et al., with invariant sites removed (Waddell et al., 1999b). 1997; Y11832), Loxodonta africana (African elephant; However, the support of the hypothesis remains equivocal Hauf, unpublished; AJ224821), Orycteropus afer (aard- from the analysis of several nuclear-encoded proteins vark; Arnason et al., 1999; Y18475). (Graur et al., 1997). If the Carnivora/Perissodactyla The 12 proteins encoded in the same strand of mtDNA clade is real, this has a great biological implication for the were prepared for phylogenetic analyses. Alignments way how apparently similar characteristics of ungulates were carefully checked by eye, and all positions with gaps evolved in Artiodactyla and Perissodactyla. Phylogenetic or ambiguous alignment plus overlapping regions between analysis generally depends on the species sampling, and ATP6 and ATP8 and between ND4 and ND4L were the conclusion drawn from a limited number of species is excluded from the phylogenetic analyses. The total num- sometimes misleading (Philippe and Douzery, 1994; ber of remaining codons is 3403. Adachi and Hasegawa, 1996c; Halanych, 1998). There- fore, we now study the problem of the Carnivora- Phylogenetic methods. The ProtML program in Perissodactyla-Cetartiodactyla trichotomy in detail by MOLPHY package (ver. 2. 3) (Adachi and Hasegawa using the updated data set of whole mitochondrial 1996b), the CodeML program in PAML package (Yang genomes. 1997) were applied to the mt-protein sequences with the In this study, we use the ML method for phylogenetic mtREV-F model (Adachi and Hasegawa 1996a). For the inference (Felsenstein, 1981) to analyze amino acid CodeML program, the discrete Γ-distribution model with sequences of mt-proteins (Kishino et al., 1990; Adachi and 8 categories was used to represent the site-heterogeneity. Hasegawa, 1996a). In spite that the ML method has a To evaluate the total evidence of separate analyses of sound statistical basis, it has been shown that the method individual genes, the TotalML program in MOLPHY was can be inconsistent when all sites are incorrectly assumed applied to the output files of the ProtML or CodeML pro- to be free to vary (Lockhart et al., 1996; Yang, 1996; grams. Sullivan and Swofford, 1997). Furthermore, not only the presence of invariable sites but also the rate heterogene- RESULTS ity among variable sites and among different genes may cause inconsistency if these factors are not taken into Fig. 2 shows the ProtML tree of the concatenated 12 account. Therefore, we take account of rate-heterogene- mt-proteins from the Fereuungulata species and ity among sites (Yang, 1997) as well as among different Chiroptera (bat) with Afrotheria (elephant + aardvark) genes. It is shown that the heterogeneity among genes is and Edentata (armadillo) as an outgroup. It has recently not sufficiently well approximated by a single discrete been suggested that Edentata is close to Fereuungulata Γ-distribution model of concatenated sequences, and that (Arnason et al., 1997) and that Afrotheria and Edentata the total evaluation of the separate analyses of indiviual form a monophyletic clade in the placental mammals and genes is preferable to the concatenated analysis. that the splitting of these two groups corresponds to the separation of African and South American Continents some 100 Myr ago (Stanhope et al., 1998; Waddell et al., MATERIALS AND METHODS 1999b, 1999c). Therefore, our choice of this outgroup Sequence data. The complete mtDNA sequences used seems reasonable, although it holds even if these sugges- in this study are from the following 18 species (only data- tions are wrong because the tree is essentially base accession numbers are given for references prior to unrooted. Fig. 2 was estimated by using the local rear- 1996); Bos taurus (cow; database accession number rangement option of the ProtML with several alternative V00654), Ovis aries (sheep; Hiendleder et al., 1998; initial tree topologies, and the figure on each edge repre- AF010406), Sus scrofa (pig; Ursing and Arnason, 1998; sents a local bootstrap probability (LBP; Adachi and AJ002189), Hippopotamus amphibius (hippopotamus; Hasegawa 1996b), which is a relative bootstrap frequency Ursing and Arnason, 1998b; AJ010957), Balaenoptera among the three alternative relationships relevant to the physalus (fin whale; X61145), Balaenoptera musculus edge when the relationships within the three subtrees at- (blue whale; X72204), Phoca vitulina (harbor seal; tached to the particular branch are fixed. 214 Y. CAO et al.

catenated and separate analyses with the Γ-distribution gave AIC values of 2 × 29,999.6 + 2 × 53 = 60,105.2 and 2 × 29,409.2 + 2 × 53 × 12 = 60,090.4, respectively. This indi- cates that the separate analysis approximates the under- lying evolutionary process better than the concatenated analysis, which does not explicitly assume heterogeneity of substitution process across genes, and this holds even if the site-heterogeneity is taken into account with the Γ-distribution. The separate analyses with the Γ-distri- bution for each of the individual proteins turned out to best approximate the data given the present models, and the result of the analyses should hopefully be more reli- able than those which give higher AIC values. Although the concatenated analysis with the Γ-distri- bution rather strongly supports the Carnivora/ Perissodactyla grouping (95%BP), we cannot rubber- stamp this suggestion, because a more appropriate model (in terms of AIC) of separate analyses of individual genes Fig. 2. The ProtML tree of the concatenated 12 mt-proteins from gives only weaker support of 72%BP (Table 1). It is the Fereuungulata species and Chiroptera with Afrotheria and interesting to note that the total evaluation of the sepa- Edentata as an outgroup. For each internal branch, a local boot- strap proportion (LBP; %) estimated by the RELL method rate analyses with site-homogeneity within each gene (Kishino et al. 1990) with 104 replications is shown. The hori- gives Tree-2 as the ML tree (although the log-likelihood zontal length of each branch is proportional to the estimated num- bers of amino acid substitutions. Table 1. Comparison of log-likelihood of trees among Car- nivora, Perissodactyla, and Cetartiodactyla for the 12 proteins with the mtREV-F model. Consistently with the previous works (Gatesy et al., concatenated separate 1996, 1999; Shimamura et al., 1997; Graur et al., 1997; Tree Ursing and Arnason, 1998; Nikaido et al., 1999), the clos- without Γ with Γ without Γ with Γ est living relative of Cetacea is hippopotamus and the sec- Tree-1 <-32,181.8> <-29,999.6> -4.3 ± 10.9 <-29,409.2> ond closest are ruminants (cow and sheep) excluding pig (0.8159) (0.9540) (0.3995) (0.7188) as an outgroup, and this relationship gains high bootstrap Tree-2 -13.5 ± 15.0 -12.7 ± 7.8 <-31,260.7> -5.7 ± 7.3 (0.1841) (0.0448) (0.5995) (0.2775) support. Tree-3 -33.9 ± 11.2 -14.7 ± 7.1 -26.2 ± 14.7 -14.6 ± 9.7 Table 1 compares log-likelihood of the three trees in Fig. (0.0000) (0.0012) (0.0010) (0.0037) 1 for the 12 mt-proteins with the mtREV-F model. It The log-likelihoods, ln L, of the ML tree are given in angle turned out that the result depends on the model used in brackets, and the differences, ∆ln L, of alternative trees from the analysis; that is, whether the site-heterogeneity that of the ML tree are shown with their SEs (following ±), is used and whether the ML estimation is done for which were estimated by the formula of Kishino and cancatenated sequences or for individual genes. Hasegawa (1989). Bootstrap proportions in parentheses were 4 The adequacy of models were compared by using the estimated by the RELL method (Kishino et al. 1990) with 10 replications. “Separate” means that ln L of each gene is sepa- Akaike Information Criterion (AIC), where the AIC score rately estimated and then is summed up. ML estimates of × × =–2 ln L + 2 (number of parameters). The model that shape parameter a of the Γ-distribution for Tree-1 (no signifi- minimizes AIC is considered to be the most appropriate cant differences for other trees; data not shown) and tree model (Akaike 1974; Sakamoto et al., 1986). The mtREV- length for Tree-1 (a measure of relative evolutionary rate of F model uses the amino acid frequencies of the data (the each protein) are 0.29 and 1.86 (concatenated sequences), 0.31 and 1.48 (ND1), 0.61 and 3.64 (ND2), 0.15 and 0.54 number of parameters is 19) and, for a tree with 18 (COI), 0.23 and 1.08 (COII), 1.04 and 7.59 (ATP8), 0.38 and species, 33 branch lengths must be estimated. For the 2.27 (ATP6), 0.20 and 1.19 (COIII), 0.32 and 1.68 (ND3), 0.57 Γ-distribution model, additionally one parameter, α (shape and 1.79 (ND4L), 0.31 and 2.25 (ND4), 0.37 and 2.55 (ND5), parameter of the distribution), must be estimated. 0.26 and 1.76 (COB). Therefore, the number of estimated parameters becomes 53 (= 33 + 19 + 1) for each ML analysis with the Γ-distribu- Table 2. Comparison of AIC of Tree-1 in Table 1 among the tion or 52 with the site-homogeneity. For Tree-1 and the different models of analyses. concatenated analysis of the 12 proteins with the site-ho- concatenated separate mogeneity, AIC was 2 × 32,181.8 + 2 × 52 = 64,467.6, while without Γ with Γ without Γ with Γ AIC for the separate analysis was reduced to 2 × 31,265.0 + 2 × 52 × 12 = 63,778.0 (Table 2). Furthermore, the con- 64,467.6 60,105.2 63,778.0 60,090.4 Model dependence of phylogenetic inference 215

Table 3. Comparison of log-likelihood of trees among Pinnipedea, Canoidea and Feloidea for the 12 proteins with the mtREV-F model. concatenated separate Tree without Γ with Γ without Γ with Γ Tree-A <-32,181.8> <-29,999.6> <-31,265.0> <-29,409.2> (0.9315) (0.9411) (0.8503) (0.7660) Tree-B -26.0 ± 16.7 -16.0 ± 10.6 -21.9 ± 15.3 -10.1 ± 10.3 (0.0489) (0.0561) (0.1376) (0.1989) Tree-C -29.8 ± 16.0 -21.5 ± 9.3 -36.2 ± 19.8 -17.1 ± 11.7 (0.0196) (0.0028) (0.0121) (0.0351) Tree-A is the same as Tree-1 in Table 1; that is, the Carnivora/Perissodactyla clade is assumed. Tree-A: ((Canoidea, Pinnipedea), Feloidea); Tree-B: ((Feloidea, Pinnipedea), Canoidea); Tree-C: ((Canoidea, Feloidea), Pinnipedea) difference from Tree-1 is very minor). Thus, the tri- or may suggest a wrong tree because of a sampling chotomy problem among Carnivora, Perissodactyla, and error. Therefore, it is necessary to scrutinize as many Cetartiodactyla could not be resolved by the mt-protein different loci as possible and to evaluate the total analysis. Since Graur et al.’s (1997) ML analysis of pro- evidence. The ML method is particularly suitable for this teins encoded by nuclear DNA as well as mtDNA supports purpose. Given the model, one can calculate the likeli- Tree-2 with 73.2%BP, while Tree-1 has only 0.9%BP, Tree- hood as the probability that one tree yielded the observed 1 might be unlikely, even though some previous analyses data, and each gene can reasonably be regarded as evolv- of mt-proteins strongly supported Tree-1. ing independently from other genes. Therefore, the Carnivora have been traditionally grouped into two total support for a particular tree can be evaluated by sim- superfamilies: the Canoidea (or Arctoidea; bear, dog, rac- ply summing up the estimated log-likelihoods of individual coon, and weasel families) and the Feloidea (or Aeluroidea; genes for that tree, and the total log-likelihoods for differ- cat, hyena, mongoose, and civet families). Although an ent trees can then be compared (Adachi and Hasegawa, independent order is sometimes assigned to the pinnipeds, 1996b; Hasegawa et al., 1997). Importantly, as the morphological evidence suggests that they should be clas- present work has demonstrated, the analyses of concat- sified within the Carnivora closely allied to the Canoidea enated sequences from several genes do not sufficiently (Simpson, 1945). Molecular evidence is in favor of this take into account the differences of tempo and mode of suggestion (Wayne et al., 1989), and the mt-protein data evolution among different genes, even if the site-heteroge- analyzed in this paper is in accord to the pinniped/canoid neity is approximated by the Γ-distribution model. On grouping excluding feloid as an outgroup (Tree-A in the other hand, if we analyze the different genes sepa- Table 3). In the analysis of Table 3, the Carnivora/ rately by the ML method, we can take account of these Perissodactyla clade (Tree-1 in Fig. 1) is assumed, but the differences. In the framework of parsimony, however, it assumption of the Perissodactyla/Cetartiodactyla clade is difficult to combine different genes because weighting (Tree-2) gives essentially the same result (data not among them must be ambiguous. Therefore, the avail- shown). In this table, while the support of the well-es- ability of the total evidence approach might be one of the tablished grouping of the Canoidea/Pinnipedea is high most important merits of ML. with the concatenated analysis, it reduces a little bit with The present study demonstrates that the phylogenetic a more appropriate model of separate analyses. There is inference depends on the model used in the ML analysis; a tendency for an ML analysis based on a more realistic i.e., whether the site-heterogeneity is taken into account model to give a result with less significance than one based and whether the rate parameters are estimated for each on a less realistic model (Hasegawa and Adachi, individual proteins or for the concatenated sequences, and 1996). Although this may seem disadvantageous, this highlights the importance of developing more appropriate apparent demerit can be taken as the cost (like an insur- models in approximating the data. ance premium) of guarding against a misleading conclu- sion (Cao et al., 1998). We thank Ms. Mitsuko Kitahara for drawing Fig. 1. We also thank Dr. Naruya Saitou and two anonymous reviewers for use- ful comments. This work was supported by grants from the Min- DISCUSSION istry of Education, Science, Sports and Culture of Japan (M. H. Although the analysis of molecular sequences has and Y. C.), Japan Society for the Promotion of Science (Y. C.), and the Korea Science and Foundation (J. H. H. and K. S. K.). become a powerful tool in elucidating the phylogenetic his- tory of organisms, a single gene does not necessarily con- tain sufficient information to resolve the problem at hand 216 Y. CAO et al.

REFERENCES mum likelihood estimate of the evolutionary tree topologies from DNA sequence data, and the branching order in Adachi, J., and Hasegawa, M. (1996a) Model of amino acid sub- Hominoidea. J. Mol. Evol., 29, 170–179. stitution in proteins encoded by mitochondrial DNA. J. Mol. Kishino, H., Miyata, T., and Hasegawa, M. (1990) Maximum like- Evol., 42, 459–468. lihood inference of protein phylogeny, and the origin of Adachi, J., and Hasegawa, M. (1996b) MOLPHY Version 2.3: Pro- chloroplasts. J. Mol. Evol., 31, 151–160. grams for Molecular Phylogenetics Based on Maximum Like- Krettek, A., Gullberg, A., and Arnason, U. (1995) Sequence analy- lihood. Computer Science Monographs, No. 28. Institute of sis of the complete mitochondrial DNA molecule of the hedge- Statistical Mathematics, Tokyo. hog, Erinaceus europaeus, and the phylogenetic position of Adachi, J., and Hasegawa, M. (1996c) Instability of quartet analy- the Lipotyphla. J. Mol. Evol., 41, 952–957. ses of molecular sequence data by the maximum likelihood Lockhart, P., Larkum,A., Steel, M., Waddell, P., and Penny,D. method: the Cetacea/Artiodactyla relationships. Mol. Phyl. (1996) Evolution of chlorophyll and bacteriochlorophyll: The Evol., 6, 72–76. problem of invariant sites in sequence analysis. Proc. Natl. Akaike, H. (1974) A new look at the statistical model iden- Acad. Sci. USA, 93, 1930–1934. tification. IEEE Trans. Autom. Contr., AC-19, 716–723. Lopez, J., Culver, M., Cevario, S., and O’Brien, S. (1996) Com- Arnason, U., Gullberg, A., and Janke,A. (1997) Phylogenetic plete nucleotide sequences of the domestic cat (Felis catus) analyses of mitochondrial DNA suggest a sister group rela- mitochondrial genome and a transposed mtDNA repeat tionship between (Edentata) and Ferungulates. (Numt) in the nuclear genome. Genomics, 33, 229–246. Mol. Biol. Evol., 14, 762–768. McKenna, M., and Bell,S. (1997) Classification of Mammals: Above Arnason, U., Gullberg,A., and Janke,A. (1999) The mitochondrial the Species Level. Columbia Univ. Press, New York. DNA molecule of the aardvark, Orycteropus afer, and the Nikaido, M., Rooney, A. P., and Okada, N. (1999) Phylogenetic position of the Tubulidentata in the eutherian tree. Proc. R. relationships among cetartiodactyls based on insertions of Soc. London, 266, 339–345. short and long interpersed elements: Hippopotamuses are the Cao, Y., Adachi, J., Janke, A., Pääbo, S., and Hasegawa, M. (1994) closest extant relatives of whales. Proc. Natl. Acad. Sci. Phylogenetic relationships among eutherian orders estimated USA, 96, 10261–10266. from inferred sequences of mitochondrial proteins: Instabil- Novacek, M. (1992) Mammalian phylogeny: shaking the tree. ity of a tree based on a single gene. J. Mol. Evol., 39, 519– Nature, 356, 121–125. 527. Penny ,D. and Hasegawa, M. (1997) Molecular systematics: the Cao, Y., Janke,A., Waddell, P., Westerman,M., Takenaka,O., platypus put in its place. Nature, 387, 549–550. Murata,S., Okada, N., Pääbo,S.. and Hasegawa, M. (1998) Philippe, H. and Douzery, E. (1994) The pitfalls of molecular phy- Conflict among individual mitochondrial proteins in resolv- logeny based on four species, as illustrated by the Cetacea/ ing the phylogeny of eutherian orders. J. Mol. Evol., 47, 307– Artiodactyla relationships. J. . Evol., 2, 133–152. 322. Pumo, D., Finamore, P., Franek, W., Phillips, C., Tarzami, S., and de Jong, W. (1998) Molecules remodel the mammalian tree. Balzarano, D. (1998) Complete mitochondrial genome of a Trends in Ecol. & Evol., 13, 270–275. neotropical fruit bat, Artibeus jamaicensis, and a new hypoth- D’Erchia, A., Gissi,C., Pesole,G., Saccone,C., and Arnason,U. esis of the relationships of to other eutherian (1996) The guinea-pig is not a rodent, Nature, 381, 597–600. mammals. J. Mol. Evol., 47, 709–717. Felsenstein, J. (1981) Evolutionary trees from DNA sequences: a Reyes, A., Pesole, G., and Saccone, C. (1998) Complete mitochon- maximum likelihood approach. J. Mol. Evol., 17, 368–376. drial DNA sequence of the fat dormouse, Glis glis: further Gatesy, J., Hayashi, C., Cronin, M., and Arctander, P. (1996) Evi- evidence of rodent paraphyly. Mol. Biol. Evol., 15, 499–505. dence from milk casein genes that cetaceans are close rela- Sakamoto, Y., Ishiguro, M., and Kitagawa, G. (1986) Akaike tives of hippopotamid artiodactyls. Mol. Biol. Evol., 13, 954– Information Criterion Statistics. D. Reidel Publ. Comp., 963. Dordrecht. Gatesy, J., Milinkovitch, M., Waddell, V., and Stanhope, M. (1999) Shimamura, M., Yasue, H., Ohshima,K., Abe, H., Kato, H., Stability of cladistic relationships between Cetacea and Kishiro, T., Goto, M., Munechika, I., and Okada, N. (1997) higher-level artiodactyl taxa. Syst. Biol., 48, 6–20. Molecular evidence from retroposons that whales form a clade Graur, D., Gouy, M., and Duret, L. (1997) Evolutionary affinities within even-toed ungulates. Nature, 388, 666–670. of the order Perissodactyla and the phylogenetic status of the Simpson, G. (1945) The principles of classification and a classifi- superordinal taxa Ungulata and Altungulata. Mol. Phyl. cation of mammals. Bull. Amer. Mus. Nat. Hist., 85, 1–350. Evol., 7, 195–200. Springer, M., Amrine, H., Burk, A., and Stanhope, M. (1999) Ad- Halanych, K. (1998) Lagomorph misplaced by more characters and ditional support for Afrotheria and Paenungulata, the per- fewer taxa. Syst. Biol., 47, 138–146. formance of mitochondrial versus nuclear genes, and the im- Hasegawa, M., and Adachi, J. (1996) Phylogenetic position of pact of data partitions with heterogeneous base composition. cetaceans relative to artiodactyls: Reanalysis of mitochondrial Syst. Biol., 48, 65–75. and nuclear sequences. Mol. Biol. Evol., 13, 710–717. Springer, M., Cleven, G., Madsen,O., de Jong, W., Waddell,V., Hasegawa, M., Adachi, J., and Milinkovitch, M. (1997) Novel phy- Amrine, H., and Stanhope, M. (1997) Endemic African mam- logeny of whales supported by total molecular evidence. J. mals shake the phylogenetic tree. Nature, 388, 61–64. Mol. Evol., 44 (Suppl. 1), 117–120. Stanhope, M., Smith, M., Waddell, V., Porter, C., Shivji, M., and Hiendleder, S., Lewalski, H., Wassmuth, R., and Janke, A. (1998) Goodman, M. (1996) Mammalian evolution and the inter- The complete mitochondrial DNA sequence of the domestic photoreceptor retinoid binding protein (IRBP) gene: convinc- sheep (Ovis aries) and comparison with the other major ovine ing evidence for several superordinal clades. J. Mol. Evol., haplotype. J. Mol. Evol., 47, 441–448. 43, 83–92. Kim, K., Lee, S., Jeong, H., and Ha, J. (1998) The complete nucle- Stanhope, M., Madsen, O., Waddell, V., Cleven, G., de Jong, W., otide sequence of the domestic dog (Canis familiaris) mito- and Springer, M. (1998a) Highly congruent molecular sup- chondrial genome. Mol. Phyl. Evol., 10, 210–220. port for a diverse superordinal clade of endemic Afican Kishino, H., and Hasegawa, M. (1989) Evaluation of the maxi- mammals. Mol. Phyl. Evol., 9, 501–508. Model dependence of phylogenetic inference 217

Stanhope, M., Waddell, V., Madsen, O., de Jong, W., Hedges, S., birds and placental mammals by using whole mitochondrial Cleven, G., Kao,D., and Springer, M. (1998b) Molecular evi- protein sequences and an extended statistical framework. dence for multiple origins of Insectivora and for a new order Syst. Biol., 48, 119–137. of endemic Afican insectivore mammals. Proc. Natl. Acad. Wayne, R., Benveniste, R., Janczewski, D., and O’Brien, S. (1989) Sci. USA, 95, 9967–9972. Molecular and biochemical evolution of Carnivora. In: Car- Sullivan, J., and Swofford, D. (1997) Are guinea pigs rodents? nivore Behavior, Ecology, and Evolution. (ed. J. Gittleman), The importance of adequate models in molecular pp. 465–494. Cornell Univ. Press, Ithaca, New York. phylogenetics. J. Mammal. Evol., 4, 77–86. Xu, X., and Arnason, U. (1997) The complete mitochondrial DNA Ursing, B., and Arnason, U. (1998) Analyses of mitochondrial sequence of the white rhinoceros, Ceratotherium simum, and genomes strongly support a hippopotamus-whale clade. comparison with the mtDNA sequence of the Indian rhinoc- Proc. R. Soc. London, B265, 2251–2255. eros, Rhinoceros unicornis. Mol. Phyl. Evol., 7, 189–194. Ursing, B., and Arnason, U. (1998) The complete mitochondrial Xu, X., Gullberg, A., and Arnason, U. (1996a) The complete mito- DNA sequence of the pig (Sus scrofa), J. Mol. Evol., 47, 302– chondrial DNA (mtDNA) of the donkey and mtDNA compari- 306. sons among four closely related mammalian species-pairs. J. Waddell, P., Okada, N., and Hasegawa, M. (1999a) Towards Mol. Evol., 43, 438–446 resolving the interordinal relationships of placental mam- Xu, X., Janke, A., and Arnason, U. (1996b) The complete mito- mals. Syst. Biol., 48, 1–5. chondrial DNA sequences of the greater Indian rhinoceros, Waddell, P., Cao, Y., Hauf, J., and Hasegawa, M. (1999b) Using Rhinoceros unicornis, and the phylogenetic relationship novel phylogenetic methods to evaluate mammalian mtDNA, among Carnivora, Perissodactyla, and Artiodactyla (+ including amino acid-invariant sites-LogDet plus site strip- Cetacea). Mol. Biol. Evol., 13, 1167–1173. ping, to detect internal conflicts in the data, with special ref- Yang, Z. (1996) Phylogenetic analysis using parsimony and like- erence to the positions of hedgehog, armadillo, and lihood methods. J. Mol. Evol., 42, 294–307. elephant. Syst. Biol., 48, 31–53. Yang, Z. (1997) PAML: a program package for phylogenetic analy- Waddell, P., Cao, Y., Hasegawa, M., and Mindell, D. (1999c) As- sis by maximum likelihood. CABIOS, 13, 555–556. sessing the superordinal divergence times within