Do habitat shifts drive diversification in teleost fishes? An example from the pufferfishes (Tetraodontidae)

Abstract

Habitat shifts are implicated as the cause of many vertebrate radiations, yet relatively few empirical studies quantify patterns of diversification following colonization of new habitats in fishes. The pufferfishes (family Tetraodon‐tidae) occur in several habitats, including coral reefs and freshwater, which are thought to provide ecological opportunity for adaptive radiation, and thus provide a unique system for testing the hypothesis that shifts to new habitats alter diversification rates. To test this hypothesis, we sequenced eight genes for 96 species of pufferfishes and closely related porcupine fishes, and added 19 species from sequences available in GenBank. We time‐calibrated the molecular phylogeny using three fossils, and performed several comparative analyses to test whether colonization of novel habitats led to shifts in the rate of speciation and body size evolution, central predictions of clades experiencing ecological adaptive radiation. Colonization of freshwater is associated with lower rates of cladogenesis in pufferfishes, although these lineages also exhibit accelerated rates of body size evolution. Increased rates of cladogenesis are associated with transitions to coral reefs, but reef lineages surprisingly exhibit significantly lower rates of body size evolution. These results suggest that ecological opportunity afforded by novel habitats may be limited for pufferfishes due to competition with other species, constraints relating to pufferfish life history and trophic ecology, and other factors.

Introduction

The colonization of novel habitats has long been thought to be one of the main causes of unequal rates of diversification across the tree of life (Simpson, 1944; Schluter, 2000; Alfaro et al., 2007). In the context of adaptive radiation theory, invasion of new habitats may provide lineages with ecological opportunity leading to rapid ecological diversification and accelerated speciation in a new adaptive zone (Simpson, 1944; Schluter, 2000; Burbrink & Pyron, 2010). Although ecological opportunity and adaptive radiation form the theoretical framework for many recent comparative studies (Slater et al., 2010; Frédérich et al., 2013), there are surprisingly few studies that have tested whether the invasion of new habitats provides ecological opportunity in closely related lineages of fishes (but see Betancur et al., 2012). Tropical coral reef ecosystems, which contain astonishingly high levels of marine fish diversity (over 600 species of fishes, 4% of the entire marine fish diversity, can be found on some individual reefs in the Indo‐West Pacific [IWP] [Bellwood et al., 2005; Froese & Pauly, 2012]), have recently been shown to promote diversification in some reef‐associated families (e.g. Alfaro et al., 2007; Cowman & Bellwood, 2011; Price et al., 2011; Frédérich et al., 2013). Colonization of freshwater ecosystems, which occupy only about 2% of the aquatic biome yet contain roughly half of the ~31 000 species of extant actinopterygian (ray‐finned) fishes (Froese & Pauly, 2012), are also thought to provide ecological opportunity for invading fish lineages (Betancur et al., 2012; Carrete Vega & Wiens, 2012). However, these hypotheses have been tested in only a handful of teleost lineages and the extent to which colonization of novel habitats mediates diversification remains unclear.

The family Tetraodontidae (pufferfishes) comprises 187 species in 28 genera, making it the most species‐rich lineage within the order Tetraodontiformes. Perhaps best known for their conspicuous defensive behaviour in which they inflate their body to a volume almost four times normal size (Brainerd, 1984; Wainwright et al., 1995; Wainwright & Turingan, 1997), these fishes are also remarkable for the diversity of habitat types which they inhabit, including pelagic open‐water, tropical reef (Alfaro et al., 2007; Yamanoue et al., 2011), temperate shallow‐water and coastal environments (Froese & Pauly, 2012), and brackish‐ or freshwaters (Dekkers, 1975; Ebert, 2001). This breadth of habitat diversity is matched only by a handful of other teleost families (e.g. gobiids, sciaenids, clupeids) and makes puffers an ideal system to investigate whether the colonization of new habitats provides ecological opportunity that promotes diversification in teleosts.

A major obstacle to the comparative analysis of pufferfish diversification is the lack of a comprehensive phylogeny for the family. Currently, the most densely sampled tree includes less than 20% of the extant diversity (Yamanoue et al., 2011). Here we construct the most complete phylogeny of pufferfishes and allies based on sequence data from two mitochondrial and eight nuclear genes for 115 species. We use multiple fossil‐based constraints to infer divergence times for pufferfishes and use the time‐calibrated phylogeny as a framework to test whether invasions of freshwater and reef habitats have led to increases in the rate of species diversification and phenotypic evolution. We also test whether diversity within the pufferfish genus Takifugu reflects ecological adaptive radiation (Schluter, 2000), as has been recently suggested (Yamanoue et al., 2009).

Materials and methods

Sampling

Tissue samples for 96 species of pufferfishes and allies were obtained through tissue loans from university or museum collections, purchases through the pet trade, and field collections by Michael E. Alfaro (MEA). When possible we sequenced two individual per each species to ensure correct identification of tissue samples. Sequences were downloaded from GenBank for 19 species for which we were unable to secure tissues, bringing the total size of our ingroup to 106 species of Tetraodontidae (56% of the extant diversity, including a relatively homogeneous coverage of all larger genera). In spite of the current uncertainty concerning the higher level relationships of the tetraodontiforms (Winterbottom, 1974; Tyler, 1980; Rosen, 1984; Santini & Tyler, 2003, 2004; Holcroft, 2005; Alfaro et al., 2007; Yamanoue et al., 2008, 2011), the Diodontidae (porcupinefishes or spiny puffers) is widely accepted as the sister group of the Tetraodontidae, and for this reason, we included nine species from three different Diodontidae genera as outgroups.

DNA Extraction, PCR Amplification and Sequencing

Muscle tissue samples or fin clips were stored in 70–90% ethanol prior to use. DNA was extracted using the Qiagen DNeasy kit (Qiagen, Valencia, CA, USA), following the protocol suggested by the manufacturer. We used the polymerase chain reaction (PCR) to amplify two mitochondrial genes, cytochrome oxidase subunit I (from here on abbreviated as cox1) and cytochrome b (Cytb), and six nuclear genes, early growth response gene 1 (EGR1); interphotoreceptor retinoid‐binding protein (IRBP); mixed‐linked Leukaemia‐like gene (MLL); cardiac muscle myosin heavy chain 6 alpha (myh6); recombination activating gene 1 (Rag1) and rhodopsin (Rh). One microlitre of genomic template was used per 25‐μL reaction containing 5 μL of 5 × Go‐Taq Flexi PCR buffer (Promega), 2 μL MgCl₂ (25 mm), 0.5 μL dNTPs (8 μm), 1.25 μL of each primer and 0.125 μL of Promega GoTaq Flexi DNA polymerase (5 U/μL). Amplification of all gene fragments was conducted with an initial denaturing step at 95 °C for 2–4 min; followed by 35–38 cycles with a 0.5‐ to 2‐min 95 °C denaturing stage; a 30‐ to 90‐s 48.5–60 °C annealing stage; and a 2‐min 72 °C extension stage. An additional 7‐min 72 °C extension stage was then followed by a 10 °C cool down. Primers and PCR conditions were obtained from the literature: (Ward et al., 2005) for cox1; (Sevilla et al., 2007) for Cytb; (Chen et al., 2008) for EGR1; (Li et al., 2009) for IRBP and MLL; (Li et al., 2007) for myh6; (López et al., 2004) for Rag1; (Chen et al., 2003) for Rh. PCRs were performed on a MJ Research PTC‐200 Peltier or an Eppendorf Mastercycler ProS thermal cycler. All products were stored at −20 °C after amplification.

We used ExoSap (Amersham Biosciences, Piscataway, NJ, USA) to remove the excess dNTPs and unincorporated primers from the PCR products; purified products were then cycle‐sequenced using the BigDye Terminator v.3.1 cycle sequencing kit (Applied Biosystem, Grand Island, NY, USA) with each gene's original or additional internal primers used for amplification. The cycle sequencing protocol consisted of 25 cycles with a 10‐s 94 °C denaturation, 5‐s of 50 °C annealing and a 4‐min 60 °C extension stage. Sequencing was conducted at the Yale University DNA Analysis Facility using an ABI 3730xl DNA Genetic Analyzer (Applied Biosystems). Sequences were deposited in Genbank with the accession numbers JQ681753–JQ681848 (cox1), JQ681849–JQ681941 (Cytb), JQ681942–JQ682030 (EGR1), JQ682031–JQ682112 (IRBP), JQ682113–JQ682195 (MLL), JQ682196–JQ682286 (myh6), JQ682287–JQ682363 (Rag1); JQ682364–JQ682454 (Rh). Additional sequences were downloaded from Genbank (AP006045, AP006742, AP009525, AP009527, AP009531, AP009534, AP009535, AP009537, AP009540, FJ237596, FJ434549, FJ434559, GQ409967 for cox1; AP006045, AP006742, AP009525, AP009527, AP009531, AP009532, AP00‐9534, AP009535, AP009537, AP009540, EF126090, EF126091, EF126093, EF126094, EF126096, EF126103, EF126104, EF126097, EU274423, EU274424, FJ82‐3446, GQ409967 for Cytb; AF036382, CR649703 for MLL; AY700325, AY700326, AY700355, AY700359, AY700360, AY700361, AY700363 for Rag1; EU637968 for Rh).

Phylogenetic analysis

Chromatograms were checked and assembled into contigs using Geneious (Drummond et al., 2010). The consensus sequences for each individual gene were aligned in Geneious using the ClustalW software (Thomson et al., 1994), and the alignments subsequently checked by eye for accuracy.

The sequences were trimmed to minimize missing characters, and our final data matrix consisted of 677 bp for cox1, 1089 bp for Cytb, 908 bp for EGR1, 817 bp for IRBP, 711 bp for MLL, 789 bp for myh6, 1448 bp for Rag1 and 829 bp for Rh, for a total of 7281 characters used in the concatenated analyses. The final concatenated alignment was deposited in the Dryad repository. The eight individual gene data sets were concatenated in Mesquite 2.7 (Maddison & Maddison, 2010), partitioned by gene, and subject to maximum likelihood (ML) analyses with RAxML (Stamatakis, 2006) and Bayesian analyses using Mrbayes 3.1.2 (Ronquist & Huelsenbeck, 2003).

We used jModelTest (Posada, 2008) to select the most appropriate model of sequence evolution that can be used in MrBayes 3.1.2 (Ronquist & Huelsenbeck, 2003) using AIC (Akaike, 1973). The GTR + G + I model was selected as the most appropriate for cox1, Cytb, EGR1 and Rh, whereas HKY + G + I was selected as the best model for IRBP, MLL, myh6 and Rag1. Due to the fact that the gamma parameter already takes invariant sites into consideration (Yang, 2006), we assigned either GTR + G or HKY + G to the individual gene partitions. The individual gene data sets were subject to ML analyses using RAxML (Stamatakis, 2006) to test for incongruence between the phylogenetic signal of the different loci. The data sets were analysed using the GTR + G model for each gene, implementing the RAxML model closest to the jModelTest results, and 500 fast bootstrap replicates using the GTR + CAT model were run. For the likelihood analyses of the concatenated data set, each partition was assigned its own GTR + G model, and we performed 1000 fast bootstrap replicates using the GTR + CAT model. Each Bayesian analysis of the concatenated matrix was run for 20 million generations, with four chains (one cold, three heated) and sampling every 1000 generations. The trace files were checked in Tracer 1.5 (Drummond & Rambaut, 2007) to ensure that the chains had reached convergence, and the first 25% of trees was discarded as burnin. All remaining trees were then combined to obtain a 50% majority rule consensus tree.

Divergence time analysis

Three fossil calibration points were integrated with the molecular data set to obtain a reliable timescale. The oldest fossil currently assigned to the Tetraodontoidei, or gymnodonts, the clade that includes all puffers is †Triodon antiquus, known from the Early Eocene of the London Clay (~53 Ma) (Tyler & Santini, 2002). This taxon was used to put a prior on the minimum age of the root of our tree, while we used the age of the oldest fossil that we currently believe to be a tetraodontiform, the stem Cretatriacanthus guidottii from the Santonian of Nardo, ~84 Ma, as a soft upper bound on the maximum age (Tyler & Sorbini, 1996; Santini & Tyler, 2003). The oldest known fossils that can be assigned to the clade formed by the tetraodontids and the diodontids are the Middle Eocene (50 Ma) stem pufferfish †Eotetraodon pigmaeus, as well as a number of extinct genera of diodontids from both Italy and the Caucasus (Tyler & Santini, 2002; Santini & Tyler, 2003, 2004; Tyler et al., 2006). †Eotetraodon pigmaeus, which shows a unique mix of derived and primitive features (e.g. it retains pleural ribs) was used to date the split between the Tetraodontidae and its sister group, the Diodontidae. Although in the only phylogenetic studies published to date that include both extant and fossil taxa (Santini & Tyler, 2003, 2004) †Eotetraodon pigmaeus does not always appear to be a stem tetraodontid, the recent discovery of a second species of †Eotetraodon in Late Eocene deposits of the Caucasus (Bannikov & Tyler, 2008) allowed for the re‐interpretation of several putative morphological synapomorphies between †Eotetraodon and the crown tetraodontids, providing additional support for the original claim of the status of †Eotetraodon as a stem pufferfish (Tyler & Santini, 2002). We thus set the minimum age of the split between tetraodontids and diodontids at 50 Ma, and used the age of the oldest crown tetraodontiform, the 59 Myr old †Moclaybalistes danekrus, to set the soft upper bound for the maximum age of this split.

Seven additional extinct species of pufferfishes are known. Six species are currently assigned to the genus †Archaeotetraodon, including the oldest crown fossil †Archaeotetraodon winterbottomi from the Oligocene of Caucausus, ~32 Ma (Tyler & Bannikov, 1994; Santini & Tyler, 2003; Carnevale & Santini, 2006; Carnevale & Tyler, 2010). We thus assigned a minimum age of 32 Ma to the crown tetraodontids, and used the age of †Eotetraodon pigmaeus to set a soft upper bound for the maximum age. An additional fossil taxon has been assigned to the extant genus Sphoeroides. †Sphoeroides hyperostosus, however, is known from a single, highly incomplete specimen characterized by a highly hyperostotic skull from the Pliocene of North Carolina (Tyler et al., 1992), and due to its young age we do not consider it a reliable calibration point.

The concatenated alignment was analysed using a relaxed clock model of uncorrelated, lognormal rates in BEAST 1.6.2 (Drummond & Rambaut, 2007), using the GTR + G or HKY + G models for the individual gene partitions, following the same scheme used for the MrBayes analyses. A birth–death prior was assigned to rates of cladogenesis. For each data set, two analyses of 100 million generations each, with sampling every 10 000 generations and burnin of the first 10% of the trees, were performed. The trace files were checked in Tracer 1.5 (Drummond & Rambaut, 2007) to ensure that the chains had reached convergence and the ESS for all parameters was greater than 200. The remaining trees were then combined in LogCombiner, and a timetree was obtained with the use of TreeAnnotator (Drummond & Rambaut, 2007).

Comparative analyses

All analyses were performed using the software package R version 2.13.0 (R Development Core Team, 2011), using functions in the packages Geiger (Harmon et al., 2008), Laser (Rabosky, 2006), APE (Paradis et al., 2004), Diversitree (FitzJohn, 2009) and Treepar (Stadler, 2011).

Body size (total length, TL, the length from the tip of the snout to the tip of the longer lobe of the caudal fin) and habitat type data (fresh‐, brackish or marine waters; and reef association) were obtained from Fishbase (Froese & Pauly, 2012) and integrated with the published literature (e.g. Dekkers, 1975; Ebert, 2001) (Table S1). Several pufferfish species occur in both freshwater and brackish or marine habitats. We coded these species as freshwater as we were interested in the evolutionary consequences following the evolution of the ability to colonize this novel habitat. We similarly coded as brackish waters species found in both brackish and marine habitats. For 42 species in our sampling Fishbase provided information for standard length (SL, length from tip of the snout to the posterior end of the hypural plate in the caudal fin), but not TL. We measured specimens in various museum collections to determine the ratio between SL and TL for most pufferfish genera, which varies between ~8% and ~20%, and used this to extrapolate TL from SL.

To determine how many times habitat shifts have occurred, we used a maximum likelihood approach to reconstruct ancestral habitats (marine vs. freshwater vs. brackish water; reef vs. nonreef for marine taxa) on the timetree using the MK1 model as implemented in Geiger (Harmon et al., 2008).

If colonization of novel habitat types provides ecological opportunity for pufferfishes, we predicted that diversification rates would be higher for lineages in these habitats. We used the MuSSE model implemented in Diversitree (FitzJohn, 2009) to test whether diversification rates in freshwater and brackish lineages were higher than marine lineages. We compared the fit of a model where the speciation rate was independent of habitat to a model where speciation rates in freshwater, brackish and marine habitats were allowed to vary using AIC scores with a threshold of four as evidence for a substantial improvement in fit of the habitat‐specific model (Burnham & Anderson, 2002) (Table 1). We allowed for separate transition rates between freshwater and brackish and brackish and marine habitats, but did not allow for transitions between freshwater and marine habitats as it is likely that adaptation to brackish habitats represents an intermediate condition between freshwater and marine species. We constrained extinction rates to be the same in the two models since extinction estimates from molecular phylogenies may be unreliable (Rabosky, 2010). To estimate parameter values of the best fitting model, we performed a Markov chain Monte Carlo analysis in Diversitree. We assessed convergence by visual inspection of the log‐likelihoods and using the heidel.diag() function in coda to implement the Heidelberger and Welch convergence diagnostic (Heidelberger & Welch, 1981) We performed a similar analysis to test whether transition to reef habitats provided ecological opportunity by comparing a model where diversification rates were independent of habitat to models that allowed reef taxa to diversify at different rates compared to all nonreef pufferfishes (Table 2). We also examined a model where reef, marine nonreef and freshwater species were assigned separate diversification rates. We allowed separate transition rates between habitats except for transitions between freshwater and coral reefs.

Table 1. Test for state‐dependent diversification of freshwater and brackish species

Model	Parameters	lnL	AIC
Habitat independent	λmarine = λbrackish = λfresh μmarine = μbrackish = μfresh qmarine‐freshwater = qfreshwater‐marine = 0 qmarine‐brackish = qbrackish‐marine qfresh‐brackish = qbrackish‐fresh	−385.41	778.82
Freshwater vs. brackish vs. marine	λmarine ≠ λbrackish ≠ λfresh μmarine = μbrackish = μfresh qmarine‐freshwater = qfreshwater‐marine = 0 qmarine‐brackish = qbrackish‐marine qfresh‐brackish = qbrackish‐fresh	−380.72	773.43

Table 2. Test for state‐dependent diversification of reef‐associated species

Model	Parameters	lnL	AIC
Habitat independent	λreef = λmarine_non_reef = λfresh+brackish μreef = μmarine_non_reef = μfresh+brackish qreef‐marine_non_reef = qmarine_nonreef‐reef qmarine_non_reef‐fresh+brackish = qfresh+brackish‐marine_non_reef qreef‐fresh+brackish = qfresh+brackish‐reef	365.61	739.2
Reef vs. nonreef	λreef ≠ λmarine_non_reef = λfresh+brackish μreef = μmarine_non_reef = μfresh+brackish qreef‐marine_non_reef = qmarine_nonreef‐reef qmarine_non_reef‐fresh+brackish = qfresh+brackish‐marine_non_reef qreef‐fresh+brackish = qfresh+brackish‐reef	365.49	741.0
Reef vs. marine nonreef vs. freshwater + brackish	λreef ≠ λmarine_non_reef ≠ λfresh+brackish μreef = μmarine_non_reef = μfresh+brackish qreef‐marine_non_reef = qmarine_nonreef‐reef qmarine_non_reef‐fresh+brackish = qfresh+brackish‐marine_non_reef qreef‐fresh+brackish = qfresh+brackish‐reef	365.20	742.4

To further examine patterns of diversification in lineages associated with different habitats, we used the method‐of‐moments estimator of diversification rate from Magallon & Sanderson (2001) and MEDUSA (Alfaro et al., 2009). The method‐of‐moments estimator, as implemented in Geiger (Harmon et al., 2008), was used to calculate the global diversification rate of the Tetraodontidae (λ_G), as well as the diversification rate of 14 focal crown subclades classified into three major habitat types. To do this, we assembled crown ages and species diversity for these 14 clades using information from our partitioned timetree and Fishbase under extinction estimates ranging from ε = 0 (pure birth model) to ε = 0.9 (Table S2). We also used MEDUSA, an AIC‐based method that uses phylogenetic and taxonomic richness data to estimate rate shifts on a dated phylogeny with incomplete taxonomic sampling (Alfaro et al., 2009), to test for diversification rate shifts. To do so, we pared down our timetree to a backbone tree where most (180 of 187 species) of the pufferfish diversity was assigned to 18 lineages (Table 3). Seven species could not be assigned to any lineage due to current lack of any information about their phylogenetic relationships. Birth–death models of increasing complexity were used to estimate the rates of speciation and extinction, and an improvement in AIC scores of four units or greater were used as the threshold for retaining rate shifts (Burnham & Anderson, 2002).

Table 3. List of pufferfish stem lineages selected for the MEDUSA analysis, and number of species that were assigned to each lineage (species numbers derived from Fishbase). Richness of SE Asian Tetraodon and of Torquigener species was evenly split among the two lineages present in each genus

Taxon	No of species
Marilyna darwinii	3
Reicheltia halsteadi	1
Tetractenos hamiltoni	2
Torquigener altipinnis	10
Torquigener pleurogramma	10
Tylerius spinosissimus	1
Takifugu ocellatus	25
Lagocephalus inermis	10
Sphoeroides testudineus	25
Pelagocephalus marki	1
Canthigaster valentini	35
Omegophora armilla	2
Chelonodon patoca	3
Tetraodon mbu	6
Arothron diadematus	15
Tetraodon cochinchinensis	9
Xenopterus naritus	5
Carinotetraodon irrubesco	15

To test whether diversification rates were higher during the initial stages of the Takifugu radiation than expected under a constant rates model, we used the MCCR test (Pybus & Harvey, 2000) assuming that the five Takifugu species missing from the tree were randomly drawn. We used the relative cladogenesis statistic (Nee et al., 1992) to test the MCCR assumption of no significant difference in diversification rates between lineages.

Alfaro et al. (2007) suggested that several lineages of tetraodontiforms dated to the Paleocene or Early Eocene, but only started diversifying into the extant species after the end of the Oligocene. To test if our more densely sampled phylogeny (~56% of extant puffers vs. ~15% in Alfaro et al. (2007)) also shows higher rates of cladogenesis during some time interval since the origin of the crown puffer clade, and especially during the Late Oligocene/Early Miocene period, we used Treepar (Stadler, 2011). Treepar (Stadler, 2011) can test for rate shifts (either increases or slowdowns in diversification) across the entire tree during specific time intervals. The timetree was analysed at intervals of 1 or 0.5 Myr, and analyses were run assuming both complete sampling and a 56% sampling of the total diversity to investigate the possible influence of the missing diversity on the results.

Adaptively radiating clades are diversifying along ecological axes (Schluter, 2000), and body size is often used as a proxy for many morphological, physiological and ecological features, such as the type and size of prey that can be consumed by gape‐limited animals such as pufferfishes, or the ability to avoid predators (Peters, 1986; LaBarbera, 1989; Harmon et al., 2010; Slater et al., 2010). If habitat shifts produced ecological opportunity for invading lineages, we predicted that those lineages would retain a signature of elevated rates of body size evolution, reflecting a burst of ecological diversification (Harmon et al., 2010). To compare rates of body size evolution across pufferfishes, we used AUTEUR (Accommodating Uncertainty in Trait Evolution Using RJ‐MCMC), a new method that allows for model‐averaged estimates of rates of continuous character evolution assuming a Brownian motion model and implements Monte Carlo permutation tests (Eastman et al., 2011). We predicted that 1) freshwater species should show accelerated rates of body size evolution compared with other puffers; 2) reef species should have faster rates of body size evolution than nonreef species and 3) the genus Takifugu, which had been singled out as an example of rapid radiation on the basis of their young age and high taxonomic richness (it contains ~15% of all pufferfish species) (Yamanoue et al., 2009), should have higher rates than their nearest relatives. To test these hypotheses we performed Monte Carlo permutation tests with 100 000 randomizations comparing rates of body size evolution in freshwater, brackish, and marine taxa, reef and nonreef species, and rates within Takifugu compared to other closely related species. We used only rate estimates from tip lineages in these tests to minimize the effects of uncertainty in ancestral state assignment on the analysis.

Results

Phylogenetic analyses

Analysis of the eight loci data set with 96 taxa and 115 taxa (including outgroups), with cox1 and/or Cytb sequences for 19 additional pufferfishes, produced identical topologies with regard to higher level relationships. Here, we present results for the 115 taxon data set to maximize taxonomic coverage; results from comparative analyses of the 96 taxon tree were qualitatively similar. Both ML and Bayesian analyses produced highly congruent results (Fig. S1). The monophyly of the tetraodontids is highly supported (bsp 100%, pp > 0.99), as is a basal split between a clade containing the species poor genera Marilyna, Reicheltia, Tetractenos, Tylerius and Polyspina, in addition to the very diverse Torquigener and Takifugu, and a second clade containing most of the taxonomic, ecological and morphological diversity of pufferfishes. Within Clade 1, Torquigener appears to be polyphyletic, with respect to Polyspina and Takifugu. Species within this clade are distributed largely in the IWP and temperate waters in the NW Pacific, and comprise mostly tropical, soft‐bottom marine species (although several species of Torquigener and Takifugu are known to enter brackish waters and at least one species of Takifugu lives in freshwater). The relationships among all the major lineages within this group are highly supported (bsp 100%, pp > 0.98) with the exception of the split between the two lineages of Torquigener and Takifugu, which has bsp < 50% and appears as a polytomy in the Bayesian tree.

One Clade 2 lineage contains a basal split between Lagocephalus, a circumglobal and largely pelagic group, and Sphoeroides, distributed predominantly in temperate waters in the eastern Pacific and the western Atlantic. Deeply nested within Sphoeroides are the two species of Colomesus, including the South American freshwater C. asellus. Support for the Sphoeroides + Colomesus + Lagocephalus clade is weak (bsp = 50%, pp = 0.75). The second major group within Clade 2 is strongly supported and shows Pelagocephalus marki + Canthigaster to be the sister clade to all remaining species. Pelagocephalus is a rare pelagic species found in the southern oceans, spanning from South Africa to New Zealand, while Canthigaster, the largest pufferfish genus with 35 species, is one of the two major reef‐associated genera of pufferfishes. Sister to this group is a clade containing most fresh‐ and brackish water taxa, as well as the second largest clade of reef‐associated puffers, the genus Arothron. Most species within this subclade are currently assigned to the genus Tetraodon, which is shown to be polyphyletic and contains at least three lineages. The first diverging ‘Tetraodon’ group includes the African freshwater puffers, which appear to be closely related to IWP Chelonodon, and with low support to the eastern Indian Ocean Omegophora. The second and third diverging ‘Tetraodon’ lineages are both found in Asia, mostly in the SE region of the continent. One of these ‘Tetraodon’ groups is composed of active swimmers found predominantly in brackish waters, and include popular pet fishes such as the Figure 8 puffer, T. biocellatus, the green puffer T. fluviatilis, and the model organism green‐spotted puffer T. nigroviridis. This group appears to be paraphyletic without the inclusion of Carinotetraodon, which include several species of freshwater fishes, including the smallest known puffer, C. travancoricus (max length ~2 cm). The last ‘Tetraodon’ lineage includes a number of freshwater SE Asian puffers that are characterized by larger size and more sedentary behaviour, including several that are sometimes assigned to the genus Monotrete, such as T. leiurus; these animals tend to sit on soft bottom waiting for prey to approach them. Although the ML analyses are inconclusive as to which are the closest relatives of this third ‘Tetraodon’ lineage, the Bayesian analysis infers a close relationship between these and the clade formed by Auriglobus and Xenopterus. These species are all found in streams in SE Asia, and are characterized by a peculiar laterally compressed body shape unique among the more tubular puffers.

Divergence time analysis

Our relaxed clock analysis inferred an age of 54.9 Myr for the split between the diodontids and the tetraodontids (Fig. 1). The crown of the pufferfishes is inferred to be 36.5 Myr (32.5–43.6 Myr 95% highest posterior density, HPD); the IWP clade is found to be 25.6 Myr old (19.6–32.6 95% HPD), while the second puffer clade is 33.7 Myr old (28.4–40.7 95% HPD). Preliminary BEAST runs sometimes showed the same topology inferred from the RAxML and MrBayes with Lagocephalus sister to Sphoeroides + Colomesus, while in other cases Sphoeroides + Colomesus appeared as sister to the remaining members of this clade. We consider the topology with Lagocephalus sister to Sphoeroides + Colomesus, which was recovered by all analyses in RAxML and Mrbayes, and also by our ongoing study of tetraodontiform interrelationships based on 22 loci (manuscript in preparation) to be the correct one. For this reason, we constrained the Lagocephalus + Sphoeroides + Colomesus clade to be monophyletic to reflect the same relationships found in the previous analyses. The Lagocephalus + Sphoeroides + Colomesus clade appears to be 31.4 Myr old (26.4–38.5 95% HPD); the age for the crown Lagocephalus is 22.4 Myr (17.7–28.4 95% HPD), the crown Sphoeroides + Colomesus is 21.6 Myr (16.2–27.6 95% HPD), and the split between Colomesus and its closely related Sphoeroides is 12.9 Myr (9.4–16.4 95% HPD). The split between Canthigaster + Pelagocephalus is 15.2 Myr (10.8–20.0 95% HPD), while the crown Canthigaster is only 7.4 Myr old (5.5–9.7 95% HPD). The remaining pufferfish subclade originated 20.4 Ma (16.0–25.6 95% HPD); the split between the IWP Chelonodon and the African Tetraodon is 13.2 Myr old (8.9–18.2 95% HPD), with the African clade being 5.4 Myr old (3.6–7.7 95% HPD); that between the Xenopterus + Auriglobus + Carinotetraodon + SE Asian Tetraodon clade is 19.9 Myr old (14.7–23.7 95% HPD).

Figure 1

Timetree of Tetraodontidae produced by the analysis of the concatenated 8‐genes matrix using BEAST 1.6.2. Bars indicate 95% HPD.

Comparative analysis of habitat‐associated diversification

Maximum likelihood reconstructions of the ancestral habitat (Fig. 2) indicate that freshwater invasions have occurred on at least eight occasions (Tetractenos glaber, several Takifugu species; Colomesus; the clade of African Tetraodon, and a large SE Asian clade that includes the remaining Tetraodon, Carinotetraodon, Auriglobus and Xenopterus). Coral reefs appear to have been invaded approximately the same number of times (up to eight), even though most of the diversity is concentrated within only two of these lineages (Arothron and Canthigaster).

Figure 2

a, b: Maximum likelihood reconstruction of ancestral habitat type (marine, freshwater and brackish water), and of reef association onto timetree. Data on habitat type obtained from Fishbase, Dekkers (1975) and Ebert (2001).

We found support for the model where freshwater, brackish and marine species diversified at different rates over a model of habitat‐independent diversification (∆AIC = 5.39, Table 1). However, contrary to our predictions, speciation rates in freshwater and brackish habitats were lower than speciation rates in marine habitats (λ_freshwater = 0.35: 95% HPD: 0.22–0.51; λ_brackish = 0.15: 95% HPD: 0.02–0.28; λ_marine = 0.47: 95% HPD: 0.33–0.63). Allowing for separate rates of speciation on coral reefs did not result in a substantial increase in model fit (Table 2). MEDUSA (Fig. 3) and method‐of‐moments analysis (Fig. S2) did not reveal exceptional diversification for freshwater or brackish lineages although MEDUSA revealed a rate increase in the lineage containing both ‘Torquigener’ lineages plus Tylerius and Takifugu. The method‐of‐moments estimator revealed four crown clades with exceptional diversification, including Torquigener + Tylerius + Takifugu, Takifugu by itself and the two major reef‐associated genera, Arothron and Canthigaster. The MCCR test failed to reject the hypothesis of a constant rate of species diversification for Takifugu (γ = −1.92, ρ = 0.015). Treepar did not detect any significant rate shifts across age intervals.

Figure 3

MEDUSA tree showing the location of the rate shift in the Torquigener + Tylerius + Takifugu + Polyspina clade. For node 1 (Tetraodontidae) r = 0.126, ε = 0.035; for node 2 (Torquigener + Tylerius + Takifugu + Polyspina) r = 0.274, ε = 0.0000001.

Maximum likelihood reconstruction of body size showed increases in length in Takifugu, Lagocephalus, some Sphoeroides, and Arothron, and miniaturization occurring in Carinotetraodon and Canthigaster (Fig. S3). Analysis of body size evolution in AUTEUR identified several young branches (i.e. those leading to Tetraodon mbu and Takifugu niphobles) with elevated probabilities for shifts in the rate of evolution (Fig. S4). Permutation tests did not show a significant difference in rates of size evolution between freshwater puffers and species in other habitats (P = 0.1726; Fig. 4a). Both freshwater and marine species showed higher rates of size evolution than brackish species although this difference was only significant for freshwater vs. brackish species (σ²_freshwater = 5.83: 95% HPD: 0.13–22.86; σ²_brackish =0.22: 95% HPD: 0.02–0.17; σ²_marine = 0.97: 95% HPD: 0.04–4.47; P_{freshwater‐brackish} = 0.0122, P_{marine‐brackish} = 0.0644; Fig. S5). Permutations test comparing reef and nonreef‐associated taxa revealed that, contrary to our predictions, reef species evolved body size at a significantly slower rate than the other marine species (σ²_reef = 0.07: 95% HPD: 0.02–0.09; σ²_{marine nonreef} = 1.27: 95% HPD: 0.04–6.00; P = 0.0104; Fig. 4b) as well as freshwater species (σ²_freshwater = 6.03: 95% HPD: 0.13–23.64, Fig. S5). Permutation tests also revealed a significant increase in the rate of size evolution for Takifugu relative to other members of Clade 1 (P = 0.0212).

Figure 4

Plots illustrating rates of body size evolution with permutation tests results, comparing (a) freshwater to brackish and marine pufferfish species and (b) comparing marine reef to marine non‐reef species.

Discussion

Our analyses produced a major new timeline for the diversification of pufferfishes, revealed novel relationships among several major lineages and provided a robust framework for examining the influence of habitat shifts on pufferfish macroevolution. Contrary to our predictions, we found evidence that colonization of novel habitats by pufferfishes sometimes resulted in lower rates of speciation and body size evolution. We interpret these results in light of recent studies examining the role of competition in mitigating ecological opportunity and consider their implications for expanding our understanding of pufferfish evolutionary history.

Phylogenetic relationships and timescale of pufferfish diversification

Our results are largely congruent with previous studies of tetraodontiform interrelationships based on Rag1, 12s and 16s (Holcroft, 2005; Alfaro et al., 2007), as well as with a recent mitogenomic study (Yamanoue et al., 2011). Both Holcroft (2005) and Alfaro et al. (2007) studies recovered a clade composed of Marilyna, Tetractenos, Torquigener and Takifugu, while Yamanoue et al. (2011) found that Tylerius and Polyspina also belong within this group. Holcroft (2005) and Alfaro et al. (2007) also retrieved a close relationship between Canthigaster and a clade formed by Arothron and the two Tetraodon species in their study. This clade also appears in Yamanoue et al. (2011), which found Pelagocephalus sister taxon to Canthigaster, and a close relationship between Chelonodon and the African Tetraodon lineage. All of these results agree with our study.

One key difference in our study concerns the position of Lagocephalus. Earlier studies recover this taxon as the sister group to other pufferfishes. In contrast, we find that the sister taxon to all remaining puffers is the IWP clade formed by Marilyna, Tetractenos, Tylerius, Polyspina, Torquigener and Takifugu, while Lagocephalus appears nested deeply within the remaining tetraodontids. Yamanoue et al. (2011) also retrieved Sphoeroides as the second pufferfish lineage to branch off after Lagocephalus in their preferred phylogeny, while the Sphoeroides + Colomesus clade is the sister group to Lagocephalus in our trees. Although previously published molecular studies had strong support for the monophyly of the tetraodontids, support for relationships within major lineages was generally weak and sensitive to gene partitioning strategies (Yamanoue et al., 2011). Another difference in our study concerns the position of Carinotetraodon, which Yamanoue et al. (2011) recovered as closely related to the ambush‐feeding SE Asian Tetraodon. In contrast, we find this species to fall within a clade of actively swimming species (Fig. 1), which are morphologically and behaviourally similar to Carinotetraodon.

Our timetree suggests an Early Eocene age for the split between Diodontidae and Tetraodontidae, followed by a long interval before the origin of the crown tetraodontids in the Late Eocene. The stem age for most major groups of pufferfishes dates to the Early Miocene, even though the extant diversity of many of these appears to date to the Late Miocene/Pliocene (e.g. Arothron, Canthigaster, Takifugu, the various ‘Tetraodon’ clades). Most freshwater invasions also appear to have occurred during this time interval. Both Colomesus (due to its distribution we believe C. psittacus to be a secondary marine invader), the African ‘Tetraodon’ clade and the Asian clade of lurking ‘Tetraodon’ have stem ages dating to the Late Miocene, with crown groups diversifying during the Pliocene. Only the largest Asian freshwater clade (including Xenopterus + Auriglobus, Carinotetraodon and the active swimming ‘Tetraodon’ clade) does not appear to follow this trend, with an Early Miocene crown age. These ages are highly congruent with those recovered for other groups found in areas colonized by freshwater puffers (e.g. Lovejoy et al., 1998), suggesting an important role of marine incursions in generating the current pufferfish distribution.

While the dates that we recovered in our study are highly congruent with these found in other studies based on Rag1 or a combination of Rag1 and ribosomal loci (Alfaro et al., 2007; Santini et al., 2009), they are much younger than those found in mitogenomic studies (Yamanoue et al., 2006, 2011), which push the age of the origin of tetraodontids into the Cretaceous. Some of this discrepancy may be due to the tendency of mitochondrial genes to estimate older ages for the deepest splits in a phylogeny (e.g. Hurley et al., 2007). However, there are also significant differences between calibration strategies employed here and in previous studies. The calibrations in this study are based on fossils originating from sites of known geological age and whose phylogenetic placement is supported by studies that incorporated both fossil and extant taxa (Tyler & Santini, 2002; Santini & Tyler, 2003, 2004); this is not the case for some of the calibrations used in the mitogenomic studies. Some calibration points used in one study (Yamanoue et al., 2006) have been assigned ages up to 100 Myr older than the correct ones, an error that results in pushing back the inferred time of origin of the pufferfishes deep into the Mesozoic era (see explanation in (Alfaro et al., 2007). Yamanoue et al. (2011) uses a secondary calibration point of 75 Myr to date the split between Takifugu rubripes and Tetraodon nigroviridis. This is a molecular age inferred in a study of cichlid relationships in which the breakup of Gondwana, instead of the fossil record, was used to date the time of origin of major cichlid lineages (Azuma et al., 2008). Calibration with this biogeographical constraint produces very old ages for deep splits within the percomorphs represented in their tree that are significantly at odds with both the fossil record for these fishes and with other time‐calibration studies (Genner et al., 2007).

Diversification rates and habitat evolution

Our study confirms previous findings that several ancestrally marine pufferfish lineages have independently invaded freshwater habitats (Yamanoue et al., 2011), with most invasions having occurred during the Early to Middle Miocene (Fig. 2a). These ages also coincide with the repeated invasion of reefs by several different pufferfish lineages (Fig. 2b). There appears, however, to have been a significant time‐lag between these invasions and the origin of the extant diversity, which may have been due to either high rates of extinction during the Mid Miocene or to delayed diversification.

We hypothesized that colonization of novel habitats has provided pufferfishes with ecological opportunity and predicted that transitions to freshwater habitats and coral reefs would be characterized by increased rates of speciation and morphological evolution. Our results reveal that the evolutionary history following these colonizations of novel habitat types has been more complicated. Pufferfish lineages that have colonized freshwater have lower rates of speciation compared to marine species and show rates of morphological evolution that are not significantly higher than species in marine habitats (Figs 4 and S5).

The observation of lower rates of speciation and unexceptional morphological evolution in freshwater species may indicate that competition with the established freshwater fish fauna has resulted in diminished ecological opportunity for pufferfishes invading South America, Africa and Asia. Betancur et al. (2012) report a similar pattern of unexceptional speciation following colonization of freshwater habitats in ariid catfishes except in Australia/New Guinea, which exhibits a depauperate freshwater fish fauna. Like the ariids, the pufferfish colonizations of freshwater habitats have occurred largely over the last 20 Myr into environments with diverse and well‐established freshwater fish communities. Our results add to a growing number of studies that indicate that competitors play an important role in determining ecological opportunity for colonizing species (de Queiroz, 2002; Gavrilets & Losos, 2009; Losos, 2010; Mahler et al., 2010) even in habitats like freshwaters (Heger et al., 2010; Logares et al., 2010; Hou et al., 2011; Betancur et al., 2012; Bloom & Lovejoy, 2012), which have been major arenas of diversification for many fishes (Carrete Vega & Wiens, 2012).

It is possible that behavioural and trophic traits that are characteristic of pufferfishes have also played a role in limiting their potential to diversify in freshwaters. Although puffers well‐tolerate conditions of low salinity, they lack life history traits as elaborate parental care that have been shown to promote diversification in many successful freshwater lineages (Seehausen, 2006). Puffer trophic morphology may be evolutionarily constrained to retain a durophagous feeding style and morphological conservativeness of their trophic anatomy (as opposed to the high degree of evolvability of pharyngeal structures in groups such as cichlids) (Seehausen, 2006) may be better suited to coastal marine ecosystems than to freshwater ones.

Colonization of reef habitats has been linked to accelerated diversification in several fish lineages (Alfaro et al., 2007; Cowman & Bellwood, 2011; Price et al., 2011; Frédérich et al., 2013), but we find only marginal evidence that reef colonization had a marked influence on pufferfish macroevolution. Although the two major crown reef lineages (Arothron and Canthigaster) have higher diversification rates than expected given their age and the average rate of diversification in pufferfishes (Fig. S2), a habitat dependent model of speciation does not fit our data substantially better than a habitat‐independent model (Table 2). Furthermore, rates of body size evolution in reef‐associated puffers are lower than rates in nonreef species including both marine‐restricted and freshwater species (Figs 4 and S5). Taken together, these results do not support the classical model of ecological adaptive radiation of pufferfishes, wherein phenotypic evolution and lineage diversification increase following colonization of a novel habitat. It is possible that processes other than ecological diversification have played a dominant role in the evolution of the two reef‐associated genera. Canthigaster species have been noted for exhibiting remarkable similarity in size (Tyler, 1967) and body proportions, and are largely classified on the basis of colour pattern (Allen & Randall, 1977; Randall et al., 2008). Due to their high toxicity to other marine fishes (Allen & Randall, 1977), it is tempting to speculate that ecological phenomena such as Mullerian mimicry might be important in constraining the evolution of body size diversity in these fishes (e.g. Alexandrou et al., 2011).

The relatively large number of species within the genus Takifugu along with their young age has led some authors to suggest that they are experiencing explosive speciation (Yamanoue et al., 2009). Our macroevolutionary analyses provide the first quantitative support for this hypothesis. The crown clade Takifugu is more species rich than expected given its age (Fig. S2) and exhibits significantly faster rate of body size evolution than other members of Clade 1. The cause or causes of this radiation remain unclear. An increase in the rate of body size evolution could signal a burst of phenotypic and ecological differentiation that is consistent with ecological adaptive radiation. However, if this is ecological adaptive radiation, the source of ecological opportunity is not readily identifiable. Takifugu do inhabit diverse environments including both subtropical and temperate waters across a very broad latitudinal spectrum spanning southern Australia to Japan across the Eastern Indian Ocean and the Western Pacific, but there are as yet no known traits possessed by this clade that explain their explosive diversification relative to other pufferfishes over the last 2–3 Myr. A possible alternative explanation is that diversification in this very young clade has been driven by Pleistocene glaciation cycles (Palumbi, 1994).

Conclusion

Our study reveals that despite an ability to colonize diverse habitats that have driven adaptive radiation in other species, pufferfishes do not explosively diversify once they reach them. In the case of freshwater habitats, a slowdown in the rate of cladogenesis could indicate that ecological opportunity is limited, either due to competition from freshwater species in these habitats, intrinsic behavioural and trophic constraints related to their life history as primarily durophagous species, or both. Rapid diversification of the two largest reef‐associated pufferfish genera suggests a role for reefs in driving pufferfish diversity. However, a significant deceleration, rather than increase in the rate of body size evolution of reef species is not consistent with diversification along an ecological axis. Surprisingly, the fastest evolving clade across tetraodontids (Takifugu, possibly including closely related sister species) is neither freshwater nor reef‐associated, but rather is found mostly on marine, soft‐bottom ecosystems in both temperate and tropical latitudes. Timing of diversification within this lineage implicates speciation induced by Pleistocene glaciation cycles, although other factors may also have been important.

Acknowledgments

This study was made possible by tissue loans from P. Wainwright (UC Davis), John Lundberg and Mark Sabaj (The Academy of Natural Sciences Philadelphia), H.J. Walker and Phil Hastings (Scripps Institute of Oceanography), Tom Near, Alex Dornburg and Greg Watkins‐Colwell (Yale Peabody Museum), Andrew Bentley and Ed Wiley (University of Kansas), Brian Bowen (Hawai'i Institute of Marine Biology), Daniel Neafsey (Harvard University), John Friel (Cornell University), David Jacobs and Greer Dolby (UCLA), Gianluca Polgar (University of Roma I and University of Malaya), Bruno Frédérich (Université de Liège), Unathi Lwana (South Africam Institute for Aquatic Biodiversity), Dianne Bray (Museum Victoria), Mark McGrouther (Australian Museum). Art North was instrumental in obtaining rare freshwater pufferfish species through the pet trade. Jose Ugalde and CONAGEBIO helped with obtaining permits and collecting in Costa Rica. Funding was provided by NSF grant DEB 0842397 ‘Systematics and Influence of Coral Reefs on Diversification in Tetraodontiform Fishes’ to MEA and FS. Graham Slater (UCLA) and Luke Harmon (UI) offered advice on some of the analyses and Graham Slater helped with some of the illustrations. We thank Amisha Gadani for the fish illustrations used in Fig. 1 and deciding editor Andy Gardner and three anonymous reviewers for many helpful comments.

1 extinct species.