Multi-locus phylogeny of dolphins in the subfamily Lissodelphininae: character synergy improves phylogenetic resolution
BMC Evolutionary Biology volume 6, Article number: 87 (2006)
Dolphins of the genus Lagenorhynchus are anti-tropically distributed in temperate to cool waters. Phylogenetic analyses of cytochrome b sequences have suggested that the genus is polyphyletic; however, many relationships were poorly resolved. In this study, we present a combined-analysis phylogenetic hypothesis for Lagenorhynchus and members of the subfamily Lissodelphininae, which is derived from two nuclear and two mitochondrial data sets and the addition of 34 individuals representing 9 species. In addition, we characterize with parsimony and Bayesian analyses the phylogenetic utility and interaction of characters with statistical measures, including the utility of highly consistent (non-homoplasious) characters as a conservative measure of phylogenetic robustness. We also explore the effects of removing sources of character conflict on phylogenetic resolution.
Overall, our study provides strong support for the monophyly of the subfamily Lissodelphininae and the polyphyly of the genus Lagenorhynchus. In addition, the simultaneous parsimony analysis resolved and/or improved resolution for 12 nodes including: (1) L. albirostris, L. acutus; (2) L. obscurus and L. obliquidens; and (3) L. cruciger and L. australis. In addition, the Bayesian analysis supported the monophyly of the Cephalorhynchus, and resolved ambiguities regarding the relationship of L. australis/L. cruciger to other members of the genus Lagenorhynchus. The frequency of highly consistent characters varied among data partitions, but the rate of evolution was consistent within data partitions. Although the control region was the greatest source of character conflict, removal of this data partition impeded phylogenetic resolution.
The simultaneous analysis approach produced a more robust phylogenetic hypothesis for Lagenorhynchus than previous studies, thus supporting a phylogenetic approach employing multiple data partitions that vary in overall rate of evolution. Even in cases where there was apparent conflict among characters, our data suggest a synergistic interaction in the simultaneous analysis, and speak against a priori exclusion of data because of potential conflicts, primarily because phylogenetic results can be less robust. For example, the removal of the control region, the putative source of character conflict, produced spurious results with inconsistencies among and within topologies from parsimony and Bayesian analyses.
Dolphins of the genus Lagenorhynchus are distributed in temperate to cool waters in the North Pacific, North Atlantic, and Southern oceans [1–4] (Figure 1). A short rostrum, relatively small, stout bodies, and flanks with horizontal flares of various contrasting patterns characterize all members of the genus. Within this group there is considerable variation in social structure and habitat, from coastal, shallow water Peale's dolphin (L. australis) that occurs in small groups within the Strait of Magellan and nearby fijords, to the meso-pelagic dusky dolphin (L. obscurus) that aggregates in groups of thousands along the continental shelves of New Zealand, South Africa, and South America.
Historically, color patterns, number of teeth, and the ratio of rostrum to brain case length were used as diagnostic characters to define dolphin genera, including the genus Lagenorhynchus [5–7]. However, these characters are now known to vary with sex and age within species . It is not surprising, therefore, that the classification of dolphin groups based on these characters proved particularly problematic for early taxonomists. For Lagenorhynchus, the ubiquitous use of these morphological characters resulted at one time or another in specimens being assigned (and re-assigned) to at least 8 different genera (e.g., Delphinus, ; Electra, ; Phocoena, ; Tursio, ; Leucopleurus, ; Clymenia, ; Sagmatius, ). Recently, molecular systematic studies of mitochondrial cytochrome b (cyt b) and control region (dloop) sequences were employed to address the taxonomic ambiguities within the family Delphinidae. The results of these studies, based on a single representative from each species, suggested a polyphyletic Lagenorhynchus, leaving many other relationships weakly supported and the overall taxonomic status of the group unresolved [12, 13]. For example, the cyt b cladogram of LeDuc et al.  indicated strong support for a monophyletic subfamily Lissodelphininae that contained L. obliquidens, L. obscurus, L. australis, L. cruciger, and the genera Cephalorhynchus and Lissodelphis but excluded the North Atlantic L. albirostris and L. acutus. A more recent Bayesian analysis of cyt b sequences by May-Collado and Agnarsson  supported the paraphyly of the genus Lagenorhynchus as suggested by LeDuc et al. , but increased taxonomic sampling (particularly of outgroups) provided increased phylogenetic resolution within the Lissodelphininae. Nevertheless, as with LeDuc et al. , some of the relationships among species of Lagenorhynchus and Cephalorhynchus within the Lissodelphininae remained unresolved.
Because of its rapid rate of coalescence and lack of recombination, mitochondrial DNA has been the molecule of choice for detailed studies of intra- and inter-specific evolution. However, the mitochondrial genome is a single, maternally inherited locus, and thus provides a perspective based on a single gene tree of female lineages. It is possible that the lack of phylogenetic resolution in LeDuc et al.'s  study was caused by limitations of a single, mitochondrial locus and/or the result of rapid divergence among taxa. However, distinguishing between alternative phylogenetic hypotheses requires the addition of more data from multiple loci and a detailed evaluation of the utility of various data partitions to resolve relationships within this group. To date, no further molecular systematic studies have attempted to address these issues.
In this paper we use four molecular markers (two nuclear and two mitochondrial) that vary in overall rates of evolution to diagnose relationships among species of dolphin in the genus Lagenorhynchus and among genera within the subfamily Lissodelphininae . To increase taxonomic representation, we include 34 individuals representing 9 species of the Lissodelphininae and North Atlantic Lagenorhynchus. In addition, we utilize a posteriori measures to quantify the interaction among data partitions in a simultaneous analysis, and to characterize the effects of removing sources of conflict on phylogenetic resolution under the optimality criterion of parsimony and with the Bayesian method. Several studies have presented methods to quantify the interaction of characters in simultaneous analyses [15–17]. These measures are either: (1) topological indices that measure the change in either tree length or structure ("topological congruence") [18, 19], and (2) measures of change in the amount of support at a particular node [15, 17, 20]. The reliability of these measures and their philosophical foundation has been questioned. In a recent review, Grant and Kluge (, pg. 409) advocated that a posteriori analyses of character partitions are heuristic only when "based on the results of the total-evidence analysis," and that there is "a great potential for the development of heuristic methods of a posteriori analysis of sets of characters." In this study, we attempt to heuristically examine with statistical analyses the contributions and interactions of data partitions in the resolution of relationships within the family Delphinidae. Our study is the first to use both nuclear and mitochondrial loci for the diagnosis of relationships within the family Delphinidae, and thus provides a unique opportunity to explore the utility of multiple process partitions in addressing the evolutionary history of this group.
Simultaneous analysis and relationships within the Delphinidae
The simultaneous analysis of two nuclear and two mitochondrial genes (a total of 3,053 characters) produced four equally parsimonious trees that differed only in their level of intra-specific resolution (Figure 2). Fourteen of the 15 species-level clades had bootstrap values >99% and high decay indices (5 to 44) (Figure 2). Our results provide unambiguous support for the monophyly of the subfamily Lissodelphininae (node iii, Figure 2) and the polyphyly of the genus Lagenorhynchus (nodes x, ix, and vi, Figure 2), first suggested by LeDuc et al.  and May-Callado and Agnarsson . Several relationships previously ambiguous in LeDuc et al.  were well-resolved in the simultaneous analysis (Figure 2) and included: (1) monophyly of L. albirostris, L. acutus and Stenella/Tursiops/Delphinus clades (nodes xv, xiv, and xiii, Figure 2); (2) increased support for the sister taxon relationship of L. obscurus and L. obliquidens (node vi); and (3) the monophyly L. cruciger and L. australis (node ix). Overall, the simultaneous parsimony analysis either resolved or improved resolution for 12 nodes (Figure 2). The Bayesian analysis produced a topology very similar to that of the parsimony analysis but with increased support at nodes that had low bootstrap values in the simultaneous parsimony analysis (e.g., nodes iv, v, xviii, Figure 2). Most notably, Bayesian analysis supported the monophyly of the Cephalorhynchus (node xviii, Figure 2), and resolved ambiguities in previous studies [13, 14] regarding the relationship of L. australis/L. cruciger to other members of the genus Lagenorhynchus. There were some topological differences between the parsimony and Bayesian trees, but these were restricted to clades outside the Lissodelphininae where taxonomic sampling was relatively poor (Figure 2).
Separate analysis of data partitions produced topologies with variation in the degree of resolution (results not shown). For example, the cyt b partition produced a topology most similar to those of the simultaneous analysis, with 100% consensus among all inter-specific clades. Overall, the mitochondrial partitions had a greater number of phylogenetically informative characters and produced fewer most parsimonious trees than nuclear DNA data. Nevertheless, in all cases the simultaneous analysis cladogram provided considerably greater resolution than any of the independent analyses. In several cases, clades supported by the simultaneous analysis were also recovered in one or more of the independent analyses, with the exception of the control region – many of the relationships in this tree were not recovered in other analyses.
Character dynamics: phylogenetic utility
Each of the four process partitions contributed in some degree to the resolution of the simultaneous analysis tree (Table 2, Figure 3). On a node-by-node basis, the amount of localized topological support varied by dataset and among regions of the simultaneous analysis tree. For example, partitioned Bremer support values for cyt b were consistently positive at all nodes, while values for the control region were positive at the tips of the tree but negative at the base (Table 2, Figure 3). Longer branches had statistically more CI = 1 characters than shorter branches in the tree (H = 7.63, df = 2, P = 0.02), and in all data partitions there was a significant, positive relationship between the frequency of CI = 1 characters and branch length (cyt b, F = 29.5, df = 1, 21, P < 0.01; dloop F = 7.2, df = 1, 21, P < 0.05; Actin, F = 16.1, df = 1, 21, P < 0.01) (Figure 4). The strength of this relationship varied among data partitions, with cyt b having the steepest positive relationship between branch length and CI = 1 characters (Figure 4). Interestingly the dloop, although generally considered to evolve at a faster rate than cyt b, had fewer CI = 1 characters overall, and a slope more similar to that of Actin (Figure 4). Cyt b consistently provided the largest proportion of CI = 1 characters at each node (Table 2, Figure 4). Despite the greater frequency and wider distribution of highly consistent characters in cyt b, the relative proportion of support contributed by each data partition did not differ statistically among categories of branch length (χ2 = 10.42, df = 6, P = 0.12), although Actin tended to contribute more to resolution along longer branches (Table 2). Put another way, each data partition consistently contributed the same relative proportion of the total number of CI = 1 characters among branch length categories, but the absolute number of these characters differed among partitions. Our results further suggest that 3rd position transitions occurred more frequently and provided more topological support than other classes of substitution (χ2 = 12.4, df = 4, P = 0.02).
Character dynamics: data interaction
Spearman's correlation analysis revealed a significantly positive relationship between RAG2 and the other data partitions (ρ2 = 0.432, P = 0.003) but failed to find evidence for interaction between cyt b, Actin, and control region partitions. However, the distribution of positive and negative partitioned Bremer support values within the simultaneous analysis topology suggested localized character conflict among data partitions (Figure 3). The greatest amount of dispersion in partitioned Bremer support was between the control region and other data partitions at the shortest nodes in the simultaneous analysis tree (Table 2), where the partitioned Bremer support for the control region were consistently negative (Figure 3). In addition, the hidden synapomorphy values for the control region were significantly negative (sign test, P = 0.007), with a frequent displacement of synapomorphies with the addition of data in the simultaneous analysis (Table 2). The implications of these findings are two-fold: (1) the control region was responsible for the majority of data conflict, and (2) the relatively weak support for some nodes was a result of this conflict. In contrast, cyt b did not demonstrate a significant pattern of gain or loss of synapomorphies (P = 0.13) due to simultaneous analysis. In fact, cyt b contributed an additional synapomorphy to 6 different nodes following simultaneous analysis (Table 2). The number of synapomorphies contributed by the nuclear DNA data partitions did not change between independent and simultaneous analyses (data not shown), which indicated they were not a source of character conflict.
Character conflict exploration
We explored further the effects of character conflict on the topology by excluding the control region and performing a parsimony and Bayesian analysis with the same parameters as the simultaneous analysis. The topologies from the truncated parsimony and Bayesian analyses were markedly different from each other and from the simultaneous analysis cladogram, and failed in both cases to recover the monophyly of key genera of the Lissodelphininae (Figure 5). In particular, neither analysis produced a monophyletic Cephalorhynchus, which has been well supported in other phylogenetic studies . These results indicated that the control region data contributed to the overall resolution of the simultaneous analysis tree, particularly within the Lissodelphininae, despite evidence for character conflict, and that this was true whether or not the data were analyzed with parsimony or Bayeisan methods. In addition, the removal of the control region data did not produce a monophyletic Lagenorhynchus, which indicated character conflict alone was not responsible for the polyphyly of this genus.
Discussion and conclusion
Lissodelphininae and Lagenorhynchus systematics
Overall, our study demonstrates that the simultaneous analysis of nuclear and mitochondrial process partitions, coupled with better taxonomic sampling, provided greater phylogenetic resolution than cyt b alone [13, 14]. The simultaneous parsimony and Bayesian analyses increased measures of support for 7 nodes, and resolved relationships among 5 clades that were previously ambiguous in the analysis of LeDuc et al.  (Figure 2). Furthermore, our study recovered a monophyletic Cephalorhynchus, and increased posterior probabilities for 4 nodes (i.e., iv, vi, viii, xvii, and xviii Figure 2) compared to the Bayesian tree of May-Callado and Agnarsson . These results are consistent with other studies that suggest phylogenetic analysis based on mitochondrial data alone can be misleading, and that single gene trees are not always accurate representations of species trees. Even in cases where there was apparent conflict among characters, our data suggest synergistic interaction in the simultaneous analysis, and that a priori exclusion of data because of potential conflict may inhibit the recovery of robust phylogenetic hypotheses. For example, the removal of the control region, the putative source of character conflict, produced spurious results with inconsistencies among and within topologies from parsimony and Bayesian analyses (Figures 2 and 5), and these inconsistencies were not trivial. In both the parsimony and Bayesian analysis without the control region, well-supported monophyletic groups were rendered poly- or para-phyletic (e.g., Cephalorhynchus), and deeper-level relationships were extensively altered (Figures 2 and 5). The reliability of Bayesian posterior probabilities as a measure of node support has been recently called into question. [23–26]. Therefore the increased nodal support derived from Bayesian analyses should be interpreted in light of this controversy, and although bootstrap values and posterior probabilities are not directly comparable, a conservative approach might be to consider them as lower and upper bounds of node reliability . Nevertheless, our results support previous studies that suggest data sets interact when analyzed simultaneously [16, 17, 20, 27, 28], and that properties of data emerge that are not detectable when partitions are analyzed separately [15, 16, 20, 27, 28].
Several methods have been proposed to quantify congruence and conflict in simultaneous analyses [15, 17, 20, 29, 30]. Although these measures are useful descriptors of inconsistency among analyses, many do not provide a quantitative measure with which to compare trees derived from independent data partitions. It has been argued (see recent review by Grant and Kluge  and references therein) that an a posteriori analysis of various character partitions against a simultaneous analysis tree is part of a heuristic approach for evaluating the utility of different partitions to resolve relationships among lineages that potentially differ in evolutionary rates and/or overall level of divergence. In this study, we have attempted through a suite of statistical tests (Table 3) to quantitate a posteriori the degree of interaction and phylogenetic utility of data partitions in combined analyses. In many of these tests we chose to use a subset of parsimony informative characters that lacked homoplasy (CI = 1) (Table 3). The logic behind this approach is that in any given topology those characters with a CI of 1 are less likely to be displaced from a node with the addition of future data (i.e., have phylogenetic "inertia") than less consistent characters, and thus represent a conservative class of characters by which to evaluate the utility of data partitions. We acknowledge that even grossly erroneous trees can contain some characters with a CI of 1, and all tree topologies are subject to change with the addition of data – this is true of any working hypothesis. However, we argue that the a posteriori evaluation of the behavior of CI = 1 characters provides one means to quantitatively evaluate the nature of data interaction and utility, and thus identify potential sources of character incompatibility.
For example, our results indicate there is a significant, positive relationship between CI = 1 characters and branch length for cyt b, dloop, and Actin (Figure 4), which is not unexpected given that longer branches are generally expected to have greater numbers of character changes than shorter branches. We also expected that the dloop, all things being equal, would have greater number of CI = 1 characters than cyt b or Actin since the dloop evolves more rapidly than other mitochondrial and nuclear DNA regions in cetaceans (and mammals in general). In contrast, we observe that both the frequency of CI = 1 characters and the strength of the CI = 1 frequency/branch length relationship is greater in cyt b than the dloop and Actin partitions (Table 2, Figure 4). One interpretation of these results is that the evolution of the dloop is so rapid that the phylogenetic information is reduced by multiple substitutions. However, if this were the case, we would expect deterioration of phylogenetic signal with increased branch length – we did not observe this pattern. Instead, we suggest an alternative interpretation – that these results indicate something about the phylogenetic utility and/or the degree of conflict of the dloop with other characters in combined analysis. For example, we have evidence that the dloop is a source of character conflict, and the deviation of the expected results from the linear regression analysis may be an additional indicator of this conflict. Yet, despite this conflict, the significantly positive relationships between branch lengths and CI = 1 characters suggest that all data partitions (to varying degrees) contribute to tree resolution, even in the presence of character conflict. Therefore, this method is a general indicator of the relative ability of each data partition to resolve relationships among OTU's with varying degrees of evolutionary divergence, and provides an example of the emergent properties of characters in combined analysis.
Our statistical tests suggest that the frequency and distribution of CI = 1 characters was not uniform for any data partition, suggesting that the rate of evolution was not constant among lineages (Figure 4). This is not surprising given that heterotachy and covariation are common phenomena in molecular data. However, we were interested to find that, although the frequency of CI = 1 characters varied along lineages, the relative proportion of CI = 1 characters contributed by each data partition was not statistically different (Figure 4). Our interpretation of this finding is that the short branches at the base of the Lissodelphininae were the result of an actual acceleration in rate of evolution along these lineages and not an artefact of the pattern of evolution of any given data partition. In this regard, we suggest this, or similar, statistical tests might be useful in identifying regions in a tree that represent rapid divergence, which are often difficult to resolve with phylogenetic analysis due to the relatively fewer number of informative characters along very short branches. The implications of these results are two-fold. First, the consistent lack of uniformity among data partitions in the number of highly consistent characters provides strong evidence for shifts in the rate of evolution of delphinid lineages since they shared a common ancestor with the Phocoenidae approximately 10 million years ago. Second, the rapid reduction in the number of highly consistent characters in the middle of the simultaneous analysis tree, and the corresponding short branch lengths and large number of lineages indicates that a shift in diversification rate occurred most recently at the base of the subfamily Lissodelphininae (Figure 4). Given that 8 of the 10 species in this monophyletic subfamily are found only in the Southern Hemisphere, this may represent a rapid anti-tropical divergence of lineages south of the equator following equatorial transgression from the north.
Our results support the monophyly of the Lissodelphininae (node iii, Figure 2), and therefore concur with Le Duc et al.  and May-Callado and Agnarsson  that the genus Lagenorhynchus is polyphyletic. Le Duc et al.  proposed that the generic name Lagenorhynchus remain with L. albirostris, the type specimen of the genus , and that the genus Leucopleurus  should be resurrected for L. acutus according to taxonomic precedence. In contrast to Le Duc et al. , the parsimony and Bayesian analyses in this study recovered the monophyly of L. acutus and L. albirostris (Figure 2, node x), and supports a close relationship of these species with the Delphinus/Stenella/Tursiops clade, although the Bayesian and parsimony analysis did not agree on the relationships among these clades (Figure 2). We therefore propose that both L. acutus and L. albirostris retain their current generic designation until further analyses are performed with better taxonomic sampling outside of the Lissodelphininae.
Second, the addition of nuclear and mitochondrial process partitions provided strong support for the paraphyly of Lagenorhynchus within the subfamily Lissodelphininae (Figure 2). In addition, both parsimony and Bayesian analyses support the sister-group relationship of L. obliquidens and L. obscurus and the monophyly of the Cepharlorhynchus, which has been proposed by other molecular phylogenetic studies [12, 13, 22, 32]. This study provides improved phylogenetic resolution for other groups including the monophyly L. cruciger and L. australis, and the placement of this clade within the Lissodelphininae (Figure 2). Our evidence concurs with the results of Le Duc et al.  that L. cruciger/L. australis are monophyletic and should be placed into a separate genus, Sagmatius, first described from Sagmatius amblodon  and later synonymized with L. australis . The monophyletic L. obliquidens and L. obscurus are more problematic and will require the description of a new genus.
Molecular data have helped to resolve some of the problems associated with dolphin taxonomy, but it is apparent that issues related to variation in the rate of evolution within the family continue to make the resolution of some relationships problematic. The next step in this study would be to combine morphology and DNA sequence characters in a simultaneous analysis. For example, Fraser and Purves  found distinct differences among genera of dolphins based on a suite of characters in the sinuses related to the structure and function hearing. A re-evaluation of these characters and additional molecular loci for all members of the Delphinidae might provide resolution of evolutionary relationships for problematic taxa.
Tissues were obtained from biopsy punches and skin swabs from living animals and post-mortem samples from beach-cast or net-caught individuals (Table 1). Skin swabs were collected following the non-invasive procedure of Harlin et al. , preserved in either 90% ETOH or a solution of 20% dimethylsulfoxide (DMSO) saturated with salt, and stored at -20°C. When possible, DNA from the same individuals was used to amplify and sequence all genes. In some cases, pre-existing sequences from GenBank were used to complete the data matrix (Table 1).
Fifty-two individuals from eight genera and 18 species in the family Delphinidae were examined, with emphasis on representation of the genus Lagenorhynchus and other members of the subfamily Lissodelphininae (Table 1). All 12 species of the putative subfamily and members of the genus Lagenorhynchus (i.e., Lagenorhynchus sp., Cephalorhynchus sp., and Lissodelphis sp.) are represented (Table 1). In addition, members of the Tursiops/Delphinus/Stenella clade identified by LeDuc et al. , and the killer whale (Orcinus orca), a taxon thought to be the basal delphinid lineage , were included. Australophocoena dioptrica and Phocoena phocoena from the family Phocoenidae were selected as outgroup taxa because of their sister-taxon relationship to the Delphinidae [37, 38].
DNA isolation and amplification
Total genomic DNA was isolated using either a standard phenol-chloroform protocol  or a Qiagen DNeasy kit (Qiagen, Valencia, California). The polymerase chain reaction (PCR) was used to amplify fragments of four genes including: (1) the complete mitochondrial cytochrome b (cyt b) gene (1040 nucleotides), (2) 474 base pairs (bp) of the mitochondrial control region, (3) 995 bp of nuclear DNA (nDNA) intron I of the muscle Actin gene, and (4) 474 bp of the coding region of the nDNA recombination activating gene 2 (RAG2). External primer sets included: (1) cyt b – 766F (5'-gaaaaaccaycgttgtwattcaact-3') and 766R (5'-gtttaattagaatytyagctttggg-3'); (2) control region – tRNA-Pro and Dlp5 of Baker et al. ; (3) Actin – Lagenorhynchus-specific LagAct1 (5'-gatttggtccctctatgtctct-3' and LagAct2 – 5'-tacttttgaacttgccacctac-3'). Actin primers were designed from published cetacean sequences  and used to amplify the majority of ingroup taxa. Act1  and Act1385H (5'-cttgtgaactgattacagtcc-3') (Palumbi, unpublished) were used to amplify fragments for outgroup taxa and others that failed to amplify with the Lagenorhynchus-specific primers. RAG2 primers were the same as those reported by Murphy et al. . PCR conditions were generally consistent across loci with adjustments made to annealing temperatures. Approximately 1–2 μl of DNA template were included in 50 μl PCR reactions containing the following: 5 μl each of 10× Amplitaq PCR buffer (Perkin Elmer, Boston, Massachusetts), MgCl (25 mM), and deoxynucleotide triphosphates (dNTP's,10 mM), and 1 μl each of bovine serum albumin (BSA, 10 mg/ml), each primer (10 μM), and 1 μl Amplitaq (Perkin Elmer, Boston, Massachusetts) DNA polymerase (5 U/μl). Thermocycler conditions were 94°C for 2 min followed by 35 cycles at 92°C for 30 s, annealing 30 s, and extension at 72°C for 30 s. Published annealing temperatures were used with the following exceptions: 765F/766R, 50°C; LagActin1/2, 58°C; Act1/1385H, 56°C. Amplicons were electrophoresed in 1.5% agarose-TBE (tris, boric acid, EDTA) and visualized under UV light, and prior to sequencing, excess oligonucleotides and dNTP's were removed with either Qiagen (Qiagen, Valencia, California) spin-columns or an Exonuclease I-Shrimp Alkaline Phosphotase (Exo-Sap) enzymatic procedure. Approximately 2 ng of cleaned PCR product per 100 bp of amplicon length was sequenced using ABI BigDye (Applied Biosystems, Foster City, California) cycle sequencing chemistry and an ABI 377 automated sequencer. All amplicons were sequenced in both directions. Internal primer pairs for cyt b (560, 5'-gcaaccctaacacgattcttcg-3'; 610, 5'-ccagtttcgtgtaggaataatagg-3') and Actin (Act5-L, 5'-ccactactttaggcag-3'; M13Act5R-H, 5'-tgtaaaacgacggccagtctgcctaaactagtgg-3' (S. Palumbi, unpublished) were used in sequencing reactions to obtain complete overlap in both directions. Sequences generated in this study are accessioned in the NCBI GenBank database: (1) Dloop: EF092925–EF092969; (2) Cytb: EF093009–EF093055; (3) Actin: EF092970–EF093008; (4) RAG2: EF093056–EF093105.
Sequence alignment and heterozygosity
Sequenced fragments were edited and compiled with the program Sequencher v. 4.1 (Gene Codes Corp., Ann Arbor, Michigan). A consensus of sense and anti-sense strands for each individual and data partition were compiled and exported to MacClade vs. 4.05 . Sequences of the four data partitions were concatenated into a single string of nucleotide characters for the same individual when possible, or a combination of fragments from members of the same species. Cyt b and RAG2 contained no length variable regions, thus alignment of these fragments was trivial. Amino acid translations of the open reading frames of RAG2 and cyt b were examined for stop codons to verify sequence orthology. Actin fragments contained minor genus-specific insertions and deletions that were revealed by alignment in Clustal X  with default parameters. The control region had one (CT)n length variable region of approximately 21 bp that was eliminated from analyses as primary homology could not be reliably assessed. There were an additional 29 sites with indels in the control region, 20 of which were present only in outgroup taxa (A. dioptrica, P. phocoena). Actin had 24 indel sites in three blocks (4, 8, and 12 sites) that were apomorphic (taxon-specific). All species were represented by fragments from the four data partitions with the exceptions: (1) C. eutropia, which lacked Actin and RAG2 fragments because we were not able to obtain tissue samples for this species, and (2) L. cruciger and L. peronii which lacked Actin fragments (Table 1).
All nucleotide ambiguities in Actin and RAG2 that resulted from two different, but equally strong, peaks on electropherograms were considered as evidence for potential heterozygous sites. All such positions were assigned IUPAC ambiguity codes, and considered as ambiguous characters in subsequent phylogenetic analyses. All potential heterozygous sites were species-specific and occurred in only a single individual of each species. PCR reactions consistently produced only one Actin or RAG2 amplicon, which were subjected to a BLAST search to verify sequence identity. In all BLAST searches, amplicons retrieved sequences from either delphinid or mammalian taxa as the closest match, providing further evidence for successful amplification of target loci.
Cladograms were generated with the program PAUP* v.4.0b10  under the optimality criterion of parsimony, with all characters equally weighted. Previous studies suggest that gaps can be phylogenetically informative in both coding and non-coding regions [45, 46], thus all gaps were considered as 5th states in parsimony analyses. The number of taxa and characters precluded the use of exhaustive search options. Therefore, we performed a heuristic search with 1000 random additions of taxa, 100 trees held at each replicate, and tree-bisection-reconnection (TBR) branch-swapping. In addition, phylogenetic analyses were performed separately for mitochondrial (cyt b and control region) and nuclear (Actin and RAG2) data sets with the same heuristic search parameters. Clade support for the simultaneous analysis tree was evaluated with bootstrap  and decay indices . The bootstrap procedure was replicated 10,000 times, each as full heuristic search with random addition of sequences and TBR branch swapping with 100 trees held at each step. Decay indices for the simultaneous analysis tree were derived from a heuristic search of constraint trees created with the program TreeRot v.2b . Heuristic search parameters were the same for the original simultaneous analysis.
There is well-documented variation in the rate of evolution within and among mitochondrial and nuclear genomes; therefore, a Bayesian analysis was performed with the program Mr. Bayes v. 3.0  in order to compare phylogenetic hypotheses derived from parsimony and Bayesian methods. Each data partition was assigned an independent (unlinked) general-time-reversal (GTR) model of substitution with a proportion of invariant sites and variation in substitution rates among sites. Two independent runs of 2 million iterations were performed, each with four chains, three hot, one cold, sampling one tree in 10. Adequate mixing and convergence of chains was examined visually with the program Tracer v.1.1.1 ; posterior probabilities for each clade were derived from trees sampled after the burn in period.
The Bayesian method does not have an option for gaps to be coded as 5th character states, and thus they are treated as missing data in our Bayesian analysis. Some studies have used a separate gap matrix to incorporate indels into Bayesian analyses (e.g., ), and recently there have been advances that allow the explicit treatment of gaps in the simultaneous derivation of sequence alignment and tree topology within parsimony, Bayesian, and likelihood analytical frameworks [48–54]. Such methods are effective particularly in dealing with regions difficult to align (e.g., length variable regions). Since our data set contains relatively few indels, the majority of which are not phylogenetically informative (see above), we opted not to further explore our data with these methods. Such topics deserve thorough treatment beyond the scope of this study.
Character dynamics: phylogenetic utility
The behavior of combining characters from different process partitions was evaluated by examining the relative contribution, or utility, of data partitions to resolving relationships within the simultaneous analysis tree. Branch support indices were partitioned (i.e., "partitioned Bremer support") with the method of Baker and DeSalle  to measure the relative contribution of each data set to node support. The larger the partitioned Bremer support is for a given partition at a particular node, the greater the relative contribution of that partition to the support of that node . We were also interested in the relative phylogenetic utility of nuclear and mitochondrial data per sequenced nucleotide. Therefore, we subsequently standardized the partitioned Bremer values for each node by the number of nucleotides sequenced.
A number of methods have been proposed to quantify a posteriori the contribution of a particular data partition to topological support in a simultaneous analysis . These methods generally include a measure of the relative number of unambiguous character changes attributable to a given data partition [15, 30, 61, 62]. In this study, we chose to quantify at each node in the simultaneous analysis tree a subset of unambiguous character changes that have a consistency index (CI) of 1, that is, they change to one of the four possible states (A, G, C, or T) once only in the tree at a single node. Because these characters lack homoplasy in a given tree, we suggest that these characters are less likely to be affected by data interaction than characters that display homoplasy, and thus are a phylogenetically conservative, and potentially useful, set of characters for a posteriori evaluation of characters.
We use CI = 1 characters to examine two aspects of data partition utility (Table 3). First, we wished to determine if the amount of phylogenetic signal in each data partition decreases over time (i.e., with level of divergence). The idea is that if the number of highly consistent CI = 1 characters that a data partition contributes to a node is constant over time (i.e., there is no decay in phylogenetic signal with increased divergence), we would expect a positive, linear relationship between the number of CI = 1 characters and branch length. This would indicate that a data partition is well suited to address hypotheses of evolutionary relationships at the levels of divergence present in this study. Second, we were interested in whether or not the relative contribution of each data partition to node resolution was consistent among branches of variable lengths (Table 3). Consistent patterns of evolution over time (i.e., the relative frequency and rate of substitutions at particular nodes) among multiple loci can provide evidence that variation in branch lengths among regions of a topology represent the actual evolutionary history of a lineage. In this manner it would be possible to determine if short branches represent periods of rapid or simultaneous taxonomic divergence (i.e., are "hard" polytomies), or are a consequence of some aspect of the data, e.g., incompatibility, or a rate of evolution ill suited to resolve short branches.
To test these hypotheses, we first estimated the number of CI = 1 characters per branch by labelling all unambiguous character changes on the simultaneous analysis and independent analysis trees with their consistency index in MacClade v. 4.01 . Branch lengths (all inter-node distances) were estimated in PAUP* v.4.0b10  by holding the simultaneous analysis tree topology constant and optimizing character state evolution along branches via maximum likelihood with empirical base frequencies, a general-time-reversible (GTR) model, and a gamma shape parameter estimated from the data (n = 8 categories). These estimates of branch lengths and the number of CI = 1 characters per branch were used in subsequent testing of hypotheses regarding phylogenetic utility. To address the question of phylogenetic signal over time, a linear regression was used to examine the relationship between branch length and number of CI = 1 characters along each branch (Table 3). To address the second question of node resolution, we performed two tests (Table 3). First, a chi-square test for homogeneity of proportions was used to determine if CI = 1 characters were evenly distributed within the simultaneous analysis tree with respect to branch length. Second, a Kruskal-Walis non-parametric statistic for multiple samples was used to test if the proportion of CI = 1 characters contributed by each data partition was uniform across branches of different lengths. We pooled data into 4 branch length categories (0–0.0019; 0.0020–0.0049; 0.0050–0.019; 0.0200–0.0500) to increase sample sizes for these tests
Additionally, the relative contribution of each partition to a node was examined by calculating the consistency index of unambiguous character changes at first, second, and third codon positions in the simultaneous analysis tree. A chi-square was used test the null hypothesis of homogeneity of highly consistent characters among codon positions. All statistical tests were performed with SPSS v. 11 statistical package. Null hypotheses were rejected if P ≤ 0.05.
Character dynamics: data interaction
A Spearman's rank correlation of partitioned Bremer support values was used to determine the nature and significance of interaction among data partitions in the simultaneous analysis topology. Partitioned Bremer values for data partitions at a particular node can be used as indicators of conflict among data partitions – positive values indicate support and negative values suggest conflict. A Spearman's rank correlation of partitioned Bremer support values from the simultaneous analysis tree  was performed to measure the level and magnitude of character interaction on a node-by-node basis. A significantly positive correlation coefficient is evidence that partitions support different nodes in similar proportions; a negative correlation indicates significant conflict among data partitions. A non-significant correlation suggests that topological support is not associated with any combined pairs of data partitions . We also calculated the "hidden synapomorphy"  index, defined as the difference in the number of unambiguous character changes at nodes in simultaneous and separate analyses. The number of unambiguous changes for each node was determined with MacClade v.4.01 . A negative hidden synapomorphy index indicated the displacement, or loss, of unambiguous synapomorphies from a node as a result of combining data, whereas positive values suggested synergistic data interaction. The greater the magnitude of the hidden synapomorphy value, the greater the interaction among characters at that node. A sign test, a non-parametric analog of the t-test, was used to determine if the simultaneous analysis had a significant effect on the frequency of synapomorphy displacement.
Würsig B, Cipriano F, Slooten E, Constantine R, Barr K, Yin S: Dusky dolphins (Lagenorhynchus obscurus) of New Zealand: status of present knowledge. Report of the International Whaling Commission. Edited by: Commission IW. 1997, Cambridge, MA , International Whaling Commission, 47: 715-722.
Goodall N: Hourglass dolphin. Molecular Genetics of Marine Mammals. Edited by: Dizon AE, Chivers SJ, Perrin WF. 2002, Cambridge, Massachusetts , International Whaling Commission, 583-585.
Goodall N: Peale's dolphin. Molecular Genetics of Marine Mammals. Edited by: Dizon AE, Chivers SJ, Perrin WF. 2002, Cambridge, Massachusetts , International Whaling Commission, 890-891.
Kinze CC: White-beaked dolphin. The Encyclopedia of Marine Mammals. Edited by: Perrin WF, Wursig B, Thewissen JGM. 2002, New York , Academic Press, 1332-1334.
Flower WH: On the characters and divisions of the family Delphinidae. Proceedings of the Scientific Meeting of the Zoological Society of London. 1883, 467-513.
True FW: A review of the family Delphinidae. Bulletin of the United States National Museum (Smithsonian Institution). 1889, Washington D.C. , Smithsonian Institution, 36: 191-
Gray JE: Sepcilegia Zoologica. 1828, London , Treuttel, Wurtz, and Co.
Gray JE: Supplement to the Catalogue of Seals and Whales in the British Museum. 1871, London , British Museum of Natural History, 103-
Peale TR: United States exploring expedition, during the years 1838-1842 under the command of Charles Wilkes, U.S.N. Mammalia and Ornithology. 1848, 8: 338-
Gray JE: Catalogue of seals and whales in the British Museum. 1866, London , British Museum of Natural History
Cope ED: Third contribution to the history of the Balaenidae and Delphinidae. Proceedings of the National Academy of Sciences, Philadelphia. 1866, 18: 293-300.
Cipriano F: Antitropical distributions and speciation in dolphins of the genus Lagenorhynchus: a preliminary analysis. Molecular Genetics of Marine Mammals. Edited by: A. E. Dizon SJCWFP. 1997, Cambridge, MA , Interntional Whaling Commission, Special Publication 3: 305-316.
LeDuc RG, Perrin WF, Dizon AE: Phylogenetic relationships among the delphinid cetaceans based on full cytochrome b sequences. Marine Mammmal Science. 1999, 15: 619-648. 10.1111/j.1748-7692.1999.tb00833.x.
May-Callado L, Angnarsson I: Cytochrome b and Bayesian inference of whale phylogeny. Mol Phyl Evol. 2006, 38: 344-354. 10.1016/j.ympev.2005.09.019.
Gatesy J, O'Grady P, Baker RH: Corroboration among data sets in simultaneous anlaysis: hidden support for phylogenetic relationships among higher-level artiodactyl taxa. Cladistics. 1999, 15: 271-313. 10.1111/j.1096-0031.1999.tb00268.x.
Baker RH, DeSalle R: Multiple sources of character information and the phylogeny of Hawaiian drosophilids. Systematic Biology. 1997, 46: 654-673. 10.2307/2413499.
Damgaard J, Cognato AI: Sources of character conflict in a clade of water striders (Heteroptera: Gerridae). Cladistics. 2003, 19: 512-526. 10.1111/j.1096-0031.2003.tb00386.x.
Naylor GJ, Brown WM: Amphioxus mitochondrial DNA, chrodate phylogeny, and the limits of inference based on comparisons of sequences. Systematic Biology. 1998, 47 (1): 61-76. 10.1080/106351598261030.
Miyamoto MM, Allard MW, Adkins RM, Janecek LL, Honeycutt RL: A congruence test of reliability using linked mitochondrial DNA sequences. Systematic Biology. 1994, 43 (2): 236-249. 10.2307/2413464.
Cognato AI, Vogler AP: Exploring data interaction and nucleotide alignment in a multiple gene analysis of Ips (Coleoptera: Scolytinae). Systematic Biology. 2001, 50 (6): 758-780. 10.1080/106351501753462803.
Grant T, Kluge AG: Data exploration in phylogenetic inference: scientific, heuristic, or neither. Cladistics. 2003, 19 (5): 379-418. 10.1111/j.1096-0031.2003.tb00311.x.
Pichler FB, Robineau D, Goodall RNP, Meyer MA, Olavarria C, Baker CS: Origin and radiation of Southern Hemisphere coastal dolphins (genus Cephalorhynchus). Mol Ecol. 2001, 10: 2215-2223. 10.1046/j.0962-1083.2001.01360.x.
Douady CJ, Delsuc F, Boucher Y, Doolittle WF, Douzery EJP: Comparison of Bayesian and Maximum Likelihood Bootstrap Measures of Phylogenetic Reliability. Mol Biol Evol %R 101093/molbev/msg042. 2003, 20 (2): 248-254. 10.1093/molbev/msg042.
Erixon P, Svenblad B, Britton T, Oxelman B: Reliability of Bayesian posterior probabilities and bootstrap frequencies in phylogentics. Syst Biol. 2003, 52 (5): 665-673. 10.1080/10635150390235485.
Simmons MP, Pickett KM, Miya M: How Meaningful Are Bayesian Support Values?. Mol Biol Evol %R 101093/molbev/msh014. 2004, 21 (1): 188-199. 10.1093/molbev/msh014.
Suzuki Y, Glazko GV, Nei M: Overcredibility of molecular phylogenies obtained by Bayesian phylogenetics. PNAS. 2002, 99 (25): 16138-16143. 10.1073/pnas.212646199.
Nixon KC, Carpenter JM: On simultaneous analysis. Cladistics. 1996, 12: 221-241. 10.1111/j.1096-0031.1996.tb00010.x.
Kluge AG: A concern for evidence and a phylogenetic hypothesis of relationships among Epicrates (Boidae, Serpentes). Systematic Zoology. 1989, 38: 7-25. 10.2307/2992432.
Wheeler W: Measuring topological congruence by extending character techniques. Cladistics. 1999, 15: 131-135. 10.1111/j.1096-0031.1999.tb00255.x.
Lee MSY, Hugall AF: Partitioned likelihood support and the evaluation of data set conflict. Systematic Biology. 2003, 52 (1): 15-22. 10.1080/10635150390132650.
Hare MP, Cipriano F, Palumbi SR: Genetic evidence on the demography of speciation in allopatric dolphin species. Evolution. 2002, 56 (4): 804-816. 10.1554/0014-3820(2002)056[0804:GEOTDO]2.0.CO;2.
Kellogg R: On the identity of Sagmatius amblodon. Zoological Series of the Field Museum of Natural History. 1941, 27: 293-311.
Fraser FC, Purves PE: Hearing in cetaceans. Bulletin of the British Museum of Natural History, Zoological Series. 1960, London , British Museum of Natural History, 7:
Harlin AD, Würsig B, Baker CS, Markowitz T: Skin swabbing for genetic analysis: application on dusky dolphins (Lagenorhynchus obscurus). Marine Mammal Science. 1999, 15: 409-425. 10.1111/j.1748-7692.1999.tb00810.x.
Barnes LG, Domming DP, Ray CF: Status of studies on fossil marine mammals. Mar Mam Sci. 1985, 1: 15-53. 10.1111/j.1748-7692.1985.tb00530.x.
Barnes LG: Evolution, taxonomy and antitropical distributions of the porpoises (Phocoenidae, Mammalia). Marine Mammmal Science. 1985, 1: 149-165. 10.1111/j.1748-7692.1985.tb00003.x.
Cassens I, Vicario S, Waddell VG, Balchowsky H, Van Belle D, Ding W, Fan C, Mohan RSL, Simoes-Lopes PC, Bastida R, Meyer A, Stanhope MJ, Milinkovitch MC: Independent adaptation to riverine habitats allowed survival of ancient cetacean lineages. Proc Natl Acad Sci U S A. 2000, 97 (21): 11343-11347. 10.1073/pnas.97.21.11343.
Maniatis T, Fritsch EF, Sambrook J: Molecular cloning: A laboratory manual. 1982, Cold Spring Harbor, New York , Cold Spring Harbor Laboratory
Baker CS, Cipriano F, Palumbi SR: Molecular genetic identification of whale and dolphin products from commercial markets in Korea and Japan. Mol Eco. 1996, 5: 671-685.
Palumbi SR, Baker CS: Contrasting population structure from nuclear intron sequences and mtDNA of humpback whales. Mol Biol Evol. 1994, 11: 426-435.
Murphy WJ, Eizirkik E, Johnson WE, Zhang YP, Ryder OA, O'Brien SJ: Molecular phylogenetics and the origin of placental mammals. Nature. 2001, 409: 614-618. 10.1038/35054550.
Maddison WP, Maddison DR: MacClade: Analysis of phylogeny and character evolution. 1992, Sunderland, MA , Sinauer Associates, Inc., 3.0
Thompson JD, Higgins DG, Gibson TJ: CLUSTAL W, improving the sensitivity of progressive multiple sequence alignment through sequence weighting, position-specific gap penalties, and weight matrix choice. Nucleic Acids Research. 1994, 22: 4673-4680.
Swofford DL: PAUP* Phylogenetic Analysis Using Parsimony (*and other methods). 1999, Sunderland, MA , Sinauer and Associates, 4
van Dijk MAM, Paradis E, Catzeflis FM, de Jong WW: The virtues of gaps: xenarthran (Edentate) monophyly supported by a unique deletion in alpha-crystallin. Syst Biol. 1999, 48: 94-106. 10.1080/106351599260463.
Giribet G, Wheeler WC: On gaps. Molecular Phylogenetics and Evolution. 1999, 13 (1): 132-143. 10.1006/mpev.1999.0643.
Felsenstein J: Confidence limits on phylogenies: an approach using the bootstrap. Evolution. 1985, 39: 783-791. 10.2307/2408678.
Bremer K: Branch support and tree stability. Cladistics. 1994, 10: 295-304. 10.1111/j.1096-0031.1994.tb00179.x.
Sorenson MD: TreeRot, version 2. 1999, Massachusetts , Boston University
Huelsenbeck JP, Ronquist F: MRBAYES: Bayesian inference of phylogeny. Bioinformatics. 2001, 17: 754-755. 10.1093/bioinformatics/17.8.754.
Drummond AJ, Rambaut A: TRACER, v.1.1.1. http://evolvezoooxacuk/software, University of Oxford, United Kingdom. 2004
Zaldivar-Riveron A, Mori M, Quicke DLJ: Systematics of the cyclostome subfamilies of braconid parasitic wasps (Hymenoptera : Ichneumonoidea): A simultaneous molecular and morphological Bayesian approach. Molecular Phylogenetics and Evolution. 2006, 38 (1): 130-145. 10.1016/j.ympev.2005.08.006.
Wheeler WC: Optimization alignment: the end of multiple sequence alignment in phylogenetics?. Cladistics. 1996, 12: 1-10. 10.1111/j.1096-0031.1996.tb00189.x.
Wheeler WC: Implied alignment: a synapomorphy-based multiple-sequence alignment method and its use in cladogram search. Cladistics. 2003, 19 (3): 261-268. 10.1111/j.1096-0031.2003.tb00369.x.
Wheeler WC, Gladstein DS, De Laet J: POY: Phylogeny Reconstruction via Direct Optimization of DNA and Other Data, 3.0.11. American Museum of Natural History, New York. 2003
Wheeler WC: Dynamic homology and the likelihood criterion. Cladistics. 2006, 22 (2): 157-170. 10.1111/j.1096-0031.2006.00096.x.
Lunter G, Miklos I, Drummond A, Jensen JL, Hein J: Bayesian coestimation of phylogeny and sequence alignment. Bmc Bioinformatics. 2005, 6:
Lunter G, Miklos I, Drummond A, Jensen JL, Hein J: Algorithms in Bioinformatics, Proceedings. 2003, 2812: 228-244. Bayesian phylogenetic inference under a statistical insertion-deletion model ,Lecture Notes in Bioinformatics,
Redelings BD, Suchard MA: Joint Bayesian estimation of alignment and phylogeny. Syst Biol. 2005, 54 (3): 401-418. 10.1080/10635150590947041.
Lee MSY: Measuring support for phylogenies: the "proportional support index". Cladistics. 1999, 15: 173-176.
Sota T, Vogler AP: Incongruence of mitochondrial and nuclear genes in the carabid beetles Ohmopterus. Syst Biol. 2001, 50: 39-59. 10.1080/106351501750107459.
This study was funded in part with a Doctoral Dissertation Improvement Grant (DDIG 0300480) from the National Science Foundation to ADHC and RLH. Tissue samples were collected in part with funding from the Earthwatch Center for Field Research and the National Geographic Society Committee for Research and Exploration. Very special thanks to T. Markowitz for aid with sample collection. This paper was greatly improved with suggestions from A.I. Cognato, A. Vogler, I. Agnarsson, H. López-Fernández, and 4 anonymous reviewers. We thank M. Meyers and D. Koetze of Sea Fisheries Research Institute, South Africa, S. Dans and K. Crespo of the Laboratorio de Mamíferos Marinos Centro Nacional Patagónico, CONICET, Argentina, and J. Gibbons from the Universidad de Magallanes, Chile for their invaluable support in the field. C. S. Baker provided access to samples of outgroup taxa and continues to be a source of support for this project. F. Pichler provided access to specimens obtained from collaborators N. Goodall, C. Olavarria, and from Te Papa Tongarewa, National Museum of New Zealand. Tissues from free-living dolphins were collected pursuant to the Marine Mammal Protection Act of New Zealand with permission from the New Zealand Department of Conservation, and with approval of Animal Use Protocol committees from the University of Auckland, New Zealand, and Texas A&M University, U.S.A. A portion of the specimens used in this study were provided by the National Marine Mammal Tissue Bank, which is maintained in the National Biomonitoring Specimen Bank at NIST, and which is operated under the direction of NMFS with the collaboration of USGS, USFWS, MMS, and NIST through the Marine Mammal Health and Stranding Response Program. We acknowledge the U.S. Fish and Wildlife Service (USFWS), U.S. Geological Survey (USGS), National Marine Fisheries Service (NMFS), National Institute of Standards and Technology (NIST), the Mineral Management Service (MMS), the National Marine Mammal Tissue Bank (NMMTB), and the specimen collector for use of the banked tissues.
ADHC designed the study, collected tissue samples, carried out the laboratory work, performed all phylogenetic and statistical analyses, and drafted the manuscript. RLH assisted with the study design and multiple drafts of the manuscript. All authors have read and approved the final manuscript.
About this article
Cite this article
Harlin-Cognato, A.D., Honeycutt, R.L. Multi-locus phylogeny of dolphins in the subfamily Lissodelphininae: character synergy improves phylogenetic resolution. BMC Evol Biol 6, 87 (2006). https://doi.org/10.1186/1471-2148-6-87