More than half of all vertebrates are ray-finned fishes (Actinopterygii; figure 1). The total number of species is now approaching 27000, the vast majority of which are teleosts (Nelson 2006). Teleosts include numerous ecologically and economically important groups, as well as model organisms used for genomic and developmental biology research. Accurate estimates of actinopterygian divergence dates are therefore vital for addressing fundamental questions about the origin of modern vertebrate biodiversity and for making meaningful interpretations of experimental data. However, theories about the timing and pattern of the early evolution of ray-finned fishes are in a state of disarray and lag far behind our understanding of tetrapod evolution.

Figure 1 Alternative trees and group names of the major living actinopterygian clades: polypterids (bichir and reedfish); chondrosteans (sturgeons and paddlefish); lepisosteids (garpike); amiids (bowfin); and teleosts. (a) Topology from morphological data, supporting (more ...)

The Actinopterygii includes five major clades (figure 1): polypterids (e.g. bichir Polypterus); chondrosteans (e.g. sturgeon Acipenser and paddlefish Polyodon); lepisosteids (e.g. gar Lepisosteus); amiids (e.g. bowfin Amia); and teleosts (e.g. salmon Salmo). Polypterids are placed consistently as the sister group of all other ray-finned fishes, but deep conflicts exist between hypotheses of relationships among the other divisions. Morphology-based phylogenies (figure 1a) unite lepisosteids, amiids and teleosts as the Neopterygii (Regan 1923; Patterson 1973), although the identity of the neopterygian sister group to the Teleostei is uncertain (Gardiner et al. 1996). Most nuclear gene analyses support a neopterygian topology (Le et al. 1993; Lecointre et al. 1993; Hoegg et al. 2004; Kikugawa et al. 2004; Crow et al. 2006), although, again, relationships within the Neopterygii are unstable. However, recent analyses of the nuclear gene RAG-1 (Venkatesh et al. 2001) and complete mitochondrial genome sequences (Inoue et al. 2003) challenge the neopterygian hypothesis. Instead, both include a monophyletic group termed the ‘Ancient Fish Clade’ that unites chondrosteans, lepisosteids and amiids (figure 1b). Phylogenetic analysis of mitochondrial genomes yields strong support for the ‘Ancient Fish Clade’ topology in both Bayesian and maximum-parsimony statistical frameworks, although likelihood ratio tests of these same data are unable to reject neopterygian monophyly (Inoue et al. 2003). Other authors suggest that the ‘Ancient Fish Clade’ topology is an artefact of low taxon sampling (Cavin & Suteethorn 2006). This disagreement about deep actinopterygian phylogeny coincides with fresh interest in ray-finned fish evolution, much of which has been prompted by the discovery and subsequent analysis of an actinopterygian-specific whole-genome duplication event (Amores et al. 1998; Prince et al. 1998; Jaillon et al. 2004; Donoghue & Purnell 2005; Hurley et al. 2005; Crow & Wagner 2006; Mank & Avise 2006).

Predictably, morphology- and molecular-based estimates of the actinopterygian evolutionary time-scale also differ (figure 4). Molecular estimates are expected to exceed morphology-based dates, which are no more than fossil-marked minima, but the incongruity is exceptional. Fossils place the minimum date for the origin of crown group teleosts at 151 million years (Myr) ago (Upper Jurassic; Arratia 2000), whereas recent mitochondrial genome data place the crown teleost date at around 285 and 334Myr ago (Early Permian or Lower Carboniferous depending upon the dataset and the method applied; Inoue et al. 2005). Fossils date the earliest crown-group neopterygian at 245Myr ago (Lower Triassic; Gardiner 1993; Grande & Bemis 1998), while mitochondrial genome estimates place the teleost divergence from amiids at around 417 and 390Myr ago (Late Silurian or Middle Devonian depending upon the dataset and the method applied; Yamanoue et al. 2006). Successive fossil-based divergence estimates for the chondrostean (crown Actinopteri) and polypterid (crown Actinopterygii) total groups date from 345 (Early Carboniferous) and 392Myr ago (Mid-Devonian), respectively (Gardiner 1993; Dineley & Metcalfe 1999). However, comparable mitochondrial estimates do not yet exist, because node dates have been calculated using the ‘Ancient Fish Clade’ tree topology (Inoue et al. 2005). In summary, fossil data present extant actinopterygian diversity as the product of consecutive and widely separated pre-teleost radiations distributed throughout the latter half of the Palaeozoic and Mesozoic eras (figure 4). In contrast, molecular estimates embed all of these events well within the Palaeozoic, with even the teleost crown divergence occurring at least 35Myr before the Mesozoic.

Figure 4 A new time-scale for actinopterygian evolution. A schematic providing comparison between teleost and neopterygian divergence date estimates derived from current and previous studies. Above geological column: previous fossil estimates (blue; Gardiner 1993 (more ...)

Given these extraordinary differences, the time is ripe for a new look at the evolution of ray-finned fishes. The aim of this paper is to examine three independent datasets composed of new and previously published data: (i) morphological material including fossils, (ii) four nuclear gene sequences, and (iii) complete mitochondrial genome sequences. These data are analysed separately, although calibrations for molecular-based node-date estimates are dependent upon fossil markers. We investigate the timing of primary events in basal actinopterygian phylogeny; we search for congruent divergence date estimates derived from these different datasets; and we provide a new estimate of the phylogenetic location and date of the whole-genome duplication.

Figure 2 Strict consensus of the 116 shortest trees from the parsimony analysis of the morphological dataset of fossil and Recent Actinopterygii. Bold lines indicate crown-group Neopterygii. (a) Discoserra pectinodon Lund (2000), from the Mississippian (Serpukhovian) (more ...)

The 33 mitochondrial sequences are from 28 genera, including all non-teleost actinopterygian clades. Novel sequences generated for this study were combined with existing sequences taken from GenBank. The overlap of non-teleost groups between the two datasets is of central importance, because the present analyses focus on the basal divergences within the actinopterygians. Database accession numbers of mitochondrial genomic sequences are given in the electronic supplementary material, table 2.

Analysis of extant taxa alone (6 taxa) was completed using the exhaustive search option. Analysis of the complete taxon set (29 taxa) was undertaken using the heuristic search option with a parsimony ‘ratchet’ (Quicke et al. 2001). Two thousand replicates of random stepwise additions were carried out, followed by TBR (tree bisection–reconnection) branch swapping, holding only one tree in memory at any one time (i.e. MAXTREES=1).

Bayesian phylogenetic analyses were performed on the first and second codon positions of each gene; analyses were implemented using MrBayes v. 3.04b (Huelsenbeck & Ronquist 2001). For each gene, the best-fitting nucleotide substitution model was selected using the Akaike information criterion (AIC) implemented in MrModeltest (v. 1.1b), modified from Modeltest (Posada & Crandall 1998) by J. A. A. Nylander. Flat Dirichlet distributions were used for substitution rates and base frequencies, and default flat prior distributions were used for all other parameters. Metropolis-coupled Markov chain Monte Carlo analyses were run with one cold and three heated chains (temperature set to the default=0.2) for 10 million generations and sampled every 1000 generations. Stationarity was confirmed by examining the plots of −lnL scores and parameter values; 8 million generations were discarded as burn-in. Single gene trees for fzd8, sox11, and tyr were rooted with the tetrapods Homo sapiens, Mus musculus and Xenopus laevis. Hoxa11 was rooted with the shark Heterodontus francisci. Resulting individual gene phylogenies were examined to assign each sequence to its appropriate paralogue group (see electronic supplementary material, figures 5–8).

Species for which two or more genes could be unambiguously assigned to paralogue groups were included in a concatenated dataset of all four genes (2520bp following exclusion of third codon positions). All concatenated sequences were formed by joining sequences from the same paralogue group and species, except for the combination of two different Lepisosteus species. A single paralogue from Oncorhynchus mykiss hoxa11, Xenopus laevis fzd8 and Xenopus laevis sox11 was chosen at random, as these paralogues are the result of more recent, lineage-specific duplication events. The concatenated dataset was divided into two data partitions (first and second codon positions). For each partition, the GTR+I+Γ model of nucleotide substitution was selected using the AIC implemented in MrModeltest (v. 1.1b). Bayesian phylogenetic analysis of the concatenated, partitioned dataset was performed as for the single gene analyses with 7.5 million generations discarded as burn-in. Maximum-likelihood analysis was performed on the concatenated dataset excluding third codon positions using PAUP^* v. 4.0b10 (Swofford 1998). The GTR+I+Γ model of nucleotide substitution was selected using the AIC implemented in Modeltest (Posada & Crandall 1998). Bootstrap proportions for clades were assessed with 100 bootstrap pseudoreplicates. Maximum-likelihood analyses were also performed on concatenated amino acid sequences (1260 amino acids) in Tree-Puzzle v. 5.2 (Schmidt et al. 2002) using the VT (Muller & Vingron 2000) substitution model and Γ-distributed rates. Amino acid frequencies and the Γ shape parameter α were estimated from the data. The Kishino & Hasegawa (1989) and Shimodaira & Hasegawa (1999) tests of an alternative phylogenetic topology including the ‘Ancient Fish Clade’ were performed on the concatenated nucleotide alignment using PAUP^* v. 4.0b10 (Swofford 1998).

The concatenated topology (figure 3) rooted with the tetrapods was used for the Bayesian divergence date estimation. The dataset was partitioned by gene. For each gene, PAML v. 3.14 (Yang 1997) was used to estimate the following model parameters under the F84+Γ model of nucleotide substitution: nucleotide frequencies, transition/transversion parameter κ, and the shape parameter α of the discrete Γ distribution of rates among sites. The maximum likelihood of branch lengths and the variance–covariance matrix of branch length estimates were calculated using the program estbranches (Thorne & Kishino 2002). Divergence times were estimated using the program multidivtime (Thorne & Kishino 2002) and our set of fossil calibrations (see electronic supplementary material). The Markov chain Monte Carlo analyses were run for 10 million generations with a burn-in of 100000 generations. The chain was sampled every 1000 generations. Convergence was verified by running multiple chains for each analysis. The following priors were used: mean time of the root, 3.7 (370Myr ago; s.d.=1.9); mean rate at the root, 0.030 substitutions per site per 100Myr (s.d.=0.015), which is the average mean distance from the in-group root to the tips divided by 370Myr; and mean Brownian motion constant ν, 0.5 (s.d.=0.5). Divergence dates were also estimated using these same parameters on a topology constrained to include a monophyletic Holostei.

Figure 3 Phylogenetic relationships among actinopterygians inferred from partitioned Bayesian analysis of 2520bp of concatenated nucleotide sequences of four nuclear genes (excluding third codon positions). All teleost concatenated sequences belong to (more ...)

Discoserra, dated at 318Myr old, branches from the apex of the neopterygian stem group. Consecutive outgroups to Discoserra plus crown Neopterygii are each confined to the Upper Permian and Lower Triassic. The phylogenetic location of Discoserra therefore implies a series of taxon range extensions for the derived stem-group neopterygians passing from the Palaeozoic–Mesozoic boundary back into the Lower Carboniferous (figure 4).

The nuclear gene analysis also corroborates the hypothesis that the actinopterygian whole-genome duplication occurred in the teleost stem group (Hoegg et al. 2004; Crow et al. 2006). All included teleost sequences belong to either paralogue groups ‘a’ or ‘b’ (figure 3), whereas amiid, lepisosteid and chondrostean sequences fall outside the teleost paralogue groups and are therefore probably pre-duplication orthologues. Hiodon alosoides was excluded from the initial concatenated gene analysis, because paralogue identity could only be assigned to one of the four genes (sox11). However, Hiodon branches from close to the base of the teleost crown group; therefore, it is especially relevant for determining the phylogenetic placement of the whole-genome duplication event. For this reason, the concatenated analysis was repeated including the H. alosoides sox11 data, with the three remaining H. alosoides genes coded as missing data. Results of this analysis also support a teleost-specific whole-genome duplication (see electronic supplementary material, figure 11).

Divergence date estimates derived from mitochondrial genomic data, using a tree topology constrained to include a monophyletic Neopterygii, are shown in figure 4 and listed in the electronic supplementary material, table 7. The 95% posterior credibility intervals for mitochondrial divergence dates are generally older than the divergence date estimates from nuclear gene data, although the two estimates overlap (figure 4). The date estimate for the crown teleost node is between 268 and 326Myr ago, 50–100Myr older than nuclear gene estimates. The date estimate for crown-group Neopterygii is between 327 and 378Myr ago, up to 50Myr older than nuclear gene estimates. The inclusion of Brachydema as a fossil calibration does not alter the overall pattern of taxon range extensions (see electronic supplementary material, table 8). Divergence date estimates using a topology constrained to include a monophyletic Holostei yield similar results (see electronic supplementary material, table 9).

This new minimum date for the crown Neopterygii is a radical departure from the current palaeontological consensus that neopterygian phylogeny concerns Mesozoic to Recent forms (Janvier 1996), but it aligns much more closely with molecular results. Brachydegma falls at the upper limit of the 95% posterior credibility interval (279–371Myr ago; figure 4), resulting from the analysis of nuclear genes using independent fossil calibrations. Mitochondrial genomic divergence date estimates obtained using the same independent calibrations predict even older dates for the Neopterygii (327–378Myr ago; figure 4), although they overlap with the estimates from nuclear gene data. In this respect, as well as identifying consistency between morphological and molecular data, our results highlight the value of applying a combined analysis of independent lines of evidence.

In contrast to the neopterygian result, fossil evidence provides no support for the earlier crown teleost divergence dates suggested by analyses of nuclear (181–265Myr ago) and mitochondrial (268–326Myr ago) genes. An incomplete fossil record could be an explanation, but, in this case, the required date shift corresponds to no obvious trough in the number of taxa documented through time (Sepkoski 2002). However, it is noteworthy that, in North America at least, the number of sedimentary formations hits an all-time low at the Permo–Triassic boundary (251Myr ago; Peters 2006). Alternatively, older dates might result from the artificial inflation of clade ages by analyses of one or both molecular datasets. Furthermore, like the neopterygian result, the mitochondrial divergence date estimate for the Teleostei is much older than the nuclear estimate. Other authors using mitochondrial sequences have obtained similarly anachronistic divergence times for the teleosts (Inoue et al. 2005; Peng et al. 2006; Yamanoue et al. 2006). In turn, these estimates are being used to constrain further molecular clock-based hypotheses of teleost clade divergences (Steinke et al. 2006). Corroboration for these dates can only be gained from vicariance biogeographic scenarios (and should be treated with caution; see Murray 2000), because the dates substantially predate any semblance of palaeontological support. Different selections of calibration markers are bound to underlie some of the resultant, controversial, date estimates, but in the present work the same set of fossil calibrations was used for both nuclear and mitochondrial analyses. The discrepancy between our nuclear and mitochondrial node-date estimates suggests that such estimates of divergence times might be affected by evolutionary rate differences between these two genomes. However, differences in taxonomic sampling and tree topologies, particularly the inclusion of osteoglossomorphs in the mitochondrial dataset, might also be influential.

This research was supported by an award from the Biotechnology and Biological Sciences Research Council to M.G.T., M.I.C. and Z.Y., by NSF grant IBN no. 0091101 to V.E.P. and fellowship DGE-0228235 to M.F., and by NIH grants R01 GM67714 and R01 DK68286 to R.K.H. R.L.M. was supported by NIH grant T32 HL 07381. We thank J. P. Bielawski, D. Jablonski and J. Liston for their comments and discussion, and A. Henrici and D. Berman (Carnegie Museum of Natural History) and C. Schaff (Museum of Comparative Zoology, Harvard) for access to fossil material. Actinopterygian tissue samples were provided by Randall Dahn (University of Chicago), Marcus Davis (University of Chicago), Matt Kourakis (University of California, Santa Barbara), Mark Westneat (Field Museum), Eric Hilton (Field Museum), Leonardo Congiu (Università di Ferrara), Dada Cottelli (Institute of Zoology, London) and Guillaume Lecointre (Muséum National d'Histoire Naturelle, Paris). Additional support was provided by the University of Chicago Biological Sciences Division Faculty Research Fund to M.I.C. and a Natural Environment Research Council grant to Z.Y.