Phylogenetic Relationships and the Radiation of Sigmodontine ...

16 downloads 176 Views 3MB Size Report
Abstract. Phylogenetic relationships among South American sigmodontine rodents were examined based on the complete sequence for the mitochondrial ...
Journal of Mammalian Evolution, Vol. 6, No. 2, 1999

Phylogenetic Relationships and the Radiation of Sigmodontine Rodents in South America: Evidence from Cytochrome b Margaret F. Smith1,2 and James L. Patton1

Phylogenetic relationships among South American sigmodontine rodents were examined based on the complete sequence for the mitochondrial cytochrome b gene [1140 base pairs (bp)] for 66 species and between 759 and 1140 bp for an additional 19 species. Thirty-eight South American genera were represented, coming from eight of nine tribes. Outgroups included the North American murid rodents Peromyscus, Reithrodontomys, Scotinomys, and Neotoma, the Old World murine rodents Mus and Rattus, and the geomyoid genera Thomomys, Geomyx, Dipodomys, and Perognathus as the most distant outgroup. The South American sigmodontines were supported as a monophyletic lineage. Within this radiation several clear-cut suprageneric groupings were identified. Many of the currently recognized tribal groupings of genera were found fairly consistently, although not always with high levels of bootstrap support. The various tribes could not be linked hierarchically with any confidence. In addition, several genera stand out as unique entities, without any apparent close relatives. The overall pattern suggests a rapid radiation of the sigmodontines in South America, followed by differentiation at the tribal and generic levels. KEY WORDS: murid rodents; Sigmodontinae; mtDNA sequences; cytochrome b; phylogeny; biogeography.

INTRODUCTION New World murid rodents have fascinated and challenged neontologists and paleontologists due to the diversity of forms, the difficulty in assessing relationships, and the tantalizing potential of the group for illuminating a chapter in the biogeographic history of North and South America. The predominantly South American sigmodontines will be treated here as a separate subfamily, the Sigmodontinae (sensu Reig, 1980), distinct from a predominantly North American subfamily Neotominae, the neotomine-peromyscines (Reig, 1980), and a Central American subfamily Tylomyinae (Reig, 1984). Steppan (1995) discusses this taxonomic arrangement in detail. South American sigmodontine rodents encompass an impressive array of forms, both in numbers and in adaptive diversity. Mem1

2

Museum of Vertebrate Zoology, 3101 Valley Life Sciences Building, University of California, Berkeley, California 94720. To whom correspondence should be addressed. e-mail: [email protected] 89 1064-7554/99/0600-0089$16.00/0 © 1999 Plenum Publishing Corporation

90

Smith and Patton

bers of the group occupy a broad variety of habitats, including dry coastal deserts, temperate grasslands, cold montane woodlands, lowland rainforest, palm swamps, and tundra. Lifestyles range from semiaquatic to fossorial, arboreal, and scansorial. The taxonomy of these rodents is uncertain at all levels, from species boundaries to tribal groupings. Over the years researchers have attempted to determine relationships among members of this group, using characters such as external morphology and craniodental characters (e.g., Hershkovitz, 1962; Reig, 1987), stomach morphology (e.g., Carleton, 1973), and features of the male phallus and reproductive tract (e.g., Hooper and Musser, 1964; Voss and Linzey, 1981). Finding appropriate characters has often been a rather frustrating enterprise. Voss and Linzey (1981) note that one general problem seems to be that often taxa have been grouped together in the past by what turn out to be shared primitive characters, rather than derived features. Genetic data, most recently DNA sequence data, have the potential for providing valuable insights into relationships among the South American sigmodontines (Smith and Patton, 1991, 1993), as well as relationships among South American, North American, and Old World murids (Engel et al., 1998). Once the phylogenetic relationships among these poorly known taxa are better resolved, it becomes possible to examine patterns of biogeographic diversification and to develop a better understanding of the timing and pattern of radiation of this group. Controversies concerning the radiation revolve around questions about the geographic location for lineage diversification, and the timing of entry of ancestral forms into South America. Voss (1988, p. 470) points out that many other Neotropical vertebrates, including some fishes, frogs, birds, and bats, are comparable to South American sigmodontines in their diversity and ecological range. For these other vertebrate groups, fossil or biochemical evidence suggests a long history in South America. For the sigmodontines the earliest known fossils in South America occur in the Montehermosan fauna of Argentina, dated at about 4-5 mya (Pardinas and Tonni, 1998). "Late-arrival" hypotheses take these fossils as evidence that the ancestors of South American sigmodontines reached South America by overland dispersal after the formation of the Panamanian land bridge. Simpson (1950, 1969) assumed a very rapid radiation of a basal stock once the ancestral form reached South America. In an alternative version of the "late-arrival" hypothesis, Patterson and Pascual (1972) and Baskin (1978) suggested that some or most of the sigmodontine genera had already differentiated in tropical North and Central America before crossing over the land bridge into South America. The second "late-arrival" alternative is more consistent with the fossils, which are derived forms rather than primitive ancestral types. Hershkovitz (1966, 1972) and Reig (1980, 1984) supported an "early-arrival" hypothesis in which the ancestor of the sigmodontines reached South America by overwater dispersal, most likely in the Miocene. They argued that the degree of differentiation of the South American sigmodontines supports an early entry of an ancestral stock into South America, with differentiation at the tribal, generic, and species level taking place autochthonously in South America. They interpret the Argentine fossils as signaling the arrival of advanced forms that had initially differentiated elsewhere in South America. The lack of rodent fossils in the Tertiary of Central America, as well as the lack of South American sigmodontine fossils prior to the advanced forms at 4–5 MYA, frustrates attempts to resolve the history of the group based on fossil evidence.

Relationships Among South American Sigmodontine Rodents

91

MATERIALS AND METHODS Whole genomic extracts of DNA were made following the sodium dodecyl sulfate-proteinase K/phenol/RNAse method (Maniatis et al., 1982, pp. 458–462) or by salt extraction (Medrano et al., 1990). One South American sigmodontine, Chroeomys jelskii, has been shown to carry a nuclear copy of a portion of the mitochondrial genome that includes part of the cytochrome b (cyt b) gene (Smith et al., 1992). There was also evidence in the course of this study for nuclear copies of cyt b in the genus Delomys. Therefore, for most samples, in addition to whole genomic extracts, purified mitochondrial DNA was obtained from frozen liver tissue using cesium chloride-ethidium bromide gradient centrifugation (Lansman et al., 1981) or by a more rapid method based on an alkaline lysis procedure, using a Promega Wizard Minipreps kit (Beckman et al., 1993). For manual sequencing gels, the most efficient approach to sequencing the cyt b gene utilized a biotinylated version of primer MVZ 14 in combination with primer MVZ 05 to produce double–stranded product using purified mtDNA as template. The strands were separated using Dynal Dynabeads M-280 streptavidin and sequenced with a combination of external and internal primers, including light-strand primers MVZ 05, MVZ 11, MVZ 45, MVZ 23 (primer sequences given by Smith and Patton, 1993, Table 2), and MVZ 67 (Patton et al., 1996), and heavy-strand primers MVZ 06 and MVZ 04 (Smith and Patton, 1993, Table 2). General procedures for amplification and manual sequencing followed those of Smith and Patton (1991). Sequences were also obtained on an Applied Biosystems Model 377 automated DNA sequencer using the Taq FS cycle sequencing kit. Additional sequencing primers used on the automated sequencer include lightstrand primers MVZ 103 [5'-CACCCTAAC(A/C)CGCTTCTTCGC-3'] and MVZ 17 [5'-ACCTCCTAGGAGA(C/T)CCAGA[C/A/T]AA(C/T)T-3'] and heavy-strand primer MVZ 14 (Smith and Patton, 1993). Sequences were edited using the Sequence Navigator software (Applied Biosystems, Inc., 1994). Typical sequences were 1140 bp long, ending in a stop codon (TAA or TAG in all of the species of Akodon, Bolomys, and Thaptomys, for example). Other taxa, including Oxymycterus, Scapteromys, Calomys, Loxodontomys, Phyllotis, henomys, Abrothrix, Chelemys, Chroeomys, Geoxus, Oecomys, Oryzomys, Scolomys, and Sigmodon, had an extra codon at the end of the sequence, followed by a T, which presumably gets polyadenylated to form a stop codon, as in the cyt b gene of Mus (Bibb et al., 1981). Sequence data were available for the first 801 bp of cyt b from 12 genera (42 species) of South American sigmodontines [Smith and Patton (1991), GenBank accession numbers M35691-M35716; Smith and Patton (1993), U03524–U03550; Patton and da Silva (1995), U58379-U58392]. Additional sequence was obtained for representative individuals from these earlier samples to complete the cyt b gene (ca. 1140 bp) in most of the species, with all 12 genera represented. The previous samples were drawn from the akodontine, oryzomyine, phyllotine, and thomasomyine tribes. New taxa were added in the current study to increase the representation of these and other tribes. Forty-three new species and 26 new genera of South American sigmodontines were added. The arrangement of South American sigmodontine taxa into tribal groupings has been the subject of considerable debate, and there is no clear consensus on the groupings. The genera recognized, and the arrangements of tribal groupings presented in Fig. 1, contrast the views of Reig (1986) in Fig. la, and Musser and Carleton (1993) in Fig. 1b. Sequence data

92

Smith and Patton

Fig. 1. South American sigmodontine rodent genera recognized, and the arrangement of those genera into tribes, contrasting the views of (a) Reig (1986) and (b) Musser and Carleton (1993). At the tribal level, Reig (1986) grouped the thomasomyines and oryzomyines together into one tribe, the oryzomyines, whereas Musser and Carleton (1993) treated them as two tribes. The two views differ in several places in terms of which genera are recognized and in the assignment of some of the genera to tribes. Genera marked with an asterisk in b have a history of changing placement, summarized briefly by Musser and Carleton (1993).

were generated in the current study for a total of 38 South American genera (85 species), representing seven of the eight tribes listed in Fig. 1b. Sequences were also obtained for four genera (seven species) of North American murid rodents: Peromyscus, Reithrodontomys, Scotinomys, and Neotoma. These represent three of the four tribal level groupings within the neotomine-peromyscine clade identified by Carleton (1980). GenBank acces-

Relationships Among South American Sigmodontine Rodents

93

sion numbers for representatives of the South American and North American genera are AF108666-AF108709. Sample localities and museum catalog numbers of voucher specimens are listed in the Appendix. Additional outgroup sequences were available from GenBank, including Old World murine rodents Mus (Bibb et al., 1981) and Rattus (Gadaleta et al., 1989) and the geomyoid rodent genera Thomomys, Geomys, Dipodomys, and Perognathus (Smith, 1998) as the most distant outgroup. The primary focus of this study is on intergeneric comparisons; details of specific intrageneric analyses will be published elsewhere. Kimura (1980) two-parameter distances were calculated using the Molecular Evolutionary Genetics Analysis (MEGA) program of Kumar et al. (1993). Maximum-parsimony analyses were run in PAUP* 4.0 (test versions 4.0d61, 4.0d63, and 4.0d64; written by David L. Swofford) and in PAUP* 4.0 bl (Swofford, 1998). Analyses were done with all sites weighted equally (with 100 random orders of addition), with transversions weighted 2 to 10 times as much as transitions (with 10-100 random orders of addition), and with transversions only (with 2-10 random orders of addition). In addition, maximum-parsimony analyses (with 100 random orders of addition) were performed on the amino acid sequences for all of the taxa. Support for the nodes was assessed with bootstrap analyses with 1000 pseudoreplicates with the fast heuristic procedure. At each hierarchical level of analysis multiple analyses were run, progressively eliminating the more distant outgroups. Additional assumptions for each analysis are detailed under Results. Maximum-likelihood analysis of the most reduced data set at the genus level was run using test version 4.0d64 of PAUP*, written by David L. Swofford. A complex model (HKY + G) optimized three parameters based on the empirical data: the proportions of the four bases, the ratio of transitions to transversions, and the shape of the gamma distribution (alpha) of rate change at different sites. Comparisons of In-likelihood values for the best tree found in the analysis, evaluated under the assumptions of several different models with fewer estimated parameters, were used to test the model of DNA substitution, as described by Huelsenbeck and Crandall (1997). Maximum-likelihood analyses were also run with the parameters set to those of the best tree, but with the tree topology constrained to match various hypotheses of relationships among groups. The maximumlikelihood score for the best tree under the constraints was compared to the value for the best unconstrained tree using the Kishino-Hasegawa (1989) test. A UPGMA tree based on third-position transversions was produced in MEGA (Kumar et al., 1993) to provide a framework for estimating times of divergence using fossil dates to set the scale. This approach is based on the premise that third position transversions accumulate nearly linearly with time for tens of millions of years in mammalian mtDNA (Irwin et al., 1991). An alternative approach, using the program QDate vl.l described by Rambaut and Bromham (1998), uses a maximum-likelihood model and takes into account lineage–specific rate heterogeneity. The parameters used in the analysis constitute an HKY-G model, by making the following choices in the QDate program: HKY model (Hasegawa et al., 1985), five categories for discrete gamma rate heterogeneity, values of alpha (the shape of the gamma distribution) of 0.50 or 0.25, a transition/transversion ratio of 2 or 6, one-rate or two-rate models, and a chi-square test of the likelihood-ratio estimates.

94

Smith and Patton

RESULTS AND DISCUSSION Detailed Sampling of Individuals and Localities For many localities the complete cyt b sequence (1140 bp) was obtained for two or more individuals of the same species. Variation among individuals within localities is minimal [N = 92 pairwise comparisons; mean, 0.005 K2p; range, 0–0.016 Kimura two-parameter distance (K2p)]. At the next hierarchical level of sampling, two or more geographic localities were sampled for 11 species. Comparisons among localities averaged 0.037 K2p (range, 0.004–0.097 K2p). In most cases only one geographic locality is used to represent each species in the next level of analysis of phylogenetic relationships. However, multiple samples are included for the taxa with the greatest amount of geographic variation. Phylogenetic Analyses Using All Species—Support for Generic Groupings Figure 2 presents an overview of relationships among all of the species for which between 759 and 1140 bp of the cyt b sequence is available. This data set includes 93 sequences from 85 South American species as currently identified, with seven species of North American Neotominae as the outgroup. The samples with less than the complete cyt b sequence have fewer informative sites on which to base their placement. However, most of the shorter sequences do cluster with their expected group in the analyses of nucleotide sites. Two of the seven genera identified below as unique genetic lineages have shorter sequences (Delomys, 1048 bp; Wiedomys, 907 bp). In a maximum-parsimony analysis in PAUP* 4.0d64, a heuristic search with all sites weighted equally, with 100 random orders of addition, uncovered one island with six trees with scores of 8175. The strict consensus of these trees is shown in Fig. 2. The distribution of random trees suggests that there is significant nonrandom information in the data set [g] = -0.211920, P < 0.01 (Hillis and Huelsenbeck, 1992)]. There were 615 variable sites in the data set, of which 556 were parsimony informative. The sequence for two or more species was obtained for 12 South American genera, and there is generally good support for generic level groupings, as detailed below. Akodontini Three genera of akodontines (Akodon, Bolomys, and Oxymycterus) have broad geographic distributions (Smith and Patton, 1993, Fig. 1) and contain numerous species. The 18 species of Akodon (s.s.) sequenced include Akodon boliviensis, the type species for the genus Akodon. The species of Akodon differ by a maximum of 0.173 K2p and group together with bootstrap support of 71% (Fig. 2). Akodon is one of the most speciose genera of sigmodontines, and our analyses include about half of the currently recognized species. A more detailed analysis of relationships and geographic groupings within the genus Akodon will be presented elsewhere. Several taxa formerly included in the subgenus Akodon, or in other subgenera included in the genus Akodon, can now be separated out based on the cyt b data. The first of these taxa is Akodon (Thaptomys) nigrita. The sequence for this species falls outside the cluster of Akodon species detailed above, thus providing phylogenetic support for the generic status of Thaptomys (see also Hershkovitz, 1998). Akodon olivaceus and Akodon

Relationships Among South American Sigmodontine Rodents

Fig. 2. Species-level analysis of phylogenetic relationships among the South American Sigmodontinae, using seven species of North American Neotominae as the outgroup. In a maximum-parsimony analysis of cyt b sequences in PAUP*4.0 with all sites weighted equally, there were 556 parsimony informative sites. The strict consensus of two trees of length 8175 is shown. Bootstrap values were obtained in a separate bootstrap analysis with 1000 replicates; only bootstrap values >50% are shown here. Locality numbers refer to the listing in the Appendix.

95

Smith and Patton

96

xanthorhinus, plus the taxa previously included in the subgenera Abrothrix and Chroeomys, are quite distinct from the genus Akodon. They belong to a clade of mice in the central and southern Andes identified previously by Smith and Patton (1993). This complex is discussed in the section on suprageneric relationships below. The pattern of geographic distribution of Bolomys is unusual. The disjunct Venezuelan form currently known as Akodon urichi is actually more closely related to Bolomys than to Akodon (Smith and Patton, 1993). Voss (1991a) assigned the form punctulatus, in eastern Ecuador, to Bolomys, based on morphological data. Assuming that urichi is indeed a Bolomys, then the geographic pattern for the genus presumably represents the remains of a formerly more widespread distribution. Divergence among the species of Bolomys is, in fact, quite high, with a maximum of 0.145 K2p. Although the genus Bolomys did not form a monophyletic group in previous analyses based on 801 bp of the cyt b sequence (Smith and Patton, 1993), monophyly is supported in the current analyses based on 1140 bp of the cyt b sequence. Bootstrap support for the grouping of Bolomys amoenus, the type species for the genus Bolomys, with B. lasiurus and Bolomys (Akodon auct.) urichi is 85% (Fig. 2). If the argument of Massoia and Pardinas (1993) that Necromys conifer and Akodon benefactus are synonymous is accepted, then the name Necromys has precedence over Bolomys for this genus. The species identifications for some of the samples of the genus Oxymycterus still remain to be established. Two species are present in the samples from Uruguay, one of which is Oxymycterus nasutus, the type species of the genus. The samples of the genus Oxymycterus differ by a maximum of 0.101 K2p and form a well-supported monophyletic unit in the analysis of cyt b sequence data, with bootstrap support of 100% (Fig. 2). A more detailed analysis of relationships within the genus Oxymycterus will be presented elsewhere. Oxymycterus iheringi, a taxon previously included in the genus Oxymycterus, is now included in a new genus, Brucepattersonius, described by Hershkovitz (1998). Sequence was obtained from samples from the coast ranges of southeastern Brazil, in the states of Sao Paulo, Rio de Janeiro, and Minas Gerais. Available data suggest the presence of two species, one that includes samples from Intervales, Boraceia, and Itatiaia (with a maximum divergence of only 0.016 K2p) and a second collected also at Itatiaia and at Passa Quatro (divergence among these localities is 0.007 K2p). The two taxa differ from each other by 0.082-0.090 K2p and group together with bootstrap support of 61% (Fig. 2). Thomasomyini Seven species were sequenced in the genus Thomasomys, about one-fourth of those listed by Musser and Carleton (1993). This set of species represents the full range of morphological variation in the genus, and each is quite distinct genetically (minimum pairwise K2p of 0.136 in all comparisons). It is likely, therefore, that most of the major genetic lineages within Thomasomys are represented in our data. They differ by a maximum of 0.185 K2p and group together consistently with bootstrap support of 69% (Fig. 2). In a recent revision of Rhipidomys based on morphological characters, Tribe (1996) divided the genus into three sections. We sequenced six species, including representatives of each of Tribe's morphological sections, and the data accord well with his views.

Relationships Among South American Sigmodontine Rodents

97

Rhipidomys mastacalis (north coastal Brazil), R. nitela (eastern Amazonia), and two undetermined species (R. sp. 1 from southeastern Peru and R. sp. 2 from central coastal Brazil), all members of the leucodactylus section, group together with pairwise K2p distances of 0.073 to 0.146. The closest pair of this section is R. mastacalis and R. nitela (0.073 K2p). Representatives of the other two sections, R. wetzeli and R. macconnelli, are quite distinct from each other (0.172 K2p) and from those of the leucodactylus section (range, 0.158-0.187 K2p). Cyt b sequences promise to be quite useful in delineating species boundaries and relationships within this genus. The lowland Rhipidomys group together with bootstrap support of 79%, but the entire genus groups with only 51% (Fig. 2). Voss (1993) presented a detailed study of morphological variation in the genus Delomys. He recognized two species, D. sublineatus and D. dorsalis; both were sequenced. They differ by the modest amount of 0.091 K2p and group together with bootstrap support of 100% (Fig. 2). Bonvicino and Geise (1995) recommended the recognition of a third species, Delomys collinus, found at elevations at or above 1750 m in the coastal ranges of Brazil, based on a unique karyotype. We have not examined this taxon. Two species of Wilfredomys, one Wilfredomys pictipes and the second currently undescribed, differ from each other by 0.144 K2p. Both of these species are small bodied, a character that distinguishes them from W. oenax. They differ from each other in a number of trenchant characters, including carotid circulation pattern and bullar characters. Despite the large K2p distance between the two included species, they form a strongly identified monophyletic clade supported by a bootstrap value of 100% (Fig. 2). Phyllotini Among the phyllotines, two of the four species of Eligmodontia listed by Musser and Carleton (1993), Eligmodontia morgani and E. typus, differ by 0.106 K2p and group together with bootstrap support of 100% (Fig. 2). Musser and Carleton (1993) list nine species of Calomys. Calomys lepidus and Calomys sorellus differ by 0.112 K2p and group together with bootstrap support of 89% (Fig. 2). Sequence from additional species would be useful in sorting out the taxonomy of Calomys. Oryzomyini Among the oryzomyines, multiple species of Oecomys [with 3 of the 13 species listed by Musser and Carleton (1993) plus an additional undetermined 1] and Oryzomys (two presumed species) were sequenced. The four Oecomys species differ by 0.091–0.103 K2p and group together with bootstrap support of 72% (Fig. 2). The two forms currently designated Oryzomys megacephalus differ by 0.149 K2p and group together with bootstrap support of 55% (Fig. 2). The Oryzomys, from Guyana and Peru, represent the eastern Amazonian and western Amazonian clades identified by Musser et al. (1998) based on morphological characters and on 401 bp of the cyt b sequence provided by J. L. Patton. Phylogenetic Analyses Using Exemplar Species—Support for Suprageneric Groupings To set the stage for the discussion of tribal groupings and their relationships, it is helpful first to consider the outgroups to the South American sigmodontines. Pocket

98

Smith and Patton

Fig. 3. Relationships of outgroup taxa to 38 genera of South American sigmodontines, based on a maximum-parsimony analysis of cyt b with transversions weighted six times as much as transitions. Bootstrap values were generated in a separate bootstrap analysis with 1000 replicates; only bootstrap values >50% are shown here.

gophers, Thomomys and Geomys, from the family Geomyidae, and pocket mice (Perognathus) and kangaroo rats (Dipodomys), from the family Heteromyidae, are used as the most distant outgroup for the analysis in Fig. 3. The figure shows a maximum-parsimony analysis with transversions weighted six times as much as transitions, to emphasize more slowly evolving sites. This is the same weighting scheme as used in the analysis for Fig. 4, discussed below. Two genera of Old World murid rodents, Mus and Rattus, currently assigned to the subfamily Murinae (Musser and Carleton, 1993), group together first. The four genera of North American murids, Peromyscus, Reithrodontomys, Scotinomys, and Neotoma, cluster together. The sampling of murid subfamilies is limited; in particular, the Central American subfamily Tylomyinae (Reig, 1984) is not represented here. Support for the nodes at this level is not always strong, but the pattern does suggest that of the available samples in this analysis, the North American Neotominae is the closest outgroup to the South American Sigmodontinae. It is important to note here that the data do provide bootstrap support for the monophyly of the South American sigmodontines (71%; Fig. 3). For other analyses of this data set, including analyses with all bases weighted equally, with transversions weighted 2, 4, 8, or 10 times as much as transitions, or with transversions only, the bootstrap support for the South American sigmodontines ranges from 59 to 81%. When the data are analyzed as amino acid sequences, bootstrap support for the monophyly of the South American sigmodontines increases to 88%. The next level of analysis focuses on suprageneric- and tribal-level groupings. These relationships can be examined in Fig. 2, where all species are represented. A more condensed data set, with a single species to represent each genus, allows a more streamlined picture of suprageneric groupings and, also, makes it feasible to run a maximum-likelihood analysis. Parsimony analyses on intermediate numbers of taxa, with a few welldifferentiated species selected to represent speciose genera, give similar pictures of relationships. The trees in Fig. 4 are based on a maximum-parsimony analysis with transversions weighted six times as much as transitions. This weighting value is close to the

Relationships Among South American Sigmodontine Rodents

Fig. 4. Generic-level analysis of phylogenetic relationships among the South American Sigmodontinae using exemplar species and the closest available outgroup, the North American Neotominae (43 taxa total). In a maximum-parsimony analysis of cyt b with transversions weighted six times as much as transitions, there were 516 parsimony informative sites, and two islands of trees were found of length 12635. Tree a is the strict consensus of the two trees in the first island; tree b is the strict consensus of the two trees in the second island. Bootstrap values were generated in a separate bootstrap analysis with 1000 replicates; only bootstrap values >50% are shown here. Locality numbers refer to the listing in the Appendix.

99

Smith and Patton

100

Fig. 4. Continued.

transition/transversion rate ratio, K, of 5.8 found in the best maximum-likelihood tree below. Analyses with transversions weighted 2, 4, 8, or 10 times as much as transitions gave varying pictures of the relationships among groups, but the genera that grouped in Fig. 4 with bootstrap support greater than 50% were similarly grouped in the other weighted analyses, regardless of the particular weighting value. For the generic level analysis there are 578 variable sites in the data set, of which 516 are parsimony informative. Two islands of two trees each were found in 100 random orders of addition; the strict consensus for each island is shown in Fig. 4. The distribution of random trees suggests

Relationships Among South American Sigmodontine Rodents

101

Fig. 5. Best tree found under the HKY + G model in a maximum-likelihood analysis of cyt b sequence for 43 taxa at the generic level, using the same exemplar species as in Fig. 4. The In-likelihood value for this tree is –18284.32063.

that there is significant nonrandom information in the data set [g1 = -0.509326, P < 0.01 (Hillis and Huelsenbeck, 1992)]. The maximum-likelihood analysis (Fig. 5) is based on a complex model that includes base composition, variable rates of change at different sites, and the ratio of transitions to transversions. The In-likelihood value under the HKY + T model for the tree in Fig. 5 is -18284.32063, with empirical base frequencies (A = 0.29729, C = 0.28416, G = 0.12407, T = 0.29448), a transition-to-transversion ratio of 2.866 (K = 5.797), and the value of the gamma shape parameter, a = 0.276. Likelihood values were calculated on the tree in Fig. 5 for a series of models of increasing complexity. Table I reports the Inlikelihood values for four models, plus those same models with rate heterogeneity added.

102

Smith and Patton

Table I. The In-Likelihood Values Under the Assumptions of Four Substitution Models, Plus Those Four Models with Rate Heterogeneity (Indicated by +T)

Modela

Base frequencies

HKY85

Equal Empirical Equal Empirical

JC69 + T F81 +T K80 + T HKY85 + T

Equal Empirical Equal Empirical

JC69

F81 K80

Substitution model (K)

Site variation

(a)

1 1

Estimated from data Estimated from data

-23281.07814 –22773.17957 -22335.60771 -21790.13683

IC I

Estimated from data Estimated from data

1 1

ln-likelihoodb

I I

Estimated Estimated Estimated Estimated

from from from from

data data data data

-20042.62627 -19392.58009 –18990.50366 -18284.32063

aModel

abbreviations: JC69, Jukes and Cantor (1969); F81, Felsenstein (1981); K80, Kimura (1980) twoparameter model; HKY85, Hasegawa, Kishino, and Yano (1985). See Swofford et al. (1996) for a detailed discussion of likelihood models. bCalculated on the tree in Fig. 5 using test version 4.0d64 of PAUP*, written by David L. Swofford. cAll sites assumed to evolve at the same rate.

Likelihood-ratio tests of models with empirical base frequencies versus equal frequencies, transition/transversion ratios estimated from the data versus equal rates, and among-site variation estimated from the data versus equal rates at all sites all showed significantly better In-likelihood values for the more complex model, which matches the tree shown in Fig. 5. Some groupings of taxa are recognized fairly consistently in all, or most, analyses. These include analyses of progressively reduced data sets with exemplar species, the earlier analyses with all species included, analyses with all sites equally weighted, with transversions weighted 2 to 10 times as much as transitions, and with transversions only. While some tribal groupings appear frequently, there is not strong bootstrap support at the basal nodes of the tribal-level groupings, except in a few instances. The deeper-level relationships among the tribal clusters are not at all well resolved and differ considerably from one analysis to the next. Many genera consistently group together in tribal-level clusters, but others stand alone, or shift frequently in their position on the trees, depending on the particular analysis. Seven genera can be identified as having the most uncertainty in their placement: Delomys, Wilfredomys, Scolomys, Reithrodon, Irenomys, Wiedomys, and Sigmodon. Several of these do not fit into the higher-level group to which they had been assigned previously (see, for example, the arrangements in Figs. la and b). Rather, these genera appear to be unique genetic lineages which differ from all the other genera in the data set by mean values of 0.197–0.262 K2p. This may explain why, historically, they have been moved around from one tribe to another in classifications. In fact, they may not have close affinities to any other currently existing taxa. Although at times these genera cluster with other unique genera, in no case is bootstrap support greater than 50%, and the pairings may be due to long-branch attraction (Felsenstein, 1978; Hendy and Penny, 1989). A proposed arrangement of genera into tribal level groupings, integrating the cyt b data of the current study with previous knowledge based primarily on morphology and karyotypes, is presented in Fig. 6. No sequence data are available from any of the gen-

Relationships Among South American Sigmodontine Rodents

103

Fig. 6. South American sigmodontine rodent genera recognized, and the arrangement of those genera into tribes, incorporating the insights gained from analyses of cyt b sequence data in this report. Sequence data were obtained for the genera in boldface type. No sequence data were available from any of the genera in the ichthyomyine tribe. All of the other tribes listed in Fig. 1b are represented by some sequence data.

era in the ichthyomyine tribe. This tribe is the only one which has been recently revised (Voss, 1988), and it stands as a monophyletic group on morphological grounds. All of the other tribes listed in Fig. 1b are represented by some sequence data. Six relatively large clades are shown in the top row in Fig. 6. Two additional clades are listed in the lower row, as well as a series of genera shown, from cyt b data, to be unique lineages. The genera listed under additional unique lines could each be recognized at the tribal level, as a way of emphasizing their genetic distinctness. Formal recognition at the tribal level of the six genera labeled "Andean clade" in Fig. 6 would emphasize their genetic distinctness from the Akodontini, where the six Andean genera have usually been placed (Fig. 1). New genera described since Musser and Carleton (1993), and listed in Fig. 6, include Pearsonomys (Patterson, 1992), Salinomys (Braun and Mares, 1995), Lundomys (Voss and Carleton, 1993), Microakodontomys (Hershkovitz, 1993), and Brucepattersonius (Hershkovitz, 1998). Braun (1993) and Steppan (1993) recognized Loxodontomys as a separate genus. The relationships of the genera are discussed in detail in the sections that follow. Akodontini and Scapteromyini The first tribal-level group listed in Fig. 6 is the Akodontini. Three genera, Akodon, Thaptomys, and Bolomys, group together with strong bootstrap support no matter what approach is taken in the analysis (Fig. 2, species level, ts + tv, 88%; Fig. 4, genus

Smith and Patton

104

level, ts + 6x tv, 97%). The genus Oxymycterus groups with the first three genera in most analyses, but with bootstrap support of less than 50%. Blarinomys and Brucepattersonius always group together, although with relatively low bootstrap support (Fig. 2, species level, ts + tv, 68%; Fig. 4, genus level, ts + 6x tv, 55%). Lenoxus typically joins Blarinomys and Brucepattersonius, but with low bootstrap support (