Phylogenetic Relationships of Esocoid Fishes (Teleostei) Based on Partial Cytochrome b and 16S Mitochondrial DNA Sequences
Abstract
The phylogenetic systematics of esocoid fishes are examined using comparisons of partial DNA sequences of the mitochondrial genes coding for the transmembrane protein cytochrome b and the 16S RNA ribosomal subunit. Nucleotide sequences from all species of umbrids, three species of esocids, and salmonid, osmerid, cypriniform, and neoteleostean outgroups were compared to determine patterns of molecular evolution and uncover evidence of phylogenetic relationships. Multiple sequence alignments for each of the two DNA regions examined were used to characterize the amount and type of change shown by the data. The sequences were analyzed under different models of molecular evolution using maximum-parsimony and maximum-likelihood optimality criteria of phylogenetic reconstruction. The phylogenetic analyses revealed previously undiscovered affinities between species of umbrids and esocids that imply the paraphyly of the Umbridae as currently defined. The following set of esocoid interrelationships is proposed: (((Esox, Novumbra), Dallia), Umbra). Esocoid classification is revised based on present findings.
THE suborder Esocoidei contains the families Esocidae and Umbridae, with Esox coextensive with Esocidae, and Dallia, Novumbra, and Umbra comprising the Umbridae. Umbridae and Esocidae are thought to be monophyletic sister groups as currently defined. Generally, the following intergeneric relationships are accepted: (Esox (Novumbra (Dallia, Umbra))). However, there are morphological, karyological, and genetic data that conflict with this hypothesis.
Esocoids are found throughout the Holarctic. Species of Esox have overlapping and abutting ranges, a pattern commonly observed among closely related species. The three species of Umbra are found in the Danube drainage of Europe (U. krameri) and in central and eastern North America (U. limi and U. pygmae, respectively). The three species of Dallia occur in the Siberian Chukotsky peninsula and western Alaska (D. pectoralis; Walters, 1955; Chereshnev and Balushkin, 1980), one locality in northern Chukotskiy Peninsula (D. delicatissima; Nordenskiold, 1882), and in the Amguema River basin in Siberia (D. admirabilis; Balushkin and Chereshnev, 1982). Novumbra is restricted to a few drainages on the Olympic peninsula of Washington State (N. hubbsi; Fitzgerald, 1957; McPhail, 1967; Harris, 1974). The spotted and disjunct distribution of umbrids is characteristic of relictual faunas (Schultz, 1929).
Consensus has not been reached regarding the phylogenetic affinities of the Esocoidei or its taxonomic status. The most generally recognized placement of esocoids is that of Greenwood et al. (1966) who included them (their Esociformes) primitively within the Protacanthopterygii. Later workers recognized esocoids as the primitive sister clade of the Euteleostei (Lauder and Liem, 1983) or as the primitive sister group of the Neoteleostei (Johnson and Patterson, 1996).
Determining esocoid intrarelationships has also been problematic. Cavender (1969) examined umbrid osteology finding evidence to support a close association between Dallia and Novumbra, with this group coordinate with Umbra (Fig. 1A). He also found characters that grouped Umbra and Novumbra to the exclusion of Dallia. On the basis of shared traits of the cephalic sensory canals, Nelson (1972) concluded that Dallia and Umbra form a clade that is the sister group of Novumbra (Fig. 1B). In a study of salmoniform phylogeny, Rosen (1974) found characters of the axial skeleton that supported a Dallia–Novumbra clade as the sister group of Esox, with this assemblage being the sister group of Umbra. Finally, Wilson and Veilleux (1982), in an extensive osteological study of umbrid species, found evidence supporting Nelson's (1972) phylogenetic hypothesis.
Citation: Ichthyology & Herpetology 2000, 2; 10.1643/0045-8511(2000)000[0420:PROEFT]2.0.CO;2
The congruence between the conclusions of Nelson (1972) and Wilson and Veilleux (1982) resulted in the acceptance of their phylogenies for the Umbridae. However, the evidence is ambiguous. Reist (1987) evaluated some of the synapomorphies used to support the accepted classification and found that most characters were reductive in nature or expressed in a mosaic fashion in the taxa in question. Reist (1987) conducted a study of the morphometrics of esocoids using phenetic methodologies to measure overall similarity. He found that Umbra and Novumbra were closer to each other phenetically than to any other esocoid, with Dallia being intermediate between Esox and the Umbra–Novumbra clade (Fig. 1C).
Several works have been published on various aspects of the molecular biology of esocoids (e.g., Ráb, 1981; Ráb and Crossman, 1994). The karyotype and NOR phenotypes of esocoids show great variation (Beamish et al., 1971; Crossman and Ráb, 1996), and show high divergence between Dallia and the other esocoids, with Novumbra and Esox sharing the most similar karyotypes. Kettler et al. (1986) and Kettler and Whitt (1986) examined structural and tissue-expressional variation of lactate dehydrogenase loci in umbrids and found lower divergence between Dallia and Novumbra than either genera showed with Umbra. No comparable data are available for esocids (M. K. Kettler, pers. comm.). Despite the noncladistic approach of the morphometric, karyological, and isozyme studies of esocoids, they show substantial evidence contradicting the currently accepted phylogeny.
Because the distribution of extant esocoid genera represents the relictual remains of a wider distribution in the past, age of divergence between these genera is difficult to estimate. However, the minimum age of esocoid genera can be roughly estimated by the earliest fossil esocid described from the Upper Cretaceous of North America (Wilson et al., 1992). The split between U. krameri and the North American members of the genus Umbra is estimated to have occurred during the Eocene when the deGeer route was broken (Wilson, 1980). The species of Dallia, Umbra of North America, and Esox probably originated more recently as a result of vicariant events associated with the advance and retreat of glaciations (Crossman, 1978).
The recent abundance of DNA sequence data in phylogenetic studies has fueled research into the evolutionary properties of these data and the implications of these properties to phylogenetic inference. One result has been the development of new approaches in a priori data analysis and phylogenetic inference (e.g., Felsenstein, 1985; Hendy and Penny, 1993), which have increased our understanding of the reliability of phylogenetic estimates (Felsenstein, 1988). It is now clear that knowledge of the rates and mode of evolution of the characters from which historical information is being retrieved must be taken into account when designing phylogenetic analyses.
Partial DNA sequences from cytochrome b and 16S rRNA were determined for six species of umbrids, the three species of esocids, and two salmonid outgroups. Cypriniform, osmeriform, and neoteleost sequences were used as alternative outgroups to assess the robustness of the phylogenetic results. Multiple sequence alignments were obtained for each region. Species pairwise comparisons of base substitutions were produced. The aligned sequences were used in parsimony and maximum-likelihood phylogenetic analyses taking into account the evolutionary characteristics of each dataset. The cytochrome b and 16S data presented in this study were treated independently because of the differences in the posttranscriptional fates of their gene products, which may result in differential retention of phylogenetic information by the two genes (Avise, 1994). For the same reason, the sequence data were partitioned according to the biological (structural and positional) properties of each molecule and differences in patterns of change within and between regions of the sequences determined.
Materials and Methods
DNA amplification and sequencing
Genomic DNA was extracted from alchohol-preserved and frozen specimens by standard phenol/chloroform extraction procedures (Palumbi, 1996). Amplifications followed standard PCR protocol.
Cytochrome b amplification reactions were in 25 µL volumes with starting reagent concentrations of 3 mM Mg2+, 0.8 mM dNTPs, 0.4 µM primer B-Gludg-L (5′-TGACCTGAARAACCAYCGTTG-3′; Lento et al., 1994), 0.4 µM primer Cyb 2 (5′- CCCTCAGAATGATATTTGTCCTCA-3′; Lento et al., 1994), 0.75 U Taq polymerase, and 100 ng of total genomic DNA as template. The amplification profile was denaturation at 94 C for 30 sec, annealing at 48 C for 30 sec, and extension at 72 C for 30 sec for 28 cycles. The DNA fragment targeted included a small portion of the transfer RNA for glutamic acid on the 5′ side of the cytochrome b coding gene, and the first 403 nucleotides of the gene. The protein coding nucleotides extended to the third transmembrane helix of cytochrome b, including intermembrane, transmembrane, and matrix domains (Degli Esposti et al., 1993).
16S rRNA amplification reactions were in 25 µL volumes with starting reagent concentrations of 1.5 mM Mg2+, 0.8 mM dNTPs, 0.4 µM primer 16Sar (5′-CGCCTGTTTATCAAAAACAT-3′; Kocher et al., 1989), 0.4 µM primer 16Sbr (5′-CCGGTCTGAACTCAGATCACGT-3′; Kocher et al., 1989), 1 U Taq polymerase, and 100 ng of total genomic DNA as template. The amplification profile was denaturation at 94 C for 30 sec, annealing at 48 C for 30 sec and extension at 72 C for 30 sec for 28 cycles. The 16S fragment targeted ranged from 565 to 570 nucleotides in length corresponding to the 3′ half of the large subunit mitochondrial ribosomal RNA gene.
Sequences were determined by automated sequencing using dye-labeled chain terminators. Sequencing reaction products were analyzed on an ABI 373 automated sequencer following manufacturer's recommendations (Sanger et al., 1977, PE Applied Biosystems).
Analyses
The rate and mode of evolution of a gene dictate the quality of phylogenetic information preserved in the structure of that gene. Because of the very different posttranscriptional fates of the 16S and cytochrome b genes, different characteristics of evolution are expected. To avoid obscuring phylogenetic information contained in either dataset, the sequences from the two genes were treated independently in all the analyses. A combined analysis was performed a posteriori to assess the robustness of the independent analyses.
Cytochrome b data were partitioned by codon position and protein domain to determine internal differences in rate or mode of evolution. The protein domain partitions were derived from the structural model of Degli Esposti et al. (1993). The 16S alignment was obtained through iterations of different gap/match weighting schemes using the Sequence Navigator (ABI) implementation of Clustal, and final manual editing based on the secondary structure of the rRNA. Alignments obtained by five different gap opening/extension weighting schemes were used in phylogenetic analyses to determine the effect of alignment ambiguities on the resulting topologies. The 16S alignment were fitted to Ortí et al.'s (1996) secondary structure model of ostaryophisan 16S rRNA to determine differences in rate and/or mode of change between stem and loop regions of the molecule.
Species pairwise comparisons of divergence and substitution type were generated in PAUP* (D. Swofford, 1999, unpubl.). To assess the extent of substitution saturation in the data, the number of transitions and transversions for all pairwise species comparisons were plotted against species divergence as estimated by percent sequence divergence. This method provides a graphical estimation of saturation (Griffiths, 1997). Information derived from graphical saturation assessments was used to determine weighting schemes in the phylogenetic analyses. Where saturation substitution was observed, differential positional and/or substitutional weighting was applied to reduce homoplastic noise (Fitch and Ye, 1991).
The extent of phylogenetic structure in the data was assessed by character state random permutation tests in the form of PTP tests (Archie, 1989) and g-statistics (Hillis and Huelsenbeck, 1992). These tests were based on a reduced set of species including only one of the species in sibling species pairs and only one of the outgroup taxa to avoid detection of phylogenetic structure relevant only to these clades.
Parsimony and maximum-likelihood analyses were performed in PHYLIP (vers. 3.57c, J. Felsenstein, 1995, unpubl.) and PAUP*. The maximum-parsimony phylogenies were produced through exhaustive searches in PAUP*. Maximum-likelihood analyses invoked the Hasegawa-Kisino-Yano (HKY85) model of molecular evolution allowing for variation in the rate of substitution between sites (Hasegawa et al., 1985). All PHYLIP searches were set to randomized species input order and global rearrangement. Alternative weighting and character exclusion schemes were applied a posteriori based on saturation assessment and initial analysis. The majority-rule consensus tree of 000 delete-half jackknife pseudoreplicates for parsimony and 100 pseudoreplicates for maximum likelihood were used to estimate the amount of support for different clades of interest (Felsenstein, 1985; Wu, 1986). Jackknife indices of support have been shown to behave in similar ways to those resulting from bootstrap pseudoreplicates (Wu, 1986). Finally, split decomposition networks were generated with Splitstree (V.2.04, D. H. Huson, 1997, unpubl.) to visualize all weakly compatible bipartitions of the taxa supported by the sequences and estimate the amount of conflict in the data.
Results
Cytochrome b
An unambiguous multiple sequence alignment of a 420 base fragment from the cytochrome b region of each of the species was obtained. Among the study species sequence divergence ranges from 0.24–26.15% with a mean divergence of 18.40%. Table 1 gives the divergence values for all pairwise comparisons.
The uncorrected sequence divergence between North American species of Umbra and U. krameri (16.71–16.95%) is comparable to the amount of divergence between Esox and Novumbra (15.98–16.46%) and much higher than that between the salmonid genera Oncorhynchus and Salvelinus (10.71%). The Alaskan and Siberian species of Dallia are virtually identical differing in only one transition. The greatest divergence is between U. krameri and Salvelinus. With the exception of comparisons among species of Esox, species of Dallia, and North American species of Umbra, divergence values are above 15%. Divergence between outgroup and esocoid taxa ranges from 18.16–26.15%.
Third codon position substitutions are much more common than changes in first and second positions, with second codon position nucleotides showing the lowest levels of divergence. Transitions at the third codon position appear to have reached the level where multiple hits at the same site may be expected, with some comparisons showing transitions at up to 36% of the third position sites (Fig. 2). All structural partitions show similar patterns of variations in rate and mode of substitution. Transitions accumulate more rapidly than transversions in all partitions (Table 2). These patterns of divergence and the substitution plots were used to devise a weighting scheme to reduce the effect of third codon position transitions in the resulting topology.
Citation: Ichthyology & Herpetology 2000, 2; 10.1643/0045-8511(2000)000[0420:PROEFT]2.0.CO;2
Uniformly weighted parsimony produces two equally parsimonious trees 400 steps long (Fig. 3A; g1 = −0.22; PTP = 0.27). The jackknife tree shows the same topology as Figure 3A, but support for nodes representing intergeneric relationships is weak (29–75%). Weighting transversions over transitions (3:1) at the third codon position and first and second codon position substitutions over third codon position (3:1; g1 = −0.25; PTP = 0.0038) produces a topology in agreement with the maximum-likelihood estimate and similar to the 16S results. The same topology is obtained adding another salmonid to the outgroup; but separate analyses using cypriniform or neoteleost outgroups produce an unresolved polytomy among the four esocoid genera (results not shown). The maximum-likelihood estimate of phylogeny and the jackknife tree share the same topology (Fig. 3A). As in parsimony analysis, intergeneric nodes show low jackknife values (< 64%). The same topology and a slightly higher likelihood are obtained from maximum-likelihood analysis allowing for unequal rates of change among sites. Split graphs derived from the cytochrome b data do not show a clear treelike organization (Fig. 4).
Citation: Ichthyology & Herpetology 2000, 2; 10.1643/0045-8511(2000)000[0420:PROEFT]2.0.CO;2
Citation: Ichthyology & Herpetology 2000, 2; 10.1643/0045-8511(2000)000[0420:PROEFT]2.0.CO;2
Table 3 summarizes the results of parsimony, maximum-likelihood, and jackknife procedures for both cytochrome b and 16S dataset as they affect some previously hypothesized esocoid clades. All parsimony and maximum-likelihood trees agree on the paraphyly of the Umbridae but disagree on other intergeneric relationships.
16S
Pairwise species divergence ranged between 0.37 and 17.88%. The divergences between European and North American species of Umbra are only slightly lower than those between esocoid genera. Divergence among esocoid genera excluding Umbra is much lower than that between Umbra and all other genera in the study. Table 1 gives uncorrected percent sequence divergences for all pairwise comparisons.
The plot of substitutions against percent divergence (Fig. 5) shows a wide range and constant progression in number of substitutions with uncorrected distance, which indicates that substitution saturation may not be prevalent in the sequences; therefore no differential weighting scheme was applied to these data. Plotting stem and loop region substitutions separately failed to show deviation from the overall pattern other than a difference in magnitude (data not shown). Substitutions, including insertion/deletions, are more common in loop regions. All pairwise comparisons involving Hypomesus show the highest levels of divergence and a very different relationship between transitions and transversions than that observed in all other comparisons. The similar numbers of transitions and transversions in Hypomesus comparisons indicate substitution saturation.
Citation: Ichthyology & Herpetology 2000, 2; 10.1643/0045-8511(2000)000[0420:PROEFT]2.0.CO;2
The single most-parsimonious tree (326 steps; g1 = −0.9; PTP = 0.0001; Fig. 3B) has the same topology as the jackknife tree. Jackknife support for individual nodes on this tree is high (> 85%) and low for nodes implied on the currently accepted hypothesis of esocoid relationships (< 3%). The currently accepted hypothesis of esocoid relationships requires 17 extra steps. Esocoid monophyly is not supported, which is likely a result of the extreme divergence and likely homoplasy of the osmerid outgroup. Substituting Oncorhynchus mykiss for Hypomesus, in parsimony analyses results in agreement between the 16S data and the weighted cytochrome b data.
The maximum-likelihood estimate of phylogeny agrees with that of parsimony (Fig. 3B). Allowing for among site rate heterogeneity has no effect on the topology of the tree but does produce a marginal improvement in the likelihood value. The accepted esocoid phylogeny has a significantly worse likelihood (Kishino-Hasegawa-Templeton test, P < 0.05; Kishino and Hasegawa, 1989). The maximum-likelihood jackknife tree shows high indices of support for all intergeneric nodes (>90%).
The topology of the tree inferred from the 16S data is robust to the use of alternative outgroups, including another salmonid, two cypriniforms, and two neoleteosts. It is also robust to the use of alignments derived from alternative gap penalty schemes. Alternative outgroups and alignments only affect the jackknife support values at different nodes but not the topology of the tree. Finally, the same set of esocoid intergenetic relationships is supported by analyses of a combined dataset (Fig. 6), which show high indices of support and resolve the outgroups and esocoid monophyly more satisfactorily than the 16S data alone. Both parsimony and maximum-likelihood jackknife procedures show no support for umbrid monophyly and very low support for the Umbrinae (parsimony only). Table 3 summarizes some results of parsimony and maximum-likelihood analyses. Split graphs derived from the 16S data are more treelike than those derived from cytochrome b data (Fig. 4).
Citation: Ichthyology & Herpetology 2000, 2; 10.1643/0045-8511(2000)000[0420:PROEFT]2.0.CO;2
Discussion
Several workers have investigated esocoid interrelationships with conflicting results. Reist (1987) showed the lack of clear synapomorphic characters to support the accepted hypothesis of esocoid phylogeny. The work presented here uses characters derived from two mitochondrial DNA gene regions to infer the phylogeny of esocoids.
DNA sequences show evolutionary characteristics that make them very useful in phylogenetic analysis at different levels of divergence (Avise, 1994). However, to draw reliable phylogenetic hypotheses from any type of characters, the amount and nature of variation in the data must be thoroughly described to establish the presence of phylogenetic information and to implement an appropriate evolutionary model in the phylogenetic analyses (Swofford et al., 1996). In this study, the evolutionary characteristics of the two molecules examined were determined prior to phylogenetic inference and used to guide parsimony and maximum-likelihood analyses.
Molecular evolution
The most striking difference between the evolution of cytochrome b and 16S among esocoids is the rate of change, as measured by the accumulation of substitutions in the regions sequenced. Cytochrome b evolves at a faster rate than 16S and among esocoids shows extensive substitution saturation at third codon positions. In both genes, change is not distributed homogeneously throughout the molecule. The heterogeneity of rate of evolution within each region can be attributed to functional constraints on the cytochrome b protein and the 16S ribosomal RNA.
The substitution plots show that, among esocoids, cytochrome b sequences have reached the levels of divergence where multiple substitutions will cause underestimates of the true extent of divergence making estimating the absolute rate of change difficult and complicate phylogenetic inference (Fig. 2).
The 16S data show a continually increasing number of substitutions with no indication of saturation. Using the earliest fossil record of an esocid to estimate the divergence between Esox and Novumbra, 16S sequences have diverged at more than 0.1% per million years; however, the rate of change is not uniform throughout the molecule. Clear differences in the patterns of mutation are observed between the salmonid-esocoid comparisons and those involving the osmerid (Fig. 3). In this last comparison, the relationship between transitions and transversions indicates substitution saturation. Therefore, phenetically, salmonids seem to be a much closer to esocoids than to osmerids. The same holds true for all other nonsalmonid outgroups examined, for both 16S and cytochrome b sequences.
Patterns of variation in the evolutionary characteristics of partitions of a gene must be a result of differential biological constrains acting on the gene product (Hillis and Dixon, 1991). Partitioning the variation in cytochrome b comparisons shows that third codon synonymous position changes are the most commonly observed type of substitution, followed by synonymous substitutions in the first codon position (Fig. 5). Mutations at the second codon position were very rare. These observations agree with previous reports on the rate and mode of change of cytochrome b (Griffiths, 1997).
Similarly, variation in the evolutionary characteristics of different regions of 16S can be attributed to the biology of the gene product. The biologically relevant information in ribosomal RNA is encoded in its secondary structure; therefore sequence changes may not disrupt the formation of this structure. In esocoids, nucleotides on the regions corresponding to stems of the rRNA transcript show higher conservation than those on the hairpin loops. As expected, insertions and deletions were restricted to loops where these types of mutations do not disrupt the base pairing necessary for stem formation.
Phylogenetics
The relationships among esocoid fishes have been studied repeatedly without producing a satisfactory hypothesis of relationships. All esocoid genera have very ancient origins, probably predating the Cretaceous-Tertiary boundary (Wilson et al., 1992), as shown by the extent of morphological and cytogenetic variation between the genera. DNA sequences also show high divergence between all esocoid genera.
Cytochrome b and 16S comparisons between esocoid genera show greater uncorrected divergence than comparisons between the two salmonid genera examined. Maximum uncorrected cytochrome b sequence divergence among esocoid fishes was higher than outgroup comparisons, which indicates substitution saturation. In fact, the divergence values for comparisons between umbrid genera are higher than the maximum cytochrome b divergence measured between genera of falconid birds (17%; Griffiths, 1997). OrtÍ et al. (1996) reported 16S sequence divergence among some characoid families at a range comparable to that between esocoid genera. The extent of divergence in mitochondrial DNA regions helps explain the paucity of morphological synapomorphies retained by extant members of esocoid clades that has made delineation of natural esocoid groups difficult (Reist, 1987).
There is some conflicting evidence regarding esocoid interrelationships between the two molecular datasets. The conflict derives from the parsimony estimates of phylogeny based on equally weighted cytochrome b sequences, with all remaining analyses being in agreement. Agreement in the conclusions supported by different sources of data increases our confidence; however, a more defensible approach is to examine the patterns present in the data independent of phylogenetic reconstruction to determine the quality of historical information. Split graphs provide one means of graphically gauging the noise-to-structure ratio present in a character matrix using parsimony-based split decomposition or spectral analysis (Bandelt and Dress, 1992; Hendy and Penny, 1993). Figure 4 shows split graphs based on split decomposition of uncorrected distance matrices from the cytochrome b and 16S sequences. The graphs provide visual indication of the different extent of nonrandom structure present in the sequences. Cytochrome b sequences are not extensively structured in the treelike manner expected from historically produced signal, whereas 16S sequences do show such organization. As mentioned above, there is clear evidence of substitution saturation in cytochrome b, which may account for the lack of resolution provided by those data. The g1-statistic and PTP values for the uniformly weighted cytochrome b dataset failed to detect significant phylogenetic structure in the data. Substitution saturation results in noise that obscures the phylogenetic information preserved in the data, especially in protein coding genes where functional constrains limit the number of states a character may take.
Simulation studies by Fitch and Ye (1991) have demonstrated the ability of weighted parsimony to retrieve the best tree from data in which multiple substitutions affected the phylogenetic signal. When substitutional and positional weighting is applied to the cytochrome b data, parsimony finds the same topology as other analyses of the molecular data, and the character state permutation test detects a significant amount of cladistic structure in the differentially weighted dataset. The conflicting topologies derived from cytochrome b unweighted parsimony may be attributed to widespread homoplasy caused by saturation of transitions at the third codon position.
The topology supported by DNA sequences radically differs from widely accepted hypotheses of esocoid relationships. The monophyly of the Umbridae has rarely been questioned in morphological studies. Studies of esocoid karyotypes and morphometrics have raised questions of umbrid paraphyly; however, how the findings from these studies can be applied to phylogeny is not clear, and umbrids are still considered to be a monophyletic group. Phylogenetic analysis of cytochrome b and 16S DNA sequences offer almost no support for the Umbridae (Fig. 4). The molecular data presented here strongly support a Dallia, Esox, and Novumbra clade to the exclusion of Umbra.
An examination of published data on esocoid morphology in light of the hypothesis of phylogeny supported by the molecular data fails to reveal significant support for the newly delineated clades with some exceptions. Rosen (1974) found modifications of the fourth and fifth epibranchials that support a Dallia, Esox, Novumbra grouping. In addition, these three genera share specialized features of gillraker morphology (pers. obs.).
Alternative outgroups
Similarities between the phylogenetic inference supported by the morphological traits examined by Wilson and Veilleux (1982), and the molecular traits treated in this study make outgroup choice critical to understanding esocoid relationships. Interestingly, the topology of the phylogenetic hypothesis presented here is equal to that of the currently accepted hypotheses, only differing in the placement of the root either with Esox or with Umbra. In other words, a reversal of the character polarities established by Wilson and Veilleux (1982) produces the same tree as that supported by the molecular data.
Further, Wilson and Veilleux's designations of plesiomorphic character states were based on the explicit assumption that Esox represented the primitive sister group to the Umbridae. This assumption is not strongly supported in the literature or by the specialized morphology and feeding habits of esocids. However, some of the character polarities established by Wilson and Veilleux are in agreement with the morphology of other basal euteleosts; therefore the molecular evidence implies that some esocoids characters show opposite evolutionary trends to those seen in other euteleosts (e.g., Nelson, 1972). It seems esocid specialization was accompanied by reversal of morphological traits, which may have misled previous phylogenetic analyses.
We have explored several alternative outgroups to root esocoid relationships. Our outgroup choices were based on previously proposed hypotheses of esocoid affinities. We found overall consistency of results independent of outgroup choice. However, neoteleost and cypriniform outgroups tended to produce unresolved polytomies within the esocoids. Phenetically, sequences from salmonids were more similar to those of esocoids. Parsimony and likelihood analyses using salmonid outgroups resulted in less conflict between the two datasets and higher indices of support for all nodes.
The lack of a clear justification for Wilson and Veilleux's polarization of character state transformations, the equality of the unrooted topologies advanced by their and our studies and the robustness of our hypothesis to alternative outgroups constitute sufficient evidence to alter the accepted hypothesis of esocoid phylogenetics.
The families Esocidae and Umbridae as currently delineated are paraphyletic groups. A natural classification of the group requires reassignment of Dallia and Novumbra to the Esocidae. We propose a rearrangement of the systematics of esociods to reflect the hypotheses of phylogeny forwarded here. Esocoidei should be divided into two families: Esocidae and Umbridae. The family Esocidae should be divided into two subfamilies: Esocinae for Esox and Novumbra, and Dallinae for Dallia. The Umbridae would then be coextensive with the genus Umbra, with the two North American species of Umbra being sister species.
We have not adopted the ordinal rank assigned to the group by Johnson and Patterson (1996) given the lack of clear evidence to support their placement of the group as the basal sister of the neoteleosts. All the names used in the classification have been used previously in the literature with the exception of the subfamily Esocinae, which is erected here to contain Esox and Novumbra.
Conservation
The likely pre-Tertiary origin of esocoid genera supported by fossil and DNA sequence evidence has significant implications for conservation. Vane-Wright (1996) suggests a role for systematics in setting conservation priorities. Species representing remnants of ancient lineages would be afforded greater attention in conservation measures to maximize the character richness being preserved. The ethical justification of this approach is debatable; however, it derives certain validity from the intuitive perception of the unique aesthetic value of “living fossils” (e.g., Latimeria spp.).
Given their very limited distribution, lack of extant diversity, and phylogenetic distinctiveness, as measured by their divergence from other esocoids, Novumbra hubbsi and Umbra krameri should be given special conservation consideration. Although including phylogenetic considerations in conservation decisions remains controversial, it is nonetheless true that N. hubbsi and to a lesser degree the species of Umbra and Dallia are relictual remnants of very ancient lineages that are threatened by human encroachment on their restricted ranges.
Materials examined
Institutional abbreviations are as listed in Leviton et al. (1985). Dallia admirabilis, UW 041670, Ieniveem River drainage, Chukchi Peninsula, Russia; Dallia pectoralis, not accessioned, Wood River, AK, USA; Esox americanus, UAIC 11819.01, Limestone Co. AL, USA; Esox niger, UAIC 10931.03, Bibb Co; UAIC 11890.02, Calhoun Co. AL, USA; Esox lucius, UW 041667, Mondsee, Obevosterreich, Austria; Hypomesus chishimaensis, UW 041862, Kunashir Island, Russia; UW 041869, Zelionyi Island, Russia; Novumbra hubbsi, UW 042434, Ocean Shores, Grays Harbor Co. WA, USA; Novumbra hubbsi, not accessioned, Cherry Cr. Snohomish Co. WA, USA; Salvelinus lecucomaenis, UW 029138, Kunashir Island, Russia; Umbra krameri, UW 041665, Budapest, Hungary; Umbra limi, UW 042432, McGillvary Creek, Manitoba, Canada; Umbra pygmae, CU 77102, Connetquot River, NY, USA. The following sequences were used as alternative outgroups to assess the robustness of the results to rooting artifacts: Cyprinus carpio (GenBank accesion X61010), Crossostoma lacustre (M91245), Oncorhynchus mykiss (L29771), Gadus morhua (X99772), and Fistularia petimba sequences from specimens KU T-1166 and KU T-1228.
Previously proposed hypotheses of esocoid relationships: (A) Cavender, 1969; (B) Nelson, 1972; (C) Reist, 1987
Cytochrome b substitution plots. Number of transitions (∘) and transversions (•) at codon positions: (A) 1, (B) 2, and (C) 3 are plotted against uncorrected percent sequence divergence for all pairwise species comparisons
Estimates of esocoid phylogeny supported by mtDNA sequence data: (A) Most-parsimonious (left and right) and maximum-likelihood (right) trees based on cytochrome b sequences. Both trees are equally parsimonious. (B) Most-parsimonious and maximum-likelihood tree based on 16S sequences. Values above tree edges are jackknife indices of support for maximum parsimony; those below tree edges are for maximum likelihood. * represent jackknife values of 100%. Nodes present on both trees have equal jackknife values
Parsimony based split-decomposition and spectral analysis graphs from Splitstree (vers. 2.4, D. H. Huson, 1997, unpubl.). Split-decomposition graphs for (A) cytochrome b and (B) 16S. Graphs are derived from uncorrected distance matrices (hamming distances)
16S substitution plot. Number of transitions (∘) and transversions (•) are plotted against uncorrected percent sequence divergence for all pairwise species comparisons. The distinct cluster of transition and transversion points at the highest levels of divergence corresponds to pairwise comparisons involving the osmerid outgroup
Consensus of (A) MP trees from 000 jackknife pseudoreplicates and (B) ML trees from 100 jackknife pseudoreplicates of the combined dataset. Values on trees correspond to percent occurrence of the clade among MP or ML trees of jackknife pseudoreplicates of the data matrices. Values above tree edges are derived from analyses on uniformly weighted data. Values below tree edges are derived from analyses where transitions at third codon positions of cytochrome b were weighted one-third relative to all other changes. Only one value is given where differential and uniform weighting yielded the same result