© 2005 Society of Systematic Biologists
Resolving Deep Phylogenetic Relationships in Salamanders: Analyses of Mitochondrial and Nuclear Genomic Data
Edited by Elizabeth Jockusch
1 Department of Biology, Campus Box 1137, Washington University St. Louis, Missouri, 63130, USA
2 Present Address: Department of Biology, University of Kentucky Lexington, Kentucky, 40506, USA; E-mail: weisrock{at}uky.edu
 |
Abstract |
|---|
 Top Abstract Materials and MethodsResultsDiscussionReferences |
|---|
Phylogenetic relationships among salamander families illustrate analytical challenges inherent to inferring phylogenies in which terminal branches are temporally very long relative to internal branches. We present new mitochondrial DNA sequences, approximately 2100 base pairs from the genes encoding ND1, ND2, COI, and the intervening tRNA genes for 34 species representing all 10 salamander families, to examine these relationships. Parsimony analysis of these mtDNA sequences supports monophyly of all families except Proteidae, but yields a tree largely unresolved with respect to interfamilial relationships and the phylogenetic positions of the proteid genera Necturus and Proteus. In contrast, Bayesian and maximum-likelihood analyses of the mtDNA data produce a topology concordant with phylogenetic results from nuclear-encoded rRNA sequences, and they statistically reject monophyly of the internally fertilizing salamanders, suborder Salamandroidea. Phylogenetic simulations based on our mitochondrial DNA sequences reveal that Bayesian analyses outperform parsimony in reconstructing short branches located deep in the phylogenetic history of a taxon. However, phylogenetic conflicts between our results and a recent analysis of nuclear RAG-1 gene sequences suggest that statistical rejection of a monophyletic Salamandroidea by Bayesian analyses of our mitochondrial genomic data is probably erroneous. Bayesian and likelihood-based analyses may overestimate phylogenetic precision when estimating short branches located deep in a phylogeny from data showing substitutional saturation; an analysis of nucleotide substitutions indicates that these methods may be overly sensitive to a relatively small number of sites that show substitutions judged uncommon by the favored evolutionary model.
Keywords: Bayesian; DNA simulation; internal fertilization; mitochondrial DNA; parsimony; ribosomal RNA; salamander
Received June 23, 2004; Revised November 16, 2004; Accepted March 25, 2005
Phylogenetic relationships among taxonomic families of salamanders are difficult to resolve because branches grouping multiple families are likely very short relative to the subsequent evolutionary history of the group (Larson et al., 2003). Parsimony-based analyses of morphological characters and nuclear-encoded rRNA sequences produce different phylogenetic topologies with statistically significant conflict between these data sets (Larson and Dimmick, 1993; Larson, 1998). Analyses of an expanded set of morphological characters and sequences from the nuclear RAG-1 gene likewise show that resolution of deep phylogenetic relationships varies among data partitions and phylogenetic methods (Wiens et al., 2005).
Analyses of combined morphological and molecular data support monophyly of internally fertilizing salamanders, suborder Salamandroidea (families Ambystomatidae, Amphiumidae, Dicamptodontidae, Plethodontidae, Proteidae, Rhyacotritonidae, and Salamandridae; Larson and Dimmick, 1993; Wiens et al., 2005), although this grouping is rejected statistically by two independent molecular data sets, the nuclear-encoded rRNA sequences and the mitochondrial genomic sequences reported here.
Nonmonophyly of Salamandroidea requires homoplastic evolution of internal fertilization, a complex syndrome of characters involving morphology of male and female cloacal glands, life history and behavior (see Sever and Brizzi, 1998) or secondary loss of this trait. Independent evolution of internal fertilization by multiple lineages seems unlikely (Sever and Brizzi, 1998). Absence of evolutionary loss of this syndrome within families indicates that it is indispensable to the species that have it and therefore highly "burdened" using the terminology of Riedl (1978) and Donoghue (1989); such characters are unlikely to be secondarily lost. External fertilization, as observed in suborder Cryptobranchoidea (families Cryptobranchidae and Hynobiidae) and strongly inferred for suborder Sirenoidea (family Sirenidae; Sever et al., 1996), is considered ancestral for salamanders and anuran amphibians.
To reexamine these relationships and the phylogenetic-information content of molecular characters, we present new mitochondrial DNA (mtDNA) sequences, approximately 2100 base pairs from the genes encoding ND1, ND2, COI, and the intervening tRNA genes. Our phylogenetic analyses apply parsimony, maximum-likelihood, and Bayesian criteria (Huelsenbeck et al., 2001; Swofford et al., 1996) to test alternative phylogenetic hypotheses using both new mtDNA data and previously published morphological and nuclear rRNA data (Larson, 1991; Larson and Dimmick, 1993). We use a variety of statistical tests to evaluate alternative phylogenetic hypotheses derived from both a priori information (e.g., monophyly of suborder Salamandroidea) as well as a posteriori hypotheses generated from our new phylogenetic results. We identify and describe the molecular characters that make large contributions to statistically significant phylogenetic results obtained from maximum-likelihood and Bayesian methods.
Because the evolutionary age of salamanders is at least 150 My (Gao and Shubin, 2001) and fossil evidence dates some extant families to the upper Cretaceous (Gao and Shubin, 2003), branches grouping two or more families are probably short relative to their phylogenetic depth. Previous attempts at using mtDNA sequence (12S and 16S rRNA) data to resolve salamander-family phylogeny produced a tree with extremely long terminal branches and short, poorly supported interfamilial branches (Hay et al., 1995), indicating that family-level cladogenesis occurred over a relatively short time interval. In such cases, misleading phylogenetic information generated by parallel substitutions among branches within taxonomic families may overwhelm the signal generated by substitutions on a lineage ancestral to two or more families. Phylogenetic signal is eroded further by multiple substitutions occurring at the sites containing the original phylogenetic signal. Phylogenetic reconstruction of such evolutionary histories can be misled by long-branch attraction (Felsenstein, 1978; Huelsenbeck, 1997; Swofford et al., 2001). In extreme cases, a bifurcating tree may be analytically indistinguishable from a polytomy (see discussion by Donoghue and Sanderson [1992], Fishbein et al. [2001], Jackman et al. [1999], Misof et al. [2001], Poe and Chubb [2004], Slowinski [2001], and Walsh et al. [1999]). We explore the limits of deep phylogenetic reconstruction in the evolutionary history of salamanders using simulations based on the inferred evolutionary models of our mtDNA sequences.
|
Materials and Methods |
|---|
|
TopAbstractMaterials and MethodsResultsDiscussionReferences |
|---|
Taxon Sampling
This study used the published morphological and nuclear rRNA sequence data matrices of Larson and Dimmick (1993). The morphological data used by Larson and Dimmick (1993) are compiled from other sources representing head and trunk morphology (Duellman and Trueb, 1986) and cloacal characters related to fertilization (Sever 1991a, 1991b; 1994). The complete data matrix containing all taxa and morphological characters from Larson and Dimmick (1993) was used in combined analyses with molecular characters.
Nuclear rRNA sequence data were taken from Larson and Dimmick (1993) and Larson (1991) and included the anuran outgroup Xenopus laevis, the caecilian outgroup, Typhlonectes compressicauda, and 24 ingroup taxa representing all ten salamander families (Table 1). Ten ingroup samples from Larson (1991) were excluded from our analyses because they are identical or nearly identical to analyzed samples representing the same taxonomic family. This reduction in sampling made heuristic parsimony and likelihood searches and bootstrap resampling computationally manageable.
|
Taxon sampling for the mtDNA data closely matched that of the nuclear rRNA sequence data (Table 1). When possible, tissue samples from the same specimen were used for the nuclear rRNA and mtDNA data; in all other cases but one (Rhyacotritonidae), another specimen of the same species was used. The mtDNA data set contained samples of 38 taxa, including 4 outgroup and 34 ingroup taxa covering all 10 salamander families. This included mitochondrial sequence for Proteus anguinus, a European troglodytic salamander grouped with the North American genus Necturus in the family Proteidae. Placement of these genera within the same family is the most dubious of all salamander-family groupings. Sequences for two outgroup taxa, Rana catesbeiana and Typhlonectes natans, and three ingroup taxa, Chioglossa lusitanica, Notophthalmus viridescens, and Salamandra salamandra, were taken from GenBank (Table 1).
Taxon sampling in combined phylogenetic analyses followed two approaches. Because not all data were available for all taxa across data sets, we assembled pruned data sets that minimized the amount of missing data. Wiens (2003) showed that accuracy of phylogenetic analyses of incomplete data may be high if each taxon sampled contains sufficient characters; therefore, we also assembled combined data sets using the full complement of taxa from the mtDNA data set with rRNA and morphological data missing for some taxa.
DNA Extraction and Sequencing
DNA extraction, polymerase chain reaction, and sequencing methods were as described by Weisrock et al. (2001) except that some sequencing reactions were performed using a Big-Dye Terminator Ready-Reaction Kit (Perkin-Elmer) and run on an ABI (PE Applied Biosystems, Inc.) 373A automated DNA sequencer. We sequenced approximately 2100 contiguous bases of mitochondrial DNA including the genes encoding ND1 (subunit one of NADH dehydrogenase), tRNAIle, tRNAGln, tRNAMet, ND2 (subunit two of NADH dehydrogenase), tRNATrp, tRNAAla, tRNAAsn, tRNACys, tRNATyr, and COI (subunit I of cytochrome c oxidase), plus the origin for light-strand replication (OL) between the tRNAAsn and tRNACys genes. This block of sequence also contains a small number of bases of intergenic sequence separating some tRNA genes. Primers used for sequencing are listed in Table 2.
|
Sequence Alignments
Alignment of the nuclear rRNA sequence data was similar to the original alignments of Larson (1991) and of Larson and Dimmick (1993) except for some length-variable segments in divergent domains of rRNA where alignment was considered ambiguous. These regions were excluded from all analyses.
Alignment of the mtDNA sequences was performed manually using amino acid sequence translations for protein-coding genes and secondary-structural models for tRNA genes (Kumazawa and Nishida, 1993). Length-variable regions whose alignment was ambiguous, including tRNA loops, all intergenic sequence, and much of the origin for light-strand replication (OL), were excluded from all phylogenetic analyses.
All new mtDNA sequences have been deposited in GenBank under accession numbers AY916014 [GenBank] to AY916045. Data alignments for the mtDNA and nuclear rRNA character matrices have been placed in TreeBase under accession number S1237.
Phylogenetic Analyses
Parsimony analysis was performed on the mtDNA and nuclear rRNA data using PAUP* v4.0 (Swofford, 2002). A heuristic search option with 100 random-addition replicates was used with equal weighting of all characters and TBR branch swapping. To assess support for branches in the parsimony trees, bootstrap analysis was applied using 1000 bootstrap replicates with 100 random additions per replicate, and decay indices (Bremer, 1994) were calculated for each internal branch using TreeRot, v2.0 (Sorenson, 1999) and PAUP* v4.0 (Swofford, 2002). To test alternative phylogenetic hypotheses using parsimony, we used the Wilcoxon signed-ranks (WS-R) test (Templeton, 1983) as implemented in PAUP* v4.0. A number of a priori phylogenetic hypotheses were tested based on current classification and results from previous phylogenetic studies. To test these hypotheses we used the most parsimonious topology compatible with the alternative hypothesis for the data being examined. To generate these alternative topologies, a constraint tree was generated using MacClade v4.0 (Maddison and Maddison, 2000) and implemented in a parsimony search using PAUP* v4.0. All alternative topologies tested in this study are presented in Appendix 1 (available at www.systematicbiology.org).
Maximum-likelihood (ML) phylogenetic analysis was performed on the individual mtDNA and nuclear rRNA data sets using PAUP* v4.0 with a heuristic search option and stepwise addition of taxa, 10 random-addition replicates, and TBR branch swapping. Likelihood-ratio tests implemented in the program Modeltest v3.06 (Posada and Crandall, 1998) were used to find the model optimal for individual data sets. The resulting models and estimated parameters were used in the maximum-likelihood analyses. Alternative maximum-likelihood phylogenetic topologies were tested using the Shimodaira and Hasegawa (SH) test with 1000 RELL bootstrap replicates (Goldman et al., 2000; Shimodaira and Hasegawa, 1999) as implemented in PAUP* v4.0. This included the same a priori hypotheses tested under parsimony, as well as two a posteriori hypotheses based on the topologies of the ML and Bayesian trees. Alternative topologies were generated using maximum likelihood as described above for the Wilcoxon signed-ranks test. To determine whether individual branches in the resulting ML tree were significantly different from zero, we used a likelihood-ratio test as implemented in PAUP* v4.0 following Slowinski (2001) with a mixed chi-square distribution (Goldman and Whelan, 2000).
Bayesian phylogenetic analysis (Huelsenbeck et al., 2001; Larget and Simon, 1999; Yang and Rannala, 1997) was performed on the individual mtDNA and nuclear rRNA data sets and on combined molecular and morphological data sets using the parallel-processor version of MrBayes v3.04 (Altekar et al., 2004; Huelsenbeck and Ronquist, 2001) run on a Silicon Graphics Origin 2000 through the Washington University Center for Scientific Parallel Computing (http://harpo.wustl.edu). In all analyses four Markov chains were used with the temperature profile at the default setting of 0.2.
In Bayesian analyses of combined data sets we used a partitioned model (Nylander et al., 2004). Combined molecular data analyses treated the mtDNA and nuclear rRNA as separate partitions, using appropriate models as determined through likelihood-ratio tests. Combined molecular and morphological analyses used three partitions and implemented the Lewis Mk model for the morphological data partition (Lewis, 2001).
In all analyses uniform priors were used for all model parameter estimates, and random trees were used to start each Markov chain. For the mtDNA data set, five million generations were run with a tree sampled every 5000th generation for a total of 1000 trees. For the rRNA data set, 10 million generations were run with a tree sampled every 5000th generation for a total of 2000 trees. For all combined analyses two million generations were run with a tree sampled every 1000th generation for a total of 2000 trees. All saved trees from generations prior to a stationary likelihood value were discarded. Trees in the posterior distribution were parsed with MrBayes to construct phylograms based upon mean branch lengths and with PAUP* v4.0 to calculate the posterior probabilities of all branches using a majority-rule consensus tree. To account for the possibility that individual analyses may not converge upon the stationary posterior distribution, two additional independent runs were performed for each data set using identical conditions. Likelihood values, tree topology, branch lengths, and posterior probabilities were compared across the replicated runs to verify that similar results were achieved.
For both molecular data sets, maximum-likelihood and Bayesian analyses reconstructed the same tree topology. Consequently, Bayesian posterior probabilities were indicated on the maximum-likelihood tree. Bayesian posterior probabilities equal to or greater than 0.95 indicate statistically significant branch support. Therefore, we treat posterior probabilities as statistical measures and apply this test to the same alternative hypotheses tested with the WS-R and SH tests. Use of Bayesian posterior probabilities as measures of phylogenetic support is a contentious issue because a number of studies suggest that it gives overconfidence under some conditions (e.g., Simmons et al., 2004; Suzuki et al., 2002; Waddell et al., 2001). In contrast, simulation studies suggest that Bayesian analysis provides reliable results (e.g., Alfaro et al., 2003; Wilcox et al., 2002).
Mitochondrial-DNA Simulations
DNA-sequence simulations were performed to examine the combined influences of the length of an internal branch and its phylogenetic depth on its recovery with significant decay indices in a parsimony analysis and with high posterior probabilities in a Bayesian analysis. We chose three subsets of salamander taxa from the mtDNA phylogeny for the mtDNA simulations. Each subset contained three confamilial ingroup species whose relationships were robustly supported in the mtDNA analysis and a single outgroup from another salamander family (Fig. 1). Branch lengths were estimated for each four-taxon subset individually using maximum likelihood. Evolutionary models that best fit the individual data sets were assessed using likelihood-ratio tests as described above for the larger data sets.
|
We generated topologies for simulation of DNA evolution (Fig. 2) using two forms of branch-length manipulation. First, we started with the initial ML phylogram for each subset of taxa and successively shortened the internal branch (y) by a length of 0.0032 in units of likelihood-corrected sequence divergence. This procedure produced a series of trees with shortened internal branches (Fig. 2, moving from left to right). Each time branch y was shortened, an equivalent length was added to the terminal branches z1 and z2 to preserve overall tree height. Reduction of the internal branch was repeated until the tree formed a polytomy. Because initial branch lengths differed among the three model trees (Fig. 1), some trees needed more rounds of internal-branch shortening than others to reach a polytomy. Second, for each tree generated in this manner, a new series of trees was created to adjust the depth of the internal branch by successively increasing the lengths of all terminal branches (w, x, z1, and z2) by 0.0032 (moving from top to bottom in Fig. 2). For each set of taxa, this lengthening of terminal branches was repeated for the same number of steps needed to reduce branch y to a polytomy. The end product of these manipulations was a large set of trees representing the starting phylogram and all pairwise combinations of a reduction in length of y and an increase in phylogenetic depth of the internal branch (Fig. 2). For each tree, 100 replicate data sets of 1829 base pairs (the total number of included mtDNA base pairs in the empirical analyses) each were simulated in Seq-Gen v1.2.3 (Rambaut and Grassly, 1997) using the model parameters that best fit the particular four-taxon data set.
|
All data sets were analyzed by parsimony in PAUP* v4.0 using the exhaustive-search option. For each replicate data set the decay index for the branch uniting the two most closely related ingroup taxa was calculated. For simulated data sets in which the true phylogeny was not the shortest recovered tree, the decay index was scored as zero. The significance of the decay index for each replicate data set was determined via the four-taxon test of Felsenstein (1985) based on the number of phylogenetically informative characters for each simulation. Significance values for decay indices depend on expectations of a molecular clock. Using likelihood-ratio tests, we did not reject a molecular clock for our three starting four-taxon ML trees; therefore, we used significant decay-index values that assumed a molecular clock (Felsenstein, 1985; Weisrock et al., 2001). The percentage of significant decay indices from 100 replicate simulated data sets was calculated for each tree. These data were then plotted as a function of the length of the internal branch and the depth of the internal branch in the phylogeny.
Lack of automation precluded Bayesian analysis of all replicate data sets for all trees. Therefore, for each of the three four-taxon subsets we analyzed 10 simulated data sets from each of 25 trees spaced evenly among the full set of manipulated trees. Bayesian analysis was performed as described above using one million generations with a burnin of 100,000 generations. The average posterior probability for the internal branch of each tree was plotted in the same fashion as the decay indices.
The ND1-COI gene region used in this study has been shown to evolve at a consistent rate in numerous vertebrate groups (Bermingham et al., 1997; Hrbek and Larson, 1999; Macey et al., 1998a, 1999), including salamanders (Weisrock et al., 2001), which exhibit a per-lineage rate of approximately 0.64% change per million years (My). We therefore used this calibration for interpreting the simulations in evolutionary time.
Analysis of Likelihood-Based Character Change
We examined the character changes that support branches in the mtDNA likelihood-based trees. Two deep interfamilial branches that were not recovered using parsimony but had significant posterior probabilities (branches A and B) were compared to a deep interfamilial branch that was not recovered using parsimony (branch C) and had a low posterior probability (< 0.5). For each branch, the maximum-likelihood tree was compared to alternative trees found through a maximum-likelihood search constrained not to contain the branch under consideration. The log-likelihood (lnL) value was calculated for each character position in the mtDNA data for each tree being compared. For the 25 positions that showed the greatest difference in lnL favoring the original maximum-likelihood tree, we identified the types of substitutional change that differed between the two trees according to maximum-likelihood reconstructions of character change obtained from the "Describe Trees" option in PAUP*. For each of the 25 positions we also calculated the gamma-shape–approximation rate category that received the highest posterior probability. The gamma-shape approximation was divided into quartiles with quartile one representing characters with the lowest probability of change and quartile four representing characters with the highest probability of change.
|
Results |
|---|
|
TopAbstractMaterials and MethodsResultsDiscussionReferences |
|---|
Nuclear rRNA Phylogeny
The nuclear rRNA data set contains 2819 aligned positions, of which 77 are excluded because of ambiguous alignment (positions 701–705, 874–885, 2109–2160, and 2408–2415). The remaining 2743 characters contain 2547 invariant positions, 68 phylogenetically uninformative positions, and 128 phylogenetically informative positions. Likelihood-ratio tests of alternative evolutionary models select the General Time-Reversible model GTR + I +
as the one that best fits the nuclear rRNA data (Table 3). Plots of absolute numbers of substitutions against ML corrected sequence divergences reveal a linear pattern indicating that there is little substitutional saturation in the rRNA data.
|
Parsimony analysis produces four most-parsimonious trees of 317 steps (results not shown). Maximum-likelihood analysis produces essentially the same phylogenetic topology with a lnL of –5701.07 (Fig. 3). Differences between the parsimony and ML topologies are restricted to short branches within the families Hynobiidae and Plethodontidae. Likelihood-ratio tests reveal that two branches in the ML tree are not statistically different from zero length (Fig. 3). However, neither of these branches groups different salamander families. The Bayesian posterior distribution produces a consensus topology almost identical to the parsimony and ML trees. A stationary lnL is reached at approximately 950,000 generations with a mean lnL of –5738.27, a maximum of –5718.29, and a variance of 53.79. For a conservative analysis, all trees from generations one through 1,000,000 are discarded. Additional Bayesian analyses yield topologically identical results, suggesting that a stationary posterior distribution is reached (results not presented).
|
0.05 level are denoted with an asterisk. Two branches (A and B) that show concordance with the mtDNA ML and Bayesian tree are labeled. Shaded boxes denote the individual salamander families.The sister-taxon relationship of Hynobiidae and Cryptobranchidae is strongly supported by parsimony and Bayesian analysis (bootstrap 99%, PP 1.0; Fig. 3). An alternative topology without this grouping is rejected by the WS-R test and by Bayesian criteria, but it is not rejected by the SH test (Table 4). The sister-taxon relationship of Ambystomatidae and Dicamptodontidae is strongly supported by parsimony and statistically supported by the Bayesian analysis (bootstrap 96%, PP 1.0; Fig. 3). Alternative topologies without this grouping are not rejected by the Wilcoxon signed-ranks test or the SH test; however, the P-values of both tests approach significance (Table 4). Placement of Salamandridae as the sister taxon to the ambystomatid and dicamptodontid clade is also strongly supported by both parsimony and Bayesian analyses (bootstrap 96%, PP 1.0; Fig. 3). However, alternative topologies in which these three families do not form a clade cannot be rejected statistically by the WS-R and SH tests (Table 4).
|
Monophyly of the internally fertilizing suborder Salamandroidea is statistically rejected using Bayesian criteria, but it is not rejected under the WS-R and SH tests (Table 4). All analyses place Amphiumidae, Plethodontidae, Rhyacotritonidae, and a clade comprising the remaining families as deep branches in salamander phylogeny. Furthermore, Amphiumidae is consistently placed as the sister taxon to all remaining salamander families across all analyses of the nuclear data set. Relationships among these four branches are not resolved with strong support by parsimony (Fig. 3), but the Rhyacotritonidae is strongly supported as the sister taxon to a clade containing all remaining families except for Amphiumidae and Plethodontidae with Bayesian posterior probabilities of 1.0 (Fig. 3).
Mitochondrial DNA Phylogeny
Alignment of mtDNA sequences produces a character matrix of 2115 base pairs, of which 286 positions are excluded because of ambiguous alignment (positions 154–183, 376–388, 402–408, 431–435, 441–448, 461–467, 518–525, 539–545, 558–563, 595–601, 867–908, 1638–1664, 1678–1683, 1715–1721, 1734–1744, 1757–1763, 1813–1819, 1832–1838, 1871–1878, 1892–1901, 1911–1932, 1956–1962, 1968–1972, 1995–1999, 2013–2014, 2027–2033, 2065–2070, and 2084–2085). The multiple alignment contains 555 invariant characters, 152 parsimony-uninformative characters, and 1122 parsimony-informative characters. Likelihood-ratio tests of alternative evolutionary models select the GTR + I + model as optimal for the mtDNA data (Table 3). Plots of absolute numbers of substitutions against ML-corrected sequence divergences reveal patterns consistent with substitutional saturation in the mtDNA protein-coding regions but not in tRNA genes.
Parsimony analysis of the mtDNA data produces a single most parsimonious tree of 8285 steps (Fig. 4a). Maximum-likelihood analysis produces an optimal tree with a lnL score of –33,780.27 (Fig. 4b). Bayesian analysis produces a phylogenetic topology identical to the ML tree. A stationary lnL occurs after approximately 55,000 generations with a mean lnL of –33,799.71, a maximum of –33,782.28, and a variance of 54.3. For a conservative analysis, all trees from generations one through 100,000 are discarded. Replicated Bayesian analyses yield identical results, suggesting that the optimal posterior distribution is reached (results not presented).
|
A monophyletic Caudata is strongly supported in both the parsimony and Bayesian analyses (bootstrap 98%; decay index 30; PP 1.0; Fig. 4a, Fig. 4b). Alternative topologies without this relationship are rejected using all statistical tests (Table 4). All analyses produce topologies inconsistent with monophyly of internally fertilizing salamanders (Fig. 4a, Fig. 4b). A monophyletic Salamandroidea is rejected using the WS-R tests and under Bayesian criteria but not by the SH test (Table 4).
Except for Proteidae, all families with multiple representatives form strongly supported monophyletic groups in all analyses (Fig. 4a, Fig. 4b). These results are particularly noteworthy for family Hynobiidae, whose monophyly has been questioned (Trueb, 1993). Branch support in the parsimony tree for relationships among salamander families is low. The clade comprising Cryptobranchidae and Hynobiidae receives the strongest support in the parsimony tree (bootstrap 74%; decay index 11) (Fig. 4a). Alternative topologies without this relationship are not rejected statistically by the WS-R test (Table 4).
Branches grouping salamander families in the ML tree are short relative to branches immediately ancestral to individual families. Likelihood-ratio tests reveal four branches not statistically different from zero length (Fig. 4b). Three of these zero-length branches represent deep interfamilial relationships. However, posterior probabilities assigned to three of the nodes in the mtDNA Bayesian tree show statistical support (PP = 1.0) for interfamilial relationships (Fig. 4b; Table 4): (1) a clade comprising Ambystomatidae, Sirenidae, Proteidae, Salamandridae, and Dicamptodontidae (labeled branch B in Fig. 4b); (2) a clade composed of the Cryptobranchidae and Hynobiidae; and (3) a clade comprising the above two groups (labeled branch A in Fig. 4b). Tests of alternative ML topologies lacking branches A and B and the Cryptobranchidae-Hynobiidae clade are not statistically rejected by the SH test (Table 4).
Monophyly of family Proteidae is not recovered in our analyses. The European species, Proteus anguinus, forms the sister taxon to a group containing all remaining salamanders in the parsimony tree (Fig. 4a). In the likelihood-based analyses, Proteus anguinus forms the sister taxon to a clade comprising Necturus and families Ambystomatidae and Sirenidae; however, this relationship does not receive a significant posterior probability, and the branch immediately ancestral to this group is not statistically different from zero length (Fig. 4b). An alternative topology in which Proteus anguinus and genus Necturusare united to form a monophyletic family Proteidae is statistically rejected by the Wilcoxon signed-ranks test but not by the SH test (Table 4). Rejection of this alternative hypothesis approaches significance using Bayesian criteria (Table 4). Recent studies using partial 12S mtDNA sequences (Trontelj and Goricki, 2003) and RAG-1 sequences (Wiens et al., 2005) support monophyly of Proteidae, although only the latter analysis seems statistically robust.
Kumazawa and Nishida (1993) suggest that the slower evolution of mtDNA tRNA stems might provide phylogenetic information for deep phylogenetic relationships, and they provide convincing examples of recovery of deep animal relationships with strong branch support. Parsimony analysis of the eight mitochondrial tRNA stem sequences yields 29 trees of 1171 steps in length. A strict consensus of these trees produces a large polytomy and fails to recover monophyly of the families Salamandridae and Hynobiidae (results not shown), clades strongly supported in the total-mtDNA parsimony analysis. Thus, restricting the analysis to slower-evolving sites in tRNA genes does not enhance deep phylogenetic resolution.
Combined Analyses
Partitioned Bayesian analysis of a combined mtDNA and nuclear rRNA data matrix using complete taxon sampling produces a tree with an average lnL of –39814.18, a maximum of –39794.51, and a variance of 51.52 (Fig. 5). This tree is nearly identical to a tree produced from a tripartitioned Bayesian analysis of a combined mtDNA, nuclear rRNA and morphological data matrix, which had an average lnL of –40172.87. A stationary distribution is reached in both analyses by the 25,000th generation, and the first 100,000 generations are treated as burn-in. The combined-data Bayesian trees are very similar to the trees produced in Bayesian analysis of the individual molecular data sets and find strong support for all interfamilial relationships supported in the single-partition trees. However, the combined trees provide increased support for the most basal interfamilial branching event between the Amphiumidae and a clade containing all remaining salamander families (Fig. 5). The only uncertain family-level relationships are the exact placement of lineages representing the Proteidae (Necturus and Proteus) and Sirenidae. However, these families are firmly placed within the clade identified by branch B in the individual molecular Bayesian trees. There was no appreciable difference in results between combined-data analyses using all taxa from all data sets and combined-data analyses using a pruned taxon sampling (results not shown).
|
Parsimony analysis of the full-taxa, combined morphological and molecular data matrix produces a single tree of 10221 steps (tree not shown). As in the mtDNA parsimony tree, the combined-data tree places Proteus anguinus outside a clade containing the remaining taxa. However, this placement and most other deep internal branches are poorly supported. Only two interfamilial relationships receive bootstrap values greater than 65% in the combined parsimony tree: (1) the clade composed of Cryptobranchidae and Hynobiidae (99%) and (2) the clade composed of Ambystomatidae and Dicamptodontidae (86%). There was no appreciable difference in results between combined-data parsimony analyses using all taxa from all data sets and combined-data analyses using a pruned taxon sampling (results not shown). The sister-group relationship of Ambystomatidae and Dicamptodontidae and their sister-group relationship to the Salamandridae are consistent with phylogenetic results of Titus and Larson (1995) based on mitochondrial genes encoding 12S and 16S rRNA.
Mitochondrial DNA Simulation Results
The relationship between branch lengths, the minimal decay index needed for statistical significance, and Bayesian posterior probabilities show similar patterns across the three separate sets of mtDNA-based simulations. Recovery of both significant decay indices and posterior probabilities decreases as the internal branch (y) moves deeper into the tree and as the length of y is shortened (Fig. 6). However, comparisons across the three sets of simulated trees reveal a consistent pattern of higher statistical power of posterior probabilities relative to decay indices.
|
Differences in the graphical patterns of Figure 6 are partly a function of differences in branch lengths among the three starting trees. For example, recovery of the true tree from the simulated data sets is higher in the Hynobiidae-based simulations than the Amphiuma and Necturus-based simulations. The greater power in recovery of the true phylogeny in the Hynobiidae-based simulations probably results from availability of a closer outgroup for this four-taxon subset (Fig. 1). This increased power may be influenced also by use of the more parameter-rich General Time-Reversible model of substitution in this simulation relative to the less complex HKY model used in Amphiuma and Necturus-based simulations.
Based on the Hynobiidae results, an internal branch must have at least 0.03 substitutions per site to be recovered with significant decay indices in 90% to 100% of the subsets when placed at a phylogenetic depth of 0.13 substitutions per site. Bayesian analysis can recover branches of similar depth with average posterior probabilities of 0.9 to 1.0 when internal branches are greater than approximately 0.015 substitutions per site. When the internal branch is placed at a phylogenetic depth of 0.18 substitutions per site, the internal branch needs to be at least 0.045 in length to be recovered with significant decay indices in 90% to 100% of the subsets. Bayesian analysis of trees with an internal-branch depth of 0.18 substitutions per site can be recovered with posterior probabilities in the 0.9 to 1.0 range when the internal-branch length is at least 0.035 substitutions per site. When the phylogenetic depth of the internal branch increases to 0.23 substitutions per site, the internal branch must have at least 0.06 substitutions per site for recovery with significant decay indices in 90% to 100% of the subsets and at least 0.035 substitutions per site to be recovered with an average posterior probability of 0.9 to 1.0.
Likelihood-Based Character Change for mtDNA Data
MtDNA character support was compared for two deep interfamilial branches receiving significant Bayesian posterior probabilities (branches A and B in Fig. 4b) and an interfamilial branch lacking such support (branch C in Fig. 4b) to identify the kinds of character changes that contribute the strongest evidence favoring clades A and B. For branch A, 20 of the 25 character positions with the highest difference in lnL between the ML tree and the alternative tree have a difference of at least 1.0 (Table 5). The top seven characters have a difference of at least 2.0 (Table 5). For branch B, 23 of the top 25 lnL differences are greater than 1.0, and six characters have a difference of at least 2.0 (Table 6). For branch C, 11 of the highest 25 lnL differences are greater than 1.0 and two are greater than 2.0 (Appendix 2, available at www.systematicbiology.org). Character changes with the highest lnL differences (particularly those with a difference greater than 2.0) usually involve transversions at character positions with low evolutionary rates, typically in 1st or 2nd codon positions or stem structures of tRNA genes (Tables 5, 6; Appendix 2).
|
|
|
Discussion |
|---|
|
TopAbstractMaterials and MethodsResultsDiscussionReferences |
|---|
Challenges in Deep Phylogenetic Reconstruction
Phylogenetic relationships among salamander families illustrate the analytical challenges inherent to inferring phylogenies in which terminal branches are temporally very long relative to internal ones (Felsenstein, 1978; Huelsenbeck, 1997; 1998). As phylogenetic depth increases relative to the length of an internal branch, characters representing true phylogenetic signal become an increasingly smaller portion of the total informative characters in a data set, and parallel changes along terminal branches increase. Together, these factors increase the difficulty of correctly recovering deep phylogenetic relationships with strong branch support and of statistically rejecting alternative phylogenetic hypotheses. This phenomenon often leads to an interpretation of simultaneous branching among multiple lineages. Under these conditions, it is difficult to distinguish analytical artifacts that yield polytomous trees (called a "soft" polytomy) from evolutionarily simultaneous branching of multiple lineages (a "hard" polytomy sensu Maddison [1989]).
Our simulations address the phylogenetic resolving power of our mtDNA data and phylogenetic methods using evolutionary models derived from the data to simulate sequence evolution on trees whose internal and terminal branches vary in length. Our mtDNA-based simulations quantify the positive relationship between the evolutionary depth of an internal branch in a phylogenetic tree and the minimum evolutionary duration needed for its recovery with strong branch support (see also Halanych, 1998 and Philippe et al., 1994).
Our results suggest that the mtDNA data have limited ability to recover relatively old and short branches and may not provide substantial support for many branching events deep in salamander phylogeny. Lineages ancestral to groupings of two or more salamander families have a phylogenetic depth of at least 50 million years and possibly as large as 150 million years (reviewed in Larson, 1991; Gao and Shubin, 2001; 2003). Model-based phylogenetic analyses are expected to be more effective in detecting relatively short branches with statistical significance than is unweighted parsimony analysis in four-taxon tests (Hillis et al., 1994; Hulesenbeck, 1995). Our Hynobiidae-based simulations, interpreted using a rough molecular clock calibrated for our mtDNA gene region, suggest that with 
1800 base pairs of aligned sequences, a lineage ancestral to taxa separated for 35 million years (branch length of 0.23 substitutions/site) must be at least 9 million years in duration to be recovered under parsimony with statistically significant decay indices 90% of the time and at least 5 million years in duration on average to be recovered with Bayesian posterior probabilities of 0.9 to 1.0. Ancestral lineages of even half these durations are much less likely to be recovered with strong support (Fig. 6). From these calculations, it seems clear that internal branches ancestral to salamander family lineages would need to have had substantial evolutionary durations to have generated sufficient phylogenetic signal in mtDNA sequences.
These estimates assume roughly constant nucleotide substitution rates over time and that appropriate outgroups are available; they are underestimates of the duration needed for recovery of an ancestral lineage if only very distant outgroups are available. Increased taxon sampling (Hillis, 1996; Poe, 2003; Poe and Swofford, 1999) and closely related outgroups (Halanych et al., 1999) should increase resolving power relative to the conditions simulated. However, increased taxon sampling across salamanders may have limited ability to reduce problems of long-branch attraction because the extant diversity found within salamander families generally represents only a small portion of the total evolutionary duration of a family-level branch. Consequently, even ancestral lineages of substantial evolutionary duration may be very difficult to resolve with confidence if they occur deep in a phylogenetic tree. This analytical problem may be avoided by using sequences that evolve more slowly than the ones modeled (e.g., nuclear rRNA versus mtDNA) due to the slower accumulation of misleading information. However, phylogenetic signal would be diminished proportionately and could substantially limit the resolution of short internal branches (Donoghue and Sanderson, 1992; Fishbein et al., 2001).
Substantial evolutionary changes in morphology and geographic distribution could have occurred early in the evolutionary history of salamanders on ancestral branches that are of millions of years in duration. However, unless these branches were of a minimum duration that could generate sufficient phylogenetic information to withstand long periods of terminal evolution, they would be nearly invisible to currently available phylogenetic methods and could be perceived as a rapid branching event.
Rapid Cladogenesis of Salamander Families?
Short internal branch lengths are frequently highlighted as potential "radiations" of lineages, often with the connotation that cladogenesis occurred simultaneously among multiple lineages. Although the evidence for such radiations has grown (e.g., Schluter, 2000), only recently have rigorous tests been applied to discriminate hard and soft polytomies (e.g., Fishbein et al., 2001; Jackman et al., 1999; Misof et al., 2001; Poe and Chubb, 2004; Slowinski, 2001; Walsh et al., 1999). Likelihood-ratio tests of nonzero branch lengths (Slowinski, 2001) offer a computationally simple way to test the null hypothesis of nonzero branch lengths. Such tests may be conservatively biased because only a single branch is allowed to vary between trees (the potentially zero-length branch). Alternative methods such as the parametric bootstrap may offer a more powerful test of nonzero branch lengths (Fishbein et al., 2001; Knowles, 2000; Poe and Chubb, 2004). However, rejection of nonzero branch lengths using a likelihood-ratio test represents robust evidence for the presence of a particular internal branch.
Perhaps the most crucial aspect in diagnosis of a hard polytomy among lineages is the signature of nonzero branch lengths in molecular phylogenies of multiple unlinked loci (Poe and Chubb, 2004; Slowinski, 2001). Using likelihood-ratio tests, we could not reject the hypothesis of zero branch length for three interfamilial branches in the salamander mtDNA ML tree; however, these same branches were found with statistical support in the rRNA ML tree. The mtDNA branch-length patterns are less likely to have arisen as a result of a multifurcating gene tree and are more likely a function of the long time since cladogenesis among salamander families. A major benefit of incorporating our simulation work into our assessment of rapid branching patterns is that it gives us insight into the phylogenetic patterns expected for resolving deep relationships. Decay of phylogenetic signal in our mtDNA data at deep branches during the long evolutionary time since cladogenesis offers the best explanation for zero-length branches deep in the mtDNA ML tree.
Salamander Phylogeny and Potentially Misleading Effects of Weighting Saturated Data
Parsimony analysis of the newly reported mitochondrial genomic sequences provides strong independent support to a number of groupings consistent with prior analyses of nuclear genomic data and morphological characters (Larson et al., 2003; Wiens et al., 2005). These groupings span a large scale of salamander phylogenetic history and include (Fig. 4a) monophyly of (1) Urodela, (2) family Ambystomatidae, (3) family Amphiumidae, (3) family Cryptobranchidae, (4) family Dicamptodontidae, (5) family Hynobiidae, (6) family Plethodontidae, (7) family Salamandridae (8) family Sirenidae, (9) genus Necturus, (10) bolitoglossine plethodontids (represented here by Parvimolge and Pseudoeurycea), (11) hynobiid genus Onychodactylus, and (12) hynobiid salamanders excluding genus Onychodactylus. Model-based analyses (maximum likelihood, Bayesian analysis) provide strong support to these same groupings (Fig. 4b). As expected from our simulations, model-based analyses demonstrate stronger resolving power in finding statistical support for some additional groupings that received only weak support from the parsimony analysis: (1) monophyly of hemidactyliine plethodontids (represented here by Eurycea, Pseudotriton, and Stereochilus); (2) monophyly of Cryptobranchoidea (Cryptobranchidae and Hynobiidae); (3) grouping of North American newts Notophthalmus and Taricha relative to other salamandrids; (4) grouping of newts Notophthalmus, Pleurodeles, and Taricha relative to other salamandrids; and (5) grouping of true salamanders (Chioglossa and Salamandra) relative to other salamandrids. Only the grouping of Hynobius and Salamandrella seems stronger in the parsimony analysis than in the Bayesian analysis (Fig. 4).
Several results indicate that most branches grouping two or more salamander families are relatively short ones located deep in the evolutionary history of salamanders. First, parsimony analysis provides no convincing phylogenetic groupings of salamander families other than the Cryptobranchoidea and the clade containing all salamanders (Urodela). Likelihood-based estimates of branch lengths show that the branches grouping two or more salamander families are very short relative to those immediately ancestral to individual families. Our simulations show that these conditions are ones in which model-based methods of phylogenetic reconstruction should outperform parsimony analysis.
Model-based methods nonetheless find statistical support for groupings of families that seem anomalous based on prior information (Larson and Dimmick, 1993; Chippindale et al., 2004; Wiens et al., 2005). Specifically, branches A and B on Figure 4b both receive strong support by the model-based analyses, and both require rejection of monophyly of salamander suborder Salamandroidea, which comprises all internally fertilizing salamanders. Both of these branches are recovered independently by model-based analyses of nuclear rRNA gene sequences, although only branch A receives statistical support from these analyses (Fig. 3). Reasons for considering these results anomalous include (1) support for a monophyletic Salamandroidea in parsimony and Bayesian analyses of nuclear RAG-1 sequences combined with the rRNA sequences and other mitochondrial genomic sequences (Chippindale et al., 2004; Wiens et al., 2005), and (2) expected evolutionary stability of the internal-fertilization system as a synapomorphy of Salamandroidea.
The internal-fertilization system of Salamandroidea is a suite of morphological and life-history characteristics considered a functionally burdened complex unlikely to show homoplastic evolution or to be secondarily lost (see discussions by Larson, 1991; Sever and Brizzi, 1998). Male and female cloacal glands associated with internal fertilization are similar among salamandroid families except that spermathecae are more complex structures in plethodontids than in other families (Sever, 1994). Although fertilization is not well documented in Sirenidae, absence of the cloacal glands that function in salamandroid fertilization and absence of sperm in the oviducts of reproductive females strongly suggest external fertilization (Sever et al., 1996) or at least a fertilization system very different from salamandroid fertilization. Internal fertilization appears not to have been lost within any salamandroid families, indicating that it is indispensable to their reproductive biology.
Given strong reasons to expect monophyly of Salamandroidea, the statistical support for branches A and B of Figure 4b from our model-based analyses provides a means to identify conditions under which model-based methods may produce misleading results. Character support for clades A and B in the maximum-likelihood analyses of mitochondrial DNA data depends largely upon a small number of character positions (Tables 5 and 6) that individually contribute relatively large amounts to total lnL in the maximum-likelihood tree. The high lnL at these positions depends on substitutions expected to be rare for the favored evolutionary model combined with a relatively low evolutionary rate for the character position. The biological interpretation of this result is that homoplasy is unlikely to produce a sharing of derived states for such characters. A major contrast between interpretations of parsimony-based analyses and model-based approaches is that parsimony analysis finds the mitochondrial genomic data equivocal regarding monophyly of Salamandroidea, whereas model-based criteria in the Bayesian analysis reject monophyly of Salamandroidea largely on patterns from heavily weighted characters.
Our results therefore suggest that despite strong evidence for greater statistical power of maximum-likelihood and Bayesian methods relative to unweighted parsimony analysis, the model-based methods may provide misleading results where a parsimony analysis would be more conservative in judging the results inconclusive. Either over- or under simplification of estimated models could produce inaccurate trees with inflated confidence (Buckley, 2002), and model-based phylogenetic methods are not immune to long-branch attraction when evolutionary models are violated (Huelsenbeck, 1997; Swofford et al., 2001).
Statistical support for a mistaken topology is perhaps most likely to occur when resolving relatively short internal branches located deep in the phylogenetic history of a group, especially when some substitutional saturation has occurred in the sequences being compared. In such situations, an analysis of the contributions of individual characters to the statistical significance of model-based results is recommended to determine whether such significance results from a relatively small number of heavily weighted characters, as seen for branches A and B in Figure 4b (Tables 5 and 6). Such outcomes would be most sensitive to errors in estimation of an evolutionary model or temporal changes in an evolutionary model (Lemmon and Moriarty, 2004; Sullivan and Swofford, 1997). This character-based analysis may be the most objective way to identify misleading statistical confidence derived from a Bayesian analysis (Buckley et al., 2001; Buckley, 2002; Goldman et al., 2000).
|
|
Acknowledgements |
|---|
This work was supported by the National Science Foundation (DDIG-0105166). We thank M. Fishbein, R. Glor, E. Jockusch, J. Kolbe, K. Kozak, M. Mahoney, K. Nicholson, J. Schulte, T. Townsend, and other members of the Washington University systematics reading group for comments on earlier drafts of this manuscript. We thank Malcolm Tobias for assistance with parallel computing and the Washington University Center for Scientific Parallel Computing for providing computer resources. Jim Wilgenbusch was helpful with PAUP technicalities. Carla Cicero, David Green, Masafumi Matsui, and Paul Moler were kind in providing tissues and museum numbers.
|
References |
|---|
|
TopAbstractMaterials and MethodsResultsDiscussionReferences |
|---|
-
Alfaro M. E., Zoller S., Lutzoni F. Bayes or bootstrap? A simulation study comparing the performance of Bayesian Markov chain Monte Carlo sampling and bootstrapping in assessing phylogenetic confidence. Mol. Biol. Evol. (2003) 20:255–266.
Altekar G., Dwarkadas S., Huelsenbeck J. P., Ronquist F. Parallel Metropolis coupled Markov chain Monte Carlo for Bayesian phylogenetic inference. Bioinformatics (2004) 20:407–415.
Anderson S., Bankier A. T., Barrell B. G., de Bruijn M. H. L., Coulson A. R., Drouin J., Eperon I. C., Nierlich D. P., Roe B. A., Sanger F., Schreier P. H., Smith A. J. H., Staden R., Young I. G. Sequence and organization of the human mitochondrial genome. Nature (1981) 290:457–465.[CrossRef][Medline]
Bermingham E., McCafferty S. S., Martin A. P. Fish biogeography and molecular clocks: Perspectives from the Panamanian Isthmus. In: Molecular systematics of fishes—Kocher T. D., Stepien C. A., eds. (1997) San Diego, California: Academic Press. Pages 113–128.
Bremer K. Branch support and tree stability. Cladistics (1994) 10:295–304.[CrossRef][Web of Science]
Buckley T. R. Model misspecification and probabilistic tests of topology: Evidence from empirical data sets. Syst. Biol. (2002) 51:509–523.
Buckley T. R., Simon C., Shimodaira S., Chambers G. K. Evaluating hypotheses on the origin and evolution of the New Zealand alpine cicadas (Maoricicada) using multiple-comparison tests of tree topology. Mol. Biol. Evol. (2001) 18:223–234.
Chippindale P. T., Bonnet R. M., Baldwin A. S., Wiens J. J. Phylogenetic evidence for a major reversal of life-history evolution in plethodontid salamanders. Evolution (2004) 58:2809–2822.[CrossRef][Web of Science][Medline]
Donoghue M. J. Phylogenies and the analysis of evolutionary sequences, with examples from seed plants. Evolution (1989) 43:1137–1156.[CrossRef][Web of Science]
Donoghue M. J., Sanderson M. J. The suitability of molecular and morphological evidence in reconstructing plant phylogeny. In: Molecular systematics of plants—Soltis P. S., Soltis D. E., Doyle J. J., eds. (1992) New York: Chapman and Hall. Pages 340–368.
Duellman W. E., Trueb L. Biology of amphibians. (1986) New York: McGraw Hill.
Felsenstein J. Cases in which parsimony or compatibility methods will be positively misleading. Syst. Zool. (1978) 27:401–410.
Felsenstein J. Confidence limits on phylogenies with a molecular clock. Syst. Zool. (1985) 34:152–161.
Fishbein M., Hibsch-Jetter C., Soltis D. E., Hufford L. Phylogeny of Saxifragales (Angiosperms, Eudicots): Analysis of a rapid, ancient radiation. Syst. Biol. (2001) 50:817–847.
Gao K. Q., Shubin N. H. Late Jurassic salamanders from northern China. Nature (2001) 410:574–577.[CrossRef]
Gao K. Q., Shubin N. H. Earliest known crown-group salamanders. Nature (2003) 422:424–428.[CrossRef]
Goldman N., Anderson J. P., Rodrigo A. G. Likelihood-based tests of topologies in phylogenetics. Syst. Biol. (2000) 49:652–670.
Goldman N., Whelan S. Statistical tests of gamma-distributed rate heterogeneity in models of sequence evolution in phylogenetics. Mol. Biol. Evol. (2000) 17:975–978.
Halanych K. M. Considerations for reconstructing metazoan history: Signal, resolution, and hypothesis testing. Am. Zool. (1998) 38:929–941.[Web of Science]
Halanych K. M., Demboski J. R., van Vuuren B. J., Klein D. R., Cook J. A. Cytochrome b phylogeny of North American hares and jackrabbits (Lepus, Lagomorpha) and the effects of saturation in outgroup taxa. Mol. Phylogenet. Evol. (1999) 11:213–221.[CrossRef][Web of Science][Medline]
Hasegawa M., Kishino H., Yano T. Dating of the human-ape splitting by a molecular clock of mitochondrial DNA. J. Mol. Evol. (1985) 21:160–174.
Hay J. M., Ruvinsky I., Hedges S. B., Maxson L. R. Phylogenetic relationships of amphibian families inferred from DNA sequences of mitochondrial 12S and 16S ribosomal RNA genes. Mol. Biol. Evol. (1995) 12:928–937.[Abstract]
Hillis D. M. Inferring complex phylogenies. Nature (1996) 383:130–131.[CrossRef][Medline]
Hillis D. M., Huelsenbeck J. P., Cunningham C. W. Application and accuracy of molecular phylogenies. Science (1994) 264:671–677.
Hrbek T., Larson A. The evolution of diapause in the killifish family Rivulidae (Atherinomorpha, Cyprinodontiformes): A molecular phylogenetic and biogeographic perspective. Evolution (1999) 53:1200–1216.[CrossRef][Web of Science]
Huelsenbeck J. P. Performance of phylogenetic methods in simulation. Syst. Biol. (1995) 44:17–48.
Huelsenbeck J. P. Is the Felsenstein zone a fly trap? Syst. Biol. (1997) 46:69–74.
Huelsenbeck J. P. Systematic bias in phylogenetic analysis: Is the Strepsiptera problem solved? Syst. Biol. (1998) 47:519–537.[Web of Science][Medline]
Huelsenbeck J. P., Ronquist F. MRBAYES: Bayesian inference of phylogenetic trees. Bioinformatics (2001) 17:754–755.
Huelsenbeck J. P., Ronquist F., Nielsen R., Bollback J. P. Bayesian inference of phylogeny and its impact on evolutionary biology. Science (2001) 294:2310–2314.
Jackman T. R., Larson A., de Queiroz K., Losos J. B. Phylogenetic relationships and tempo of early diversification in Anolis lizards. Syst. Biol. (1999) 48:254–285.[Abstract]
Knowles L. L. Tests of Pleistocene speciation in montane grasshoppers (genus Melanopus) from the sky islands of western North America. Evolution (2000) 54:1337–1348.[CrossRef][Web of Science][Medline]
Kumazawa Y., Nishida M. Sequence evolution of mitochondrial transfer RNA genes and deep-branch animal phylogenetics. J. Mol. Evol. (1993) 37:380–398.[Web of Science][Medline]
Larget B., Simon D. L. Markov chain Monte Carlo algorithms for the Bayesian analysis of phylogenetic trees. Mol. Biol. Evol. (1999) 16:750–759.[Web of Science]
Larson A. A molecular perspective on the evolutionary relationships of the salamander families. Evol. Biol. (1991) 25:211–277.
Larson A. The comparison of morphological and molecular data in phylogenetic systematics. In: Molecular ecology and evolution: Approaches and applications—DeSalle R., Schierwater B., eds. (1998) Basel, Switzerland: Birkhäuser Verlag. Pages 275–296.
Larson A., Dimmick W. W. Phylogenetic relationships of the salamander families: An analysis of congruence among morphological and molecular characters. Herpetol. Monogr. (1993) 7:77–93.[CrossRef]
Larson A., Weisrock D. W., Kozak K. H. Phylogenetic systematics of salamanders (Amphibia: Urodela), a review. In: Reproductive biology and phylogeny of Urodela (Amphibia)—Sever D. M., ed. (2003) Enfield, New Hampshire: Science Publishers, Inc. Pages 31–108.
Lemmon A. R., Moriarty E. C. The importance of proper model assumption in Bayesian phylogenetics. Syst. Biol. (2004) 53:265–277.
Lewis P. O. A likelihood approach to estimating phylogeny from discrete morphological character data. Syst. Biol. (2001) 50:913–925.
Macey J. R., Larson A., Ananjeva N. B., Fang Z. L., Papenfuss T. J. Two novel gene orders and the role of light-strand replication in rearrangement of the vertebrate mitochondrial genome. Mol. Biol. Evol. (1997a) 14:91–104.[Abstract]
Macey J. R., Larson A., Ananjeva N. B., Papenfuss T. J. Replication slippage may cause parallel evolution in the secondary structures of mitochondrial transfer RNAs. Mol. Biol. Evol. (1997b) 14:30–39.[Abstract]
Macey J. R., Schulte J. A., Ananjeva N. B., Larson A., Rastegar-Pouyani N., Shammakov S. M., Papenfuss T. J. Phylogenetic relationships among agamid lizards of the Laudakia caucasia species group: Testing hypotheses of biogeographic fragmentation and an area cladogram for the Iranian Plateau. Mol. Phylogenet. Evol. (1998a) 10:118–131.[CrossRef][Web of Science][Medline]
Macey J. R., Schulte J. A., Larson A., Ananjeva N. B., Wang Y. Z., Pethiyagoda R., Rastegar-Pouyani N., Papenfuss T. J. Evaluating trans-Tethys migration: An example using acrodont lizard phylogenetics. Syst. Biol. (2000) 49:233–256.
Macey J. R., Schulte J. A., Larson A., Fang Z. L., Wang Y. Z., Tuniyev B. S., Papenfuss T. J. Phylogenetic relationships of toads in the Bufo bufo species group from the eastern escarpment of the Tibetan Plateau: A case of vicariance and dispersal. Mol. Phylogenet. Evol. (1998b) 9:80–87.[CrossRef][Web of Science][Medline]
Macey J. R., Schulte J. A., Larson A., Tuniyev B. S., Orlov N., Papenfuss T. J. Molecular phylogenetics, tRNA evolution, and historical biogeography in anguid lizards and related taxonomic families. Mol. Phylogenet. Evol. (1999) 12:250–272.[CrossRef][Web of Science][Medline]
Macey J. R., Strasburg J. L., Brisson J. A., Vredenburg V. T., Jennings M., Larson A. Molecular phylogenetics of western North American frogs of the Rana boylii species group. Mol. Phylogenet. Evol. (2001) 19:131–143.[CrossRef][Web of Science][Medline]
Maddison W. P. Reconstructing character evolution on polytomous cladograms. Cladistics (1989) 5:365–377.[CrossRef][Web of Science]
Maddison W. P., Maddison D. R. MacClade4: Analysis of phylogeny and character evolution, version 4.0. (2000) Sunderland, Massachusetts: Sinauer Associates.
Misof B., Rickert A. M., Buckley T. R., Fleck G., Sauer K. P. Phylogenetic signal and its decay in mitochondrial SSU and LSU rRNA gene fragments of Anisoptera. Mol. Biol. Evol. (2001) 18:27–37.
Nylander J. A. A., Ronquist F., Huelsenbeck J. P., Nieves-Aldrey J. L. Bayesian phylogenetic analysis of combined data. Syst. Biol. (2004) 53:47–67.
Philippe H., Chenuil A., Adoutte A. Can the Cambrian explosion be inferred through molecular phylogeny? Development (Cambridge) (1994) (Supplement):15–25.
Poe S. Evaluation of the strategy of long-branch subdivision to improve accuracy of phylogenetic methods. Syst. Biol. (2003) 52:423–428.
Poe S., Chubb A. L. Birds in a bush: Five genes indicate explosive evolution of avian orders. Syst. Biol. (2004) 58:404–415.
Poe S., Swofford D. L. Taxon sampling revisited. Nature (1999) 398:299–300.
Posada D., Crandall K. A. Model Test: Testing the model of DNA substitution. Bioinformatics (1998) 14:817–818.
Rambaut A., Grassly N. C. Seq-Gen: An application for the Monte Carlo simulation of DNA sequence evolution along phylogenetic trees. Comput. Appl. Biosci. (1997) 13:235–238.
Riedl R. Order in living organisms. (1978) New York: Wiley.
Salim M., Maden B. E. H. Nucleotide sequence of Xenopus laevis 18S ribosomal RNA inferred from the gene sequence. Nature (1981) 291:205–208.[CrossRef][Medline]
Schluter D. The ecology of adaptive radiation. (2000) Oxford, England: Oxford University Press.
Sever D. M. Comparative anatomy and phylogeny of the cloacae of salamanders (Amphibia, Caudata). 1. Evolution at the family level. Herpetologica (1991a) 47:165–193.[Web of Science]
Sever D. M. Comparative anatomy and phylogeny of the cloacae of salamanders (Amphibia, Caudata). 2. Cryptobranchidae, Hynobiidae, and Sirenidae. J. Morphol. (1991b) 207:283–301.[CrossRef][Web of Science]
Sever D. M. Observations on regionalization of secretory activity in the spermathecae of salamanders and comments on phylogeny of sperm storage in female amphibians. Herpetologica (1994) 50:383–397.[Web of Science]
Sever D. M., Brizzi R. Comparative biology of sperm storage in female salamanders. J. Exp. Zool. (1998) 282:460–476.[CrossRef][Web of Science][Medline]
Sever D. M., Rania L. C., Krenz J. D. Reproduction of the salamander Siren intermedia Le Conte with special reference to oviducal anatomy and mode of fertilization. J. Morphol. (1996) 227:335–348.[CrossRef][Web of Science]
Shimodaira H., Hasegawa M. Multiple comparisons of log-likelihoods with applications to phylogenetic inference. Mol. Biol. Evol. (1999) 16:1114–1116.[Web of Science]
Simmons M. P., Pickett K. M., Miya M. How meaningful are Bayesian support values. Mol. Biol. Evol. (2004) 21:188–199.
Slowinski J. B. Molecular polytomies. Mol. Phylogenet. Evol. (2001) 19:114–120.[CrossRef][Web of Science][Medline]
Sorenson M. D. TreeRot, version 2. (1999) Boston University.
Sullivan J., Swofford D. L. Are guinea pigs rodents? The utility of models in molecular phylogenetics. J. Mammal. Evol. (1997) 4:77–86.[CrossRef]
Suzuki Y., Glazko G. V., Nei M. Overcredibility of molecular phylogenies obtained by Bayesian phylogenetics. Proc. Natl. Acad. Sci. USA (2002) 99:16138–16143.
Swofford D. L. PAUP*. In: Phylogenetic analysis using parsimony (* and other methods), version 4. (2002) Sunderland, Massachusetts: Sinauer Associates.
Swofford D. L., Olsen G. J., Waddell P. J., Hillis D. M. Optimality criteria II: Methods based on models of evolutionary change. In: Molecular systematics—Hillis D. M., Moritz C., Mable B. K., eds. (1996) Sunderland, Massachusetts: Sinauer Associates. Pages 426–503.
Swofford D. L., Waddell P. J., Huelsenbeck J. P., Foster P. G., Lewis P. O., Rogers J. S. Bias in phylogenetic estimation and its relevance to the choice between parsimony and likelihood methods. Syst. Biol. (2001) 50:525–539.
Templeton A. R. Phylogenetic inference from restriction endonuclease cleavage site maps with particular reference to the evolution of humans and the apes. Evolution (1983) 37:221–244.[CrossRef][Web of Science]
Titus T. A., Larson A. A molecular phylogenetic perspective on the evolutionary radiation of the salamander family Salamandridae. Syst. Biol. (1995) 44:125–151.
Trontelj P., Goricki S. Monophyly of the family Proteidae (Amphibia: Caudata) tested by phylogenetic analysis of mitochondrial 12S rDNA sequences. Nat. Croat. (2003) 12:113–120.
Trueb L. Patterns of cranial diversity among the Lissamphibia. In: The skull—Hanken J., Hall B. K., eds. (1993) Chicago, Illinois: University of Chicago Press. Pages 255–343.
Waddell P. J., Kishino H., Ota H. A phylogenetic foundation for comparative mammalian genomics. Genome Informatics (2001) 12:141–154.[Medline]
Walsh H. E., Kidd M. G., Moum T., Friesen V. L. Polytomies and the power of phylogenetic inference. Evolution (1999) 53:932–937.[CrossRef][Web of Science]
Weisrock D. W., Macey J. R., Ugurtas I. H., Larson A., Papenfuss T. J. Molecular phylogenetics and historical biogeography among salamandrids of the "true" salamander clade: Rapid branching of numerous highly divergent lineages in Mertensiella luschani associated with the rise of Anatolia. Mol. Phylogenet. Evol. (2001) 18:434–448.[CrossRef][Web of Science][Medline]
Wiens J. J. Missing data, incomplete taxa, and phylogenetic accuracy. Syst. Biol. (2003) 52:528–538.
Wiens J. J., Bonett R. M., Chippindale P. T. Ontogeny discombobulates phylogeny: Paedomorphosis and higher-level salamander relationships. Syst. Biol. (2005) 54:91–110.
Wilcox T. P., Zwickl D. J., Heath T. A., Hillis D. M. Phylogenetic relationships of the dwarf boas and a comparison of Bayesian and bootstrap measures of phylogenetic support. Mol. Phylogenet. Evol. (2002) 25:361–371.[CrossRef][Web of Science][Medline]
Yang Z. H. Maximum likelihood phylogenetic estimation from DNA sequences with variable rates over sites: Approximate methods. J. Mol. Evol. (1994) 39:306–314.[CrossRef][Web of Science][Medline]
Yang Z. H., Rannala B. Bayesian phylogenetic inference using DNA sequences: A Markov Chain Monte Carlo method. Mol. Biol. Evol. (1997) 14:717–724.[Abstract]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  V. Campbell and F.-J. Lapointe The Use and Validity of Composite Taxa in Phylogenetic Analysis Syst Biol, December 1, 2009; 58(6): 560 - 572. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. E. Horvath, D. W. Weisrock, S. L. Embry, I. Fiorentino, J. P. Balhoff, P. Kappeler, G. A. Wray, H. F. Willard, and A. D. Yoder Development and application of a phylogenomic toolkit: Resolving the evolutionary history of Madagascar's lemurs Genome Res., March 1, 2008; 18(3): 489 - 499. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. P. Townsend Profiling Phylogenetic Informativeness Syst Biol, April 1, 2007; 56(2): 222 - 231. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 P. Zhang, Y.-Q. Chen, H. Zhou, Y.-F. Liu, X.-L. Wang, T. J. Papenfuss, D. B. Wake, and L.-H. Qu Phylogeny, evolution, and biogeography of Asiatic Salamanders (Hynobiidae) PNAS, May 9, 2006; 103(19): 7360 - 7365. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. L. Mueller Evolutionary Rates, Divergence Dates, and the Performance of Mitochondrial Genes in Bayesian Phylogenetic Analysis Syst Biol, April 1, 2006; 55(2): 289 - 300. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||