Background

With approximately 28,872 species, teleost fishes constitute the dominant radiation of vertebrates on our planet [1]. One common explanation for this diversity is that a complete duplication of the entire genome [2] facilitated teleost diversification. This event is also known as the fish specific genome duplication, or FSGD [3]. Many studies have corroborated the occurrence of the genome duplication event [3–7], and several workers have hypothesized that the FSGD enabled the subsequent explosive diversification of teleosts by providing massive opportunities for evolutionary experimentation via gene duplication and decoupling [3–6]. This hypothesis, which we refer to as FSGD-FD (FSGD-facilitated diversification) has never been quantitatively tested.

The fact that teleosts contain over 99% of the total diversity found in ray-finned fishes might be taken as evidence that the diversification rate in teleosts is higher than in their close relatives. However a closer examination of species richness within teleosts reveals that the history of their diversification has been complex. Several teleost orders possess low species richness (Table 1), while the bulk of the biodiversity is concentrated in two large groups: the Ostariophysi, an almost exclusively freshwater clade that includes carps, danios, piranhas, and catfish, and the Perciformes (or perch-like fishes), a group of spiny-rayed fish that includes the majority of coastal and pelagic marine fish as well as some large freshwater lineages like cichlids and perches. Both of these groups appear in the fossil record 150–250 My after the estimated time of the FSGD [3, 7, 8]. One central prediction of the FSGD-FD hypothesis is that the diversification rate accelerated with the origin of teleosts. An alternative hypothesis is that major rate shifts are more recent, and correspond to the appearance of the species-rich teleost subclades described above. If true, this alternative hypothesis would suggest that the FSGD did not play a major role in generating fish biodiversity, due to the long interval between genome duplication and accelerated diversification.

Table 1 Ray-finned fish Species Richness
Full size table

Recently, a number of molecular timescales for ray-finned fishes have been published. However most of these studies sampled a limited number of taxa and used relatively few fossil calibrations [7, 9–11], making it difficult to date the origin of more than a few major actinopterygian crown groups. Here we present a large scale molecular timescale for actinopterygians that allows us to estimate the divergence times of most major lineages as well as the origin of many crown groups within them. We used this timescale along with information about taxonomic richness of unresolved actinopterygian clades to test the hypothesis that teleosts experienced an increase in diversification rates as predicted by the FSGD with a recently developed comparative method, MEDUSA (Modeling Evolutionary Diversification Using Stepwise AIC) ([16]; Additional file 1).

Results

Timetree

We downloaded and aligned 227 vertebrate RAG1 sequences from GenBank (221 actinopterygians, 4 sarcopterygians, 2 elasmobranchs, Additional File 2), and used Bayesian methods to infer divergence times with the ages of 44 clades constrained by fossils (Table 2). Our timetree (Fig 1, 2, 3) is the most comprehensive divergence time study of actinopterygians to date: it includes representatives of 39 of the 44 orders of ray-finned fish and 127 teleost families (which, taken together, represent over 80% of the total teleost species diversity); in addition, many of the 45 fossil calibration points used in this study, identified after a comprehensive review of the actinopterygian fossil record, have never before been integrated in a divergence time analysis (Additional file 3).

Figure 1
figure 1

Timetree of ray-finned fish. Timetree of ray-finned fish based on 227 RAG1 sequences and 45 fossil calibration points. Includes taxa from Polypteriformes to Ostariophysi from Fig. 4.

Full size image
Figure 2
figure 2

Timetree of ray-finned fish. Timetree of ray-finned fish based on 227 RAG1 sequences and 45 fossil calibration points. Includes taxa from Esociformes to part of Percomorpha from Fig. 4.

Full size image
Figure 3
figure 3

Timetree of ray-finned fish. Timetree of ray-finned fish based on 227 RAG1 sequences and 45 fossil calibration points. Includes part of Percomorpha from Fig. 4.

Full size image
Table 2 Priors Used in Divergence Time Analysis
Full size table

Among the Actinopterygii, the crown ray-finned fishes (Fig. 1, node 4) have a mean age of 298 Ma (with a 95% High Posterior Density, HPD: 284 to 337 Ma). This is approximately 100 My older than the oldest fossil, but also ~100 My younger than recent mitogenomic studies [9–11]. The key node in this study is the most recent common ancestor of teleosts (Fig. 1, node 6). We found that teleosts separated from their sister taxon, which in our analysis is a clade formed by gars and the bowfin (from here onwards we refer to this clade as Holostei) about 230 Ma, (95% HPD: 225–243 Ma), and radiated 193 Ma (95% HPD: 173–214 Ma). Our age estimate overlaps with the revised estimate for the FSGD of Hurley et al. [7] (226–316 Ma) but is not congruent with earlier estimates of 300–350 Ma based on less complete sampling [4, 6, 12]. The short fuse of 37 My between the origin of crown neopterygians and the origin of crown teleosts also suggests a relatively brief window of time for the occurrence of the FSGD. Within teleosts, we found that the two largest clades are both Cretaceous in origin. Crown Ostariophysi appeared 128 Ma (95% HPD: 125–134 Ma) (Fig. 1), and the crown Percomorpha (which differs from the Percomorpha of Nelson [13] because it includes also the Atherinomorpha, and includes over 50% of all teleosts) appeared 104 Ma (95% HPD: 93–115 Ma) (Fig. 2).

Our estimated ages for both the origin of the teleosts, as well as for the main splits among their major lineages (Table 3), are much younger than those inferred in mitogenomic studies [7, 11], but are in fairly close agreement with dates provided by time-calibrated nuclear gene divergences in Hurley et al. [7]. This discrepancy might be due to an overall higher rate of evolution in mitochondrial genomes as discussed by Hurley et al. [7]. Within the more derived teleosts, our age estimates are generally younger than those previously published, but are in relatively good agreement with Inuoe et al. [9] for the origin of the acanthomorphs, with largely overlapping 95% HPDs in both studies. The age of two important percomorph clades, the cichlids and the tetraodontiforms, are drastically different between our study and previous work [10, 11] (Fig. 3). The crown cichlids appear to have originated 57 Ma, with the split between the African and Neotropical lineages only 49 My old, dates that are consistent with those inferred by Genner et al. [14]. These ages appear to rule out a major role of the breakup of Gondwana [which dates to the Cretaceous] in determining the present distribution of this group. The Tetraodontiformes, the group that includes the pufferfish, an important model for vertebrate genomics, originated ~65 Ma, towards the end of the Cretaceous. This age is in agreement with the estimate based the previous analysis of a multigene dataset of Alfaro et al. [15], but is almost 100 Ma younger than Yamanoue et al.'s estimate based on mitogenomes [10].

Table 3 Divergence time estimates of focal ray-finned fish nodes
Full size table

Diversification rate study

To test whether diversification rate shifts supports the FSGD-diversification hypothesis, we applied a recently developed comparative method [MEDUSA, Modeling Evolutionary Diversification Using Stepwise AIC ([16], Additional file 1) to a 'diversity tree' derived from both the chronogram and species richness data compiled from the literature [Fig. 4]. Our stepwise procedure, based on the flexible rate model of Rabosky et al. [17], integrates both phylogenetic and taxonomic data [Fig. 4], and involves the assignment of rate shifts [both birth and death rates] to the optimal branches on the phylogeny with unresolved tips until additional rate changes do not substantially improve the AIC score. We tabulated the total species richness of actinopterygians, and partitioned it among representative stem lineages in our phylogeny. We pruned the chronogram down to representatives of each stem lineage with taxonomic information and estimated diversification statistics for the tree under the assumption of rate homogeneity across lineages. Then we tested for rate heterogeneity across lineages by implementing MEDUSA, to identify lineages representing significant departures from an expected background of diversification [16].

Figure 4
figure 4

Diversity tree for analyses of lineage diversification in ray-finned fish. Diversity tree for analyses of lineage diversification in ray-finned fish. Clades from Fig. 1, 2, 3 are collapsed to 27 representative stem lineages and colored by extant species diversity. Clades with unusual diversification rates are denoted with numbers; yellow and blue numbers denote exceptionally fast and slow rates respectively, compared to background rates. Estimates for net diversification rate (r = b-d) and relative extinction rate (e = d/b) are included in the lower right table. Asterisk indicates FSGD event. Abbreviations is figure as follows: Percopsif.: Percopsiformes, Gadiif.: Gadiiformes.

Full size image

Our study reveals that actinopterygian biodiversity has been profoundly shaped by four diversification events. The most statistically significant of these occurred at the base of modern teleosts, as predicted by the FSGD-FD hypothesis, and involved a four-fold increase in net diversification rates [net rate r = birth rate – death rate] over the background rates estimated from the closest evolutionary relatives of teleosts. Additionally we find evidence for secondary rate increases in two lineages. The first of these preceded the radiation of percomorph fishes comprising most of the diversity of acanthomorphs or spiny-rayed fishes [13], including most of the coral reef-associated teleost families as well as most other marine fish diversity. The second increase preceded the radiation of a clade containing most ostariophysans, including the cypriniforms (carps and minnows), characiforms (pirhanas) and siluriforms (catfish). The final rate shift is a deceleration which gave rise to the denticled herring, the sole member of the family Denticipitidae, an ancient lineage that is the sister taxon of the Ostariophysi.

The teleost rate shift is characterized by a 3.7 fold increase in the rate of net diversification. Surprisingly, despite a net increase in diversification rate, estimated extinction rates in teleosts is higher than in nonteleosts (death rate, dteleosts = 6.98 × 10-5, dnonteleosts = 1.21 × 10-5). This contradicts suggestions that genome duplication in teleosts would have contributed to their diversification by making them more resistant to extinction [18]. Instead, turnover (e, the ratio of death to birth rate) is 1.5 times higher in teleosts than in non-teleosts. In comparison, the rate shift that gave rise to the percomorphs was less pronounced with a net diversification rate 1.98 times greater than the teleost rate. The rate shift leading to the ostariophysans reveals a period of increased volatility in the history of actinopterygians. Birth rates increased by more than 2.6 fold over teleost birth rates, but this rise in cladogenesis was checked by a substantial increase in extinction rates. This resulted in turnover rates in ostariophysans that were ~218 times higher than other teleosts. An increase in clade volatility may also have accompanied the increase in the percomorph diversification rate, but we were unable to calculate independent birth and death rates from the net diversification rate due to a lack of phylogenetic resolution within this large clade. In contrast to these three major rate accelerations, the shift underlying the denticled herring was characterized by a ~12 orders of magnitude decrease in net diversification rate. This result provides additional empirical evidence for the unusual nature of ancient clades of small size. These clades are too small and persist too long to be plausible outcomes under typical birth-death models unless the birth and death rates approach 0 [19, 20].

Discussion

Our study provides two lines of evidence in support of the FSGD-FD hypothesis. First, we find a significant increase in the diversification rate of teleosts. Second, the window in time between the split of teleosts and their sister taxon, the Holostei (230 Ma, 95% HPD:225–243 Ma, Fig. 1) and the subsequent radiation of crown teleosts (193 Ma, 95% HPD:173–214 Ma, Fig. 1) overlaps with the estimated age of the genome duplication itself (316-226 Ma) derived from dating of gene paralogs [7]. Furthermore, since teleost diversification is characterized by increases in the birth rate and not by decreases in the death rate (Fig. 4), our study suggests that genome-facilitated mechanisms of divergence, like lineage-specific nonfunctionalization [5, 21], have played a larger role in teleost diversification than extinction resistance imparted by functional redundancy [18].

Our results also caution against the broad interpretation of the FSGD as the primary explanation for extant teleost diversity, of which approximately ~88% derives from the secondary diversification events in the percomophs and ostariophysans. One limitation of MEDUSA is that the assignment of rate shifts is limited to the level of phylogenetic resolution. Thus, the rate shifts leading to the ostariophysans and percomorphs might reflect a series of rate changes within these unresolved groups. As the teleost tree of life is uncovered, it will become possible to more precisely identify subclades or time intervals where diversification rates have changed. However given these caveats, we suggest that the radiation of teleosts is best understood as consisting of at least three pulses. Initial diversification may have been facilitated by mechanisms related to the FSGD [5, 21], though further studies are needed to clarify how genome duplication can lead to sustained, elevated rates of diversification within a clade.

The second pulse (or series of pulses), the diversification within the largely freshwater ostariophysans, occurred about 100 My after the FSGD and coincides with the breakup of the supercontinent Gondwana during the Cretaceous. This geologic upheaval may have created opportunities for ecological diversification by creating new environments. Although many percomorph stem lineages also appear at this time, both the fossil record [8] and this molecular study suggest that the third pulse (or series of pulses) of teleost diversification occurred mostly in the Paleogene (65 to 23 Ma). Possible triggers of increased percomorph diversification include the establishment of scleractinian coral reefs and other tropical shallow water habitats like sea grasses [15, 22], the fragmentation of the marine biotas due to geological events such as the progressive closing of the Tethys sea [23], sea-level fluctuations [24], and the establishments of steeper temperature gradients across the world's oceans, primarily due to the rearrangement of oceanic currents [25].

Conclusion

Whole genome duplications are not uncommon in the tree of life and have been implicated in the diversification of other large clades, including most flowering plants (eudicots) and vertebrates [18, 26]. Without quantitative comparison of diversification rates, however, it is difficult to identify the correlates of an evolutionary radiation, genomic or otherwise. The approach outlined here provides a mean for testing the central predictions of macroevolutionary hypotheses, including those linked to genome duplications, and represents an important step towards identifying the correlates of evolutionary radiations. Theoretical explanations of how genome duplication may lead to speciation have been proposed [27], and a number of examples are now known in which a genome duplication has been shown to have been linked to rounds of cladogenesis in yeasts and flowering plants [28, 29]. However, additional work that elucidates the interplay between genomic isolating mechanisms and ecological opportunity is needed to more completely evaluate the role of genome duplication in shaping patterns of biodiversity.

Methods

Timetree inference

RAG1 sequences for 225 species of bony fish (including three species of lungfish, one species of coelacanths and 221 species of ray-finned fish), and two species of sharks, which we used as outgroups, were downloaded from GenBank (Additional file 2). The sampling was selected in order to both maximize the number of taxonomic groups that we could include in our analysis, and the number of fossil calibration points that could be assigned to the phylogeny. Sequences were aligned automatically using ClustallW [30], and the alignment was then refined by eye using MEGA 4 [31]. A survey of the fossil fish literature allowed us to identify 45 calibration points that were used to date 44 clades identified in the tree as well as the root of the tree (Additional file 3). We used BEAST v 1.4.6 [32] to estimate divergence times under a model of uncorrelated but log-normally distributed rates. We assigned soft upper bounds to the prior distributions of all fossil calibrations using log-normal distributions as described in Table 2. We specified a Yule prior on the rates of cladogenesis. The data set was assumed to have evolved under a GTR model with invariant sites and gamma-distributed rate heterogeneity. We constrained the monophyly of a number of groups in order to reflect generally accepted phylogenetic relationships. Five independent analyses of 20,000,000 generations each were run. Output from each run was analyzed using TRACER 1.4 [32]; 25% of the trees were discarded as burnin, and the remaining were combined using TreeAnnotator 1.4.6 to produce the timescale.

Diversification rate analysis

MEDUSA [16] is an extension of the flexible rate shift model introduced by Rabosky et al. [17]. Rabosky's approach combines two likelihoods. The first is called the phylogenetic likelihood and uses the timing of splits along the resolved backbone of a phylogenetic tree to find maximum likelihood estimates for birth and death rates following equations developed by Nee et al. [33]. The second is called the taxonomic likelihood and uses information about the total species richness of an unresolved tip clade on a phylogeny along with the age of the split between the unresolved clade and its sister group to estimate diversification rates following methods developed by Magallon and Sanderson [34]. Rabosky et al. [17] presented a likelihood ratio test for a model where birth and death rates are allowed to shift on one branch of a phylogeny with unresolved tip clades to a model where birth and death rates are held constant across the tree. MEDUSA extends this procedure by adding rates in a stepwise fashion. First, the AIC score of a model with a single birth and death rate is calculated for the unresolved tree using the combined likelihood estimator presented by Rabosky et al. [17]. This two parameter model is then compared to the best four parameter model (two birth rates and two death rates) where the birth rate and the death rate are allowed to shift on the branch in the unresolved tree that produces the greatest improvement in the likelihood score. If the difference in AIC score between the two and four parameter models is substantial (ΔAIC ≥ 4, [35]) then this rate shift is retained. Next the four parameter model is compared to the best six parameter model by finding the optimal place on the tree for a third rate shift. The process is continued until additional rate shifts no longer produce a substantial improvement in AIC score. Full description of MEDUSA is present in Additional file 1.

To implement MEDUSA with the actinopterygian data, first we assembled taxonomic richness data from FISHBASE [1] for major lineages of fishes. Then we pruned the timetree in Fig. 1, 2, 3 down to 27 representative lineages. Our goal in pruning down the timetree was to preserve as much of the backbone of the timetree as would still permit us to assign species richness unambiguously to tip lineages. Thus, for example, we did not retain splitting events within Percomorpha because, although it was possible to assign species richness to some percomorph subclades such as tetraodontiforms, we could not confidently assign the entire species richness of other percomorphs to lineages included in our sampling. We used this pruned chronogram plus the taxonomic richness to estimate birth and death rates for ray-finned fishes and tested for rate shifts across the tree in R [36] using the LASER [37] and GEIGER [38] packages.