DO FRESHWATER FISHES DIVERSIFY FASTER THAN MARINE FISHES? A TEST USING STATE-DEPENDENT DIVERSIFICATION ANALYSES AND MOLECULAR PHYLOGENETICS OF NEW WORLD SILVERSIDES (ATHERINOPSIDAE)

Authors

  • Devin D. Bloom,

    1. Department of Biological Sciences, University of Toronto Scarborough, Toronto, Canada
    2. Department of Ecology and Evolutionary Biology, University of Toronto, Toronto, Canada
    Search for more papers by this author
  • Jason T. Weir,

    1. Department of Biological Sciences, University of Toronto Scarborough, Toronto, Canada
    2. Department of Ecology and Evolutionary Biology, University of Toronto, Toronto, Canada
    Search for more papers by this author
  • Kyle R. Piller,

    1. Department of Biological Sciences, Southeastern Louisiana University, Hammond, Louisiana
    Search for more papers by this author
  • Nathan R. Lovejoy

    1. Department of Biological Sciences, University of Toronto Scarborough, Toronto, Canada
    2. Department of Ecology and Evolutionary Biology, University of Toronto, Toronto, Canada
    Search for more papers by this author

Abstract

Freshwater habitats make up only ∼0.01% of available aquatic habitat and yet harbor 40% of all fish species, whereas marine habitats comprise >99% of available aquatic habitat and have only 60% of fish species. One possible explanation for this pattern is that diversification rates are higher in freshwater habitats than in marine habitats. We investigated diversification in marine and freshwater lineages in the New World silverside fish clade Menidiinae (Teleostei, Atherinopsidae). Using a time-calibrated phylogeny and a state-dependent speciation–extinction framework, we determined the frequency and timing of habitat transitions in Menidiinae and tested for differences in diversification parameters between marine and freshwater lineages. We found that Menidiinae is an ancestrally marine lineage that independently colonized freshwater habitats four times followed by three reversals to the marine environment. Our state-dependent diversification analyses showed that freshwater lineages have higher speciation and extinction rates than marine lineages. Net diversification rates were higher (but not significant) in freshwater than marine environments. The marine lineage-through time (LTT) plot shows constant accumulation, suggesting that ecological limits to clade growth have not slowed diversification in marine lineages. Freshwater lineages exhibited an upturn near the recent in their LTT plot, which is consistent with our estimates of high background extinction rates.

All sequence data are currently being archived on Genbank and phylogenetic trees archived on Treebase.

Explaining the disparity in species richness among clades and areas is a major goal of evolutionary biology. One of the more intriguing patterns of biodiversity is the extraordinary number of species on continents compared to oceans (Vermeij and Grosberg 2010; Mora et al. 2011). In fact, it has been estimated that species diversity on continents is 25 times more than that found in oceans (Briggs 1994; Benton 2001), despite the fact that oceans are vastly greater in size. For aquatic organisms, freshwater (continental) habitats make up only ∼0.01% of available aquatic habitat and yet harbor 40% of all fish species, whereas marine habitats comprise >99% of available aquatic habitat and yield only 60% of fish diversity (Horn 1972; Lundberg et al. 2000; Leveque et al. 2008; Eschmeyer et al. 2010). The question of why continental diversity greatly exceeds than that found in oceans remains largely unanswered.

To explain the discrepancy between marine and freshwater diversity, several hypotheses emphasizing ecological factors (e.g., productivity, size of primary producers, and ecological specialization) have been proposed (May 1994; Vermeij and Grosberg 2010). Although ecological factors undoubtedly play a role in shaping biodiversity, ultimately disparity in species richness among clades and areas is the result of differences in net diversification (speciation minus extinction) and transition (or dispersal) rates between habitats (Barraclough and Nee 2001; Wiens and Donoghue 2004; Ricklefs 2007; Wiens et al. 2011), and the age of clades (McPeek and Brown 2007). Indeed, Benton (2001) hypothesized that the disparity in species richness among continental and oceanic lineages is an outcome of “rocketing diversification rates” in continental lineages compared to “the more sluggish rates of diversification of life in the sea.” A largely unconsidered possibility is biased transition rates between marine and freshwater lineages over evolutionary time; there may be a high number of freshwater species because marine to freshwater transition rates are higher than freshwater to marine transitions (Maddison 2006; Maddison et al. 2007). Alternatively, freshwater lineages may be older than marine lineages, suggesting continents have simply had more time to accumulate diversity (McPeek and Brown 2007).

There are a number of reasons for predicting that freshwater lineages should have higher diversification rates than marine lineages. Causes of differential diversification may include environmental parameters, as well as intrinsic properties of organisms from each environment. Freshwater environments are generally expected to have more barriers limiting population connectivity, to have greater habitat complexity, and to be more heavily influenced by tectonic activity than marine environments (Strathmann 1990; May 1994). Marine organisms have greater geographic range size and less genetic structuring (Palumbi 1994; Bierne et al. 2003). All of these factors are known to influence speciation and extinction rates (Cracraft 1982; Jablonski 1987; Barraclough et al. 1998; Ribera et al. 2001; Jablonski 2008; Badgley 2010). However, the interaction of some of these parameters may yield unexpected patterns. For example, the probability of speciation should increase as geographic range size increases, and decrease as levels of gene flow increase (Kisel and Barraclough 2010; Kisel et al. 2011). So although oceans are vastly greater in size, thus predicting higher speciation rates, marine taxa also have higher levels of gene flow (Palumbi 1994; Bierne et al. 2003; Puebla 2009), which is thought to reduce the probability of speciation events (Kisel and Barraclough 2010). Benton (2001) also suggested that marine environments have strong ecological limits on clade growth, causing a slow-down in lineage accumulation, whereas continental environments have few ecological constraints. The first step toward disentangling how these various habitat parameters affect macroevolutionary patterns is testing the prediction that freshwater lineages have faster diversification rates.

Previous studies have demonstrated that macrohabitat (e.g., coral reefs vs. pelagic ocean) can influence diversification rates in marine fishes (Ruber et al. 2003; Ruber and Zardoya 2005; Alfaro et al. 2007). However, comparative studies on the diversification of marine and freshwater fishes have proven challenging. A recent study (Vega and Wiens 2012) addressed patterns of aquatic diversity but did not detect differences in diversification rates between marine and freshwater ray-finned fishes (Actinopterygii). However, due to the large number of actinopterygian species (∼30,000), the authors were not able to use methods that explicitly estimate speciation and extinction rates for marine or freshwater character states (Maddison et al. 2007; FitzJohn et al. 2009; Vega and Wiens 2012). This makes it difficult to use clades that have mixed character states to infer diversification dynamics in a comparative framework (Maddison et al. 2007).

Testing for the effects of marine versus freshwater habitat on macroevolutionary patterns is possible using a time-calibrated phylogenetic framework that allows speciation, extinction, and character transition rates to be parameterized independently. By using a state-dependent model, we can explicitly estimate speciation and extinction rates as a function of a particular character state (Maddison et al. 2007). Ideally, the clade to be analyzed should contain multiple freshwater and marine lineages. However, many clades of fishes (and other taxa) are restricted to either marine or freshwater habitats (Lee and Bell 1999; Vermeij and Dudley 2000; Vermeij and Wesselingh 2002; Bloom and Lovejoy 2011). For this reason, New World silversides (Atherinopsidae) are an excellent system for investigating patterns of aquatic diversification because current taxonomy and (albeit limited) phylogenetic data suggests multiple marine/freshwater transitions have occurred. In this study, we focus on the subfamily Menidiinae, which is composed of 44 freshwater and 25 marine species. Marine species are distributed in the western Atlantic from southern Canada to northern Brazil, and in the eastern Pacific from Baja California to Peru. Marine silversides are generally near-shore fishes, with a few representatives in pelagic ocean habitats (Dyer 2006). Freshwater species are distributed in eastern United States and Canada, central and southern Mexico, and throughout Central America (Lee et al. 1980; Reis et al. 2003; Miller 2005; Dyer 2006). Freshwater silversides are largely riverine fishes, rarely occurring in natural lakes. The Central Mexican lakes clade of silversides is an exception with up to 13 species occurring in three lakes (Barbour 1973a, 1973b; Bloom et al. 2009). Previous phylogenetic studies utilizing morphological (Chernoff 1986a; White 1986; Dyer and Chernoff 1996; Dyer 1998) and molecular data (Bloom et al. 2012) have supported the monophyly of Menidiinae. Although several studies have investigated species level relationships in subclades within Menidiinae (Gosline 1948; Barbour 1973b; Echelle and Echelle 1984; Chernoff 1986b; Dyer 1998; Bloom et al. 2009), no molecular study has investigated the relationships among the major lineages of Menidiinae.

Here we use a phylogenetic approach to test for differences in diversification rates between marine and freshwater lineages of menidiine silversides. We generate a time-calibrated molecular phylogeny for Menidiinae, and reconstruct the number and timing of marine and freshwater habitat transitions. We ask whether freshwater lineages are older than marine lineages, and we use state-dependent models of diversification to test for differences in rates of speciation, extinction, and character state change between marine and freshwater lineages. Finally, we use lineage-through time (LTT) plots (Weir 2006; Rabosky and Lovette 2008a) to evaluate whether marine and freshwater lineages fit similar or different patterns of clade growth. Together, these tests and analyses provide novel insights into the macroevolutionary processes that determine differences in large-scale patterns of diversity between continents and oceans.

Materials and Methods

TAXON SAMPLING, DNA EXTRACTION, POLYMERASE CHAIN REACTION (PCR) AMPLIFICATION, AND SEQUENCING

Our taxon sampling included 50 out of 74 species and representatives of all genera currently recognized in Menidiinae (Table S1). We also included, as outgroups, six species from Atherinopsinae, Notocheirus hubbsi (Notocheirinae), and Atherinomorus stipes (Atherinidae) for a total of 58 species in the dataset. Our taxon sampling includes representatives from every genus in the family Atherinopsidae except Colpichthys and comprises the most comprehensive molecular phylogenetic study of Atherinopsidae to date. Some data were available from previous studies (Bloom et al. 2009, 2012), whereas most were newly sequenced for this study; all new sequences were deposited in Genbank (Table 1).

Table 1. Summary of specimens used in this study, including habitat (M = marine, F = Freshwater) and associated Genbank and museum numbers
GenusSpeciesHabitatCytBND2TMORAG1Museum numberCollection locality
AtherinellabalsanasFKC736414KC736421KC633404KC669427SLU 6637Rio Cancita, Michoacan, Mexico
AtherinellabalsanasFKC736415KC736422KC633405KC669428SLU 6637Rio Cancita, Michoacan, Mexico
AtherinellabrasiliensisMKC736412KC736425KC633402KC669431ROM 88861Chatham, Trinidad
AtherinellabrasiliensisMKC736413KC736426KC633403KC669432ROM 88861Chatham, Trinidad
AtherinellachagresiFKC736363KC736429KC633337KC669435STRI-02090Rio Canazas, Panama
AtherinellaalvareziFKC736374KC736416KC633347KC669424SLU 6772Laguna de Caobas, Mexico
AtherinellaalvareziFKC736375KC736417KC633348naSLU 6772Laguna de Caobas, Mexico
AtherinellaammophilaFKC736391KC736418KC633366KC669425SLU-TC 380Rio la Palma, Mexico
AtherinellaargenteaMJQ282017KC736419KC633335JQ282062ROM 91571Playla Cruzas, Panama
AtherinellaargenteaMKC736360KC736420KC633336KC669426ROM 91571Playla Cruzas, Panama
AtherinellablackburniMKC736356KC736423KC633333KC669429ROM 93997Playa Baraka, Galeta, Panama
AtherinellablackburniMKC736357KC736424KC633334KC669430ROM 93997Playa Baraka, Galeta, Panama
AtherinellacrystallinaFKC736346KC736427KC633321KC669433SLU 6087Rio Acaponeta, Nayarit, Mexico
AtherinellacrystallinaFKC736347KC736428KC633322KC669434SLU 6087Rio Acaponeta, Nayarit, Mexico
AtherinellapellosemionFKC736348KC736432KC633323KC669438SLU 6115Rio Mancuernas, Nayarit, Mexico
AtherinellapellosemionFKC736349KC736433KC633324KC669439SLU 6115Rio Mancuernas, Nayarit, Mexico
AtherinellacolombiensisFKC736364KC736430KC633338KC669436STRI 02097Rio San Juan, Colombia
AtherinellacolombiensisFKC736365KC736431KC633339KC669437STRI 02097Rio San Juan, Colombia
AtherinellaguatamalanensisFKC736380EF602045KC633353KC669443SLU 5012Rio Tehuantepec, Mexico
AtherinellaguatamalanensisFKC736381KC736434KC633354KC669444SLU 5085Laguna Coyucan, Mexico
AtherinellaguatamalanensisFKC736386KC736435KC633360KC669445SLU 6074El Teucan Lagoon, Mexico
AtherinellaguatamalanensisFKC736387KC736436KC633361KC669446SLU 6074El Teucan Lagoon, Mexico
AtherinellahubbsiFKC736388KC736437KC633362KC669447SLU 6853Costa Rica
AtherinellahubbsiFJQ282020KC736438KC633363JQ282065SLU 6853Costa Rica
AtherinellamarvalaeFJQ282021KC736439KC633355JQ282066SLU 6690Rio Papaloapan, Mexico
AtherinellamarvalaeFKC736382KC736440KC633356KC669448SLU 6690Rio Papaloapan, Mexico
AtherinellamilleriFKC736378KC736441KC633351KC669449SLU 5104Cangrejal river, Honduras
AtherinellamilleriFKC736379KC736442KC633352KC669450SLU 5104Cangrejal river, Honduras
AtherinellaserrivomerMKC736358KC736450naKC669458ROM 93998Play Peten, Boca Parita, Panama
AtherinellaserrivomerMKC736359KC736451naKC669459ROM 93998Play Peten, Boca Parita, Panama
AtherinellapanamensisMKC736361KC736443naKC669451ROM 93999Playa la Cruzas, Panama
AtherinellapanamensisMKC736362KC736444naKC669452ROM 93999Playa la Cruzas, Panama
AtherinellasalleiFKC736383KC736445KC633357KC669453SLU 5005Rio Hueyapan, Mexico
AtherinellasalleiFKC736384KC736446KC633358KC669454SLU 5005Rio Hueyapan, Mexico
AtherinellasardinaFKC736389KC736447KC633364KC669455ROM 94000L. Apoyo, Nicaragua
AtherinellasardinaFKC736390KC736448KC633365KC669456ROM 94000L. Apoyo, Nicaragua
AtherinellaschultziFKC736376EF602044KC633349naSLU 5103Rio Palenque, Mexico
AtherinellaschultziFKC736377KC736449KC633350KC669457SLU 5103Rio Palenque, Mexico
Atherinellan. sp.MKC736350KC736453KC633327KC669460SLU 6853Pacific Ocean, Mazatlan, Mexico
Atherinellan. sp.MKC736351KC736454KC633328KC669442SLU 6853Pacific Ocean, Mazatlan, Mexico
AtherinellaschultziFKC736385KC736452KC633359KC669461SLU TC5008Rio Almoloya, Mexico
AtherinellastarksiMKC736352KC736455KC633329KC669462ROM 94001Tabago Island, Panama
AtherinellastarksiMKC736353KC736456KC633330KC669463ROM 94001Tabago Island, Panama
AtherinomorusstipesMJQ282023KC736457KC633315JQ282068ROM 91573Barbados
AtherinomorusstipesMKC736343KC736458KC633316KC669464ROM 91573Barbados
AtherinopsaffinisMnaKC736459KC633397JQ282061SIO 0581Pacific Ocean, California, USA
AtherinopsiscaliforniensisMJQ282018KC736460KC633398JQ282063SIO 03458Pacific Ocean, California, USA
BasilichthyssemotilusFJQ282024EF602042naJQ282069ANSP 180736R. Santuario, Peru
ChirostomaargeFnaEF602099KC633389naSLU 5110Laguna Negritas, Mexico
ChirostomaattenuatumFKC736404EF602083KC633387KC669465SLU 5036L. Zirahuen, Mexico
ChirostomaattenuatumFKC736405EF602082KC633388KC669466SLU 5036L. Patzcuaro, Mexico
ChirostomachapalaeFKC736397EF602075KC633375KC669467SLU 5016L. Chapala, Mexico
ChirostomaconsociumFJQ282025EF602078KC633381JQ282070SLU 5023L. Chapala, Mexico
ChirostomaconsociumFKC736401KC736461KC633382KC669468SLU 5023L. Chapala, Mexico
ChirostomacontrerasiFKC736406EF602098KC633390KC669469SLU 5080Rio Laja, Mexico
ChirostomaestorFKC736403EF602068KC633385KC669470SLU 5114L. Patzcuaro, Mexico
ChirostomagrandoculeFKC736369EF602061nanaSLU 5118L. Patzcuaro, Mexico
ChirostomahumboldtianumFKC736402EF602070KC633383naSLU 5095San Pedro Lagunillas, Mexico
ChirostomahumboldtianumFJQ282026EF602071KC633384JQ282071SLU 5011Lago de Zacapu, Mexico
ChirostomajordaniFKC736407EF602086KC633391KC669471SLU 5033L. Chapala, Mexico
ChirostomajordaniFJQ282027EF602090KC633392JQ282072SLU 5033L. Cuitzeo, Mexico
ChirostomalabarcaeFKC736399EF602084KC633378KC669472SLU 5017L. Chapala, Mexico
ChirostomalabarcaeFJQ282073EF602085KC633379JQ282028SLU 5017L. Chapala, Mexico
ChirostomaluciusF EF602059 naSLU 5022L. Negritos, Mexico
ChirostomapatzcuaroFJQ282029EF602063KC633386JQ282074SLU 5117L. Patzcuaro, Mexico
ChirostomapromelasFKC736368EF602060KC633342naSLU-TC 925Tizaplan Hatchery, Mexico
ChirostomariojaiFKC736398EF602096KC633376KC669473SLU 5079L. Guadalupe Victoria, Mexico
ChirostomariojaiFJQ282030EF602097KC633377JQ282075SLU 5079L. Guadalupe Victoria, Mexico
ChirostomasphyraenaFKC736400EF602065KC633380KC669474SLU 5025L. Chapala, Mexico
LabidesthesvanhyningiFKC736409EF602057KC633395KC669475SLU 5106Pine Log Creek, FL, USA
LabidesthessicculusFJQ282031KC736462KC633396JQ282077SLU-TC 607Duck River, TN, USA
LeuresthestenuisMJQ282032KC736463KC633399JQ282078SIO 0563Pacific Ocean, California, USA
MelanorhinusmicropsMKC736344KC736474KC633317KC669482ROM 91572Archers Bay, Barbados
MelanorhinusmicropsMJQ282037KC736475KC633318JQ282083ROM 91572Archers Bay, Barbados
MembrasgilbertiMJQ282034KC736469KC633331JQ282080ROM 91569Tabago Island, Panama
MembrasgilbertiMKC736355KC736470KC633332KC669479ROM 91569Tabago Island, Panama
MembrasmartinicaMJQ282035KC736471KC633369KC669480SLU 5102Wrightsville Beach, NC, USA
MembrasmartinicaMKC736393KC736472KC633370JQ282081SLU 5102Wrightsville Beach, NC, USA
MenidiaberyllinaFKC736408EF602049KC633393KC669476SLU 5108Bayou Lacombe, LA, USA
MenidiaberyllinaFJQ282033KC736464KC633394JQ282079ROM 91570L. Ponchartrain, LA, USA
MenidiacoleiMKC736372KC736465KC633345KC669477SLU-TC 1914Laguna de Caobas, Mexico
MenidiacoleiMKC736373KC736466KC633346KC669478SLU-TC 1915Laguna de Caobas, Mexico
MenidiaextensaFKC736370KC736467KC633343nanaLake Waccamaw, NC, USA
MenidiaextensaFKC736371KC736468KC633344nanaLake Waccamaw, NC, USA
MenidiamenidiaMJQ282036KC736473KC633367JQ282082SLU-TC 2240Wrightsville Beach, NC, USA
MenidiamenidiaMKC736392EF602050KC633368KC669481SLU-TC 2241Wrightsville Beach, NC, USA
MenidiapeninsulaeMJQ282038KC736476KC633319JQ282084SLU 5107Panama City, FL, USA
MenidiapeninsulaeMKC736345KC736477KC633320KC669483SLU 5107Panama City, FL, USA
NotocheirushubbsiMJQ282012nanaJQ282054naAtlantic Ocean, Argentina
OdonthesthesmauleanumFKC736410KC736478KC633400KC669485naRio Itata, Chile
OdonthesthessmittiMKC736411KC736479KC633401KC669486naPuerto Madryn, Argentina
PoblanaletholepisFKC736367EF602105KC633341KC669490SLU 5116L. Preciosa, Mexico
PoblanasquamataFKC736366EF602112KC633340naSLU 5115L. Preciosa, Mexico
PoblanaalchichicaFKC736395EF602109KC633373KC669487SLU 5034L. Alchichica, Mexico
PoblanaalchichicaFKC736396EF602110KC633374KC669488SLU 5034L. Alchichica, Mexico
PoblanaferdebueniFKC736394EF602100KC633371KC669489SLU 5028L. Chignahuapan, Mexico
PoblanaferdebueniFJQ282039EF602101KC633372JQ282085SLU 5028L. Chignahuapan, Mexico

Whole genomic DNA was extracted using DNeasy tissue kit (Qiagen, Valencia, CA). We PCR amplified fragments of two mitochondrial (nd2 and cytb) and two single-copy protein-coding nuclear (tmo4C4 and rag1) genes. For amplification and sequencing, we used protocols and primers from Bloom et al. (2012, 2009) for nd2, cytb, and rag1, and Lovejoy et al. (2004) for tmo4c4. Sequences were edited using the computer software Geneious v5.4 (Drummond et al. 2010) and aligned using the MUSCLE module (Edgar 2004) implemented in Genious. Following alignment, coding regions were translated to amino acids to confirm the integrity of reading frames and absence of stop codons. To test the monophyly of species and as a measure of quality control, we sequenced multiple individuals for nearly all newly sequenced species. We included multiple individuals of each species in the MrBayes analyses, but removed duplicate species representatives for all subsequent BEAST and diversification analyses described later.

PHYLOGENETIC ANALYSES AND DIVERSIFICATION TIME ESTIMATION

Best-fit models of nucleotide substitution were selected for each gene using Akaike information criteria in the program jModelTest (Posada 2008). We used BEAST v1.6.1 (Drummond and Rambaut 2007) to jointly estimate phylogeny and divergence times under a relaxed clock, uncorrelated lognormal model (Drummond et al. 2006) that allows rates to vary among branches. The dataset was partitioned by gene, with each partition unlinked and set to a general time reversible (GTR)ur model with γ-distributed rate heterogeneity. We used a birth-death prior for rates of cladogenesis and ran two independent analyses of 100 million generations sampling every 1000 generations. We used Tracer 1.5 (Drummond and Rambaut 2007) to evaluate convergence and mixing of runs and to verify that effective sample sizes were >200 for all parameters. We determined that the first 30 million generations from the Markov chain Monte Carlo (MCMC) sample were a conservative burn-in. The two converged runs were combined using LogCombiner v1.6.1 (Drummond and Rambaut 2007) and the maximum credibility tree (MC tree) was generated in TreeAnnotator v1.6.1 (Drummond and Rambaut 2007).

We used three fossil constraints in the BEAST analysis. A fossil Basilichthys is dated to the late Miocene (Rubilar 1994) and was used to constrain the Basilichthys/Odontesthes clade to a hard minimum bound of 5.33 ma with a soft upper bound of 75 ma based on previous molecular estimates that show Atherinomorpha is less that 75 million years old (Alfaro et al. 2009; Santini et al. 2009). A fossil Menidia from Oklahoma was dated to the lower Pliocene (Hubbs 1942). This fossil is a crown Menidia and was used to constrain the clade including Menidia beryllina, Menidia colei, and Menidia peninsulae with a hard minimum bound of 1.8 ma and a soft upper bound of 75 ma. A fossil Chirostoma is known from the Lerma Basin in Central Mexico and dated to the Pleistocene with a minimum age reported of 46,000 years before present (Bradbury 1971; Piller and Barbour, in press). The fossil Chirostoma is a member of the crown “humboldtianum” clade of large-bodied Chirostoma (Bradbury 1971; Echelle and Echelle 1984; Bloom et al. 2009). This fossil was used to date the clade including C. humboldtianum, C. sphryaena, C. lucius, C. chapalae, C. estor, C. grandocule, and C. consocium to a hard minimum age of 0.0046 ma and a soft upper bound of 5.33 ma based on the fossil Menidia (Hubbs 1942). We used exponential priors for all fossil constraints because this distribution is appropriate for crown fossils and when no information is available to determine the shape of a lognormal distribution (Ho and Phillips 2009).

We estimated phylogenetic relationships using a mixed-model partitioned-by-gene Bayesian analysis implemented in MrBayes v3.1.2 software (Ronquist and Huelsenbeck 2003). We ran two MrBayes analyses, which consisted of four independent chains run for 10 million generations, sampling every 1000 generations with all parameters unlinked and default priors. Convergence of MrBayes analyses was assessed by comparing likelihood states over generations using the sump command in MrBayes and by confirming that standard deviation of split frequencies remained below 0.01 and potential scale reduction factors were 1.0. Adequate mixing of chains was confirmed by determining that acceptance rates were between 10% and 70%. MrBayes tree searches were used to confirm the topology recovered in our BEAST analyses, but subsequent diversification analyses and trait reconstructions were estimated using the trees from the BEAST analyses.

LINEAGE DIVERSIFICATION ANALYSES

Our diversification analyses were conducted on Menidiinae only by pruning all non-Menidiinae from the tree. We used a state-dependent speciation and extinction model to estimate diversification rates for lineages with marine or freshwater states. For binary characters, the BiSSE model estimates the probability that a lineage evolved as observed given a set of speciation (λ), extinction (μ), and character transition (q01 and q10) parameters (Maddison et al. 2007). Using this framework, we compared the fit of models with unconstrained parameters (speciation, extinction, and transition rates allowed to vary) to models with these parameters constrained to be equal to explicitly test the hypothesis that speciation and extinction rates were different between marine and freshwater habitats, and that there were asymmetrical transition rates between habitats (i.e., test for asymmetrical character transitions). Recent implementations of BiSSE correct for a known proportion of missing species in each character state. We accounted for missing species using the proportions of included and missing taxa for each character state: 13 marine (proportion = 0.5), 11 freshwater (proportion = 0.2292), assuming missing species represent a random sample. The BiSSE models we implemented assume that parameters are constant through time. Although more complex models in which parameters change as a function of time can also be used, these greatly increase the number of estimated parameters. Given the size of our tree (50 species) we felt it was unwise to test models with time variable parameters.

Marine lineages might transition to a freshwater state (and vise versa) via an intermediate state defined as occupancy of both freshwater and marine habitats. Indeed, two silverside lineages that are primarily freshwater are known to occasionally occur estuarine habitats (M. beryllina and A. guatamalanensis). We coded these species as transitional and used GeoSSE (Goldberg et al. 2011), which allows for transitional states, as an alternative to BiSSE.

Model parameters of our BiSSE analysis were optimized in a maximum likelihood framework on our MC tree and best-fit models were selected using Akaike Information Criteria (AIC) following Maddison et al. (2007). We also estimated parameters on our MC tree in a Bayesian framework using MCMC sampling to more effectively search for optimal parameter estimates. Our MCMC parameter searches consisted of 10,000 iterations with 2500 discarded as burn-in. We used maximum likelihood parameter estimates as starting values in the MCMC analyses. Exponential priors were used following the examples in the DiversiTree package (FitzJohn et al. 2009). To incorporate branch length and topology uncertainty, we obtained parameter estimates across a sample of 97 of our Bayesian trees generated in BEAST. Trees were sampled every 700,000 generations following the burn-in period. For each tree, we performed MCMC parameter searches of 5000 iterations, with a 2000 iteration burn-in. All BiSSE analyses were conducted in the R package DiversiTree (FitzJohn et al. 2009).

Our silverside tree has only 50 species, yet our best-fit model using BiSSE has five parameters (Table 2). We ran a power analysis to determine if our best-fit model is appropriate and if our data provided sufficient signal to obtain reasonable parameter estimates. Using our maximum likelihood parameter estimates, we simulated 1700 trees and their associated character state data (marine or freshwater) under the best-fit model (CRλ01, CRμ01, CRq). Each tree was simulated with 74 tips (the total number of species in our group), and 24 tips were randomly pruned, resulting in trees with 50 tips (the actual level of sampling we obtained). Maximum likelihood was then used to fit all the models in Table 2 to each simulated tree, and log-likelihoods and parameter estimates were recorded. If the parameter estimates we obtained under our best-fit model are reasonable, then these parameter estimates should be similar to those from simulated data. Similarly, if the best-fit model represents a reasonable model for our data, then it should also be chosen as the best-fit model in simulated datasets. We used AIC and likelihood ratio tests to determine model fit. For the likelihood ratio test, twice the difference in log-likelihoods between constrained and unconstrained models should follow a χ2 distribution with degrees of freedom equal to the difference in the number of free parameters between the two models. To test this assumption, we obtained empirical estimates of the critical cutoff value using simulation (see supporting online methods). The empirical cutoff value (3.43) was similar to the expected cutoff value (3.84) under a χ2 distribution with 1 degree of freedom. Given the similarity, we consider the χ2 approximation appropriate for significance testing with likelihood ratio tests (see S1).

Table 2. Summary of model statistics from BiSSE analyses of speciation, extinction, and habitat transitions of silversides. Marine habitats are coded as state 0 and freshwater habitats as state 1. Speciation rates are λ, extinction rates are μ, and character transition rates are q. An “=” sign indicates model parameters constrained as equal, and a ““sign indicates parameters estimated independently for each habitat state. The best-fit model is in bold
ModelLn LParametersAICΔAICexp(−ΔAIC/2)Akaike weights
λ0 = λ1, q01 = q10−160.64442325.288827.611.01299E−060.0000
λ0 = λ1, μ0 = μ1, q01 = q10−153.24003312.480014.800.0006123540.0004
λ0 ≠ λ1, q01 = q10−153.17213312.344214.660.0006553770.0005
λ0 ≠ λ1, q01 ≠ q10−152.42434312.848615.170.0005092860.0004
λ0 ≠ λ1, μ0 = μ1, q01 = q10−152.23634312.472614.790.0006146240.0004
λ0 ≠ λ1, μ0 ≠ μ1, q01 = q10143.84185297.68360.0010.7294
λ0 ≠ λ1, μ0 ≠ μ1, q01 ≠ q10−143.84016299.68022.000.3685053680.2688

ANCESTRAL CHARACTER RECONSTRUCTIONS AND LINEAGE DIVERSIFICATION THROUGH TIME ANALYSIS

We coded marine and freshwater habitats as discrete, unordered character states. All character reconstructions were conducted on the chronogram resulting from the BEAST analyses. We used maximum parsimony (MP) and maximum likelihood (ML) in Mesquite v2.6 (Maddison and Maddison 2011) to reconstruct ancestral character states and determine the number of transitions between marine and freshwater habitats. Maximum likelihood reconstructions were estimated using the Mk model (Pagel 1999).

We generated LTT plots to evaluate patterns of clade growth. The log number of LTT indicates the relative rate of lineage accumulation (Weir 2006). If there are no limits to clade growth, lineages are expected to accrue exponentially through time yielding an LTT plot with a straight slope. A slow-down in lineage accumulation over time is characterized by a decline in slope through time and indicates the existence of a diversity-dependent threshold on clade growth. A strong upturn in the slope of LTT plots near the recent is the signature of high background extinction (although a recent increase in speciation rate can mirror this pattern; Weir 2006; Rabosky and Lovette 2008a). We explored patterns of clade growth for the entire Menidiinae clade as well as clade growth separately in marine and freshwater habitats. We followed the approach of Weir (2006) to generate LTT plots for marine and freshwater clades separately; this method uses information from ancestral state reconstructions to determine lineage accumulation patterns based on character state (assuming transitions occur at nodes, nodes representing a transition to, or diversification within the marine state were included in the marine LTT plot; nodes representing transition to, or diversification within a freshwater state were included in the freshwater LTT plot; see Weir 2006 for details). We also calculated the γ statistic (Pybus and Harvey 2000) for marine and freshwater lineages independently (using the same separation of nodes used done in the LTT analysis) using the R package Laser (Rabosky 2006b).

Results

MOLECULAR DATA AND PHYLOGENETICS

Our final dataset consisted of 1047bp from nd2, 1121bp of cytb, 534bp of tmo4c4, and 1141bp of rag1. The combined dataset included 4143 characters, 1441 of which were parsimony informative. Figure 1 shows our time-calibrated phylogeny, which is well resolved and has strong support for clades associated with habitat transitions, indicating phylogenetic uncertainty is not problematic for character reconstructions (Figs. S1 and S2). We find strong support for a monophyletic Atherinopsidae, and our focal clade, Menidiinae. Within Menidiinae, we found two major species groups: the Menidiini, distributed in North America, the Caribbean, and the Central Mexican Plateau, and the Membradini, distributed in Central America and southern Mexico. Our analyses place Atherinella brasiliensis as sister to all other members of Menidiinae; previously this taxon was thought to be a member of the Membradini clade (Chernoff 1986a). We also recovered Melanorhinus microps as sister to all other Menidinii, rather than in its previously proposed position as a member of Membradini (Chernoff 1986a; Dyer 1997). Many currently recognized genera were not recovered as monophyletic and are likely in need of taxonomic revision. Aside from these differences, our results are consistent with previous studies on silverside phylogenetics at or above the genus level (Chernoff 1986a; Dyer and Chernoff 1996; Dyer 1998, 2006; Bloom et al. 2009; Bloom et al. 2012).

Figure 1.

Time-calibrated maximum clade credibility phylogeny of silversides from BEAST analysis of four-gene dataset (referred to in the text as MC tree). Branch colors and circles at nodes indicate maximum likelihood ancestral character reconstructions for marine (red) and freshwater (blue) habitats. Branches are proportional to absolute time and the x-axis is in millions of years (ma) before present day.

ANCESTRAL CHARACTER RECONSTRUCTIONS

Ancestral character reconstructions indicate a marine state for the most recent common ancestor (MRCA) of Menidiinae and both of the major clades Menidiini and Membradini. Within Menidiinae, there were four independent transitions from marine to freshwater habitats, occurring in the following geographic regions: (1) Mississippi and Atlantic drainages of eastern North America and the central plateau region of Mexico (within Menidiini); (2) Atlantic drainages along the Isthmus of Tehuantepec and western margins of the Yucutan Peninsula in southern Mexico (Atherinella schultzi and A. ammophila); (3) coastal and southern Mexico and Central America, including Atlantic and Pacific drainages (clade including 11 species of Atherinella); and (4) the Pacific slope of Colombia (Atherinella colombiensis). There were also three reversals from freshwater back to marine habitats. These freshwater to marine transitions occurred: (1) along the Atlantic coast of North America (Menidia menidia), (2) in the Gulf of Mexico (Menidia peninsulae and Menidia colei), and (3) along the Atlantic coast of Central America (Atherinella argentea). Qualitatively, this suggests that asymmetrical transition rates do not explain freshwater species richness; quantitative evidence is provided later.

DIVERSIFICATION TIMES

Our time-calibrated phylogeny is deposited on TreeBase 13962. Our diversification time analysis dates the MRCA of Atherinopsidae to 36.94 ma, and the MRCA of Menidiinae to 26.89 ma (Figs. 1 and S2). The oldest node reconstructed as freshwater was the MRCA of Melanorhinus and members of Menidiini, dated to 19.59 ma. Within the Membradini clade, the oldest reconstructed freshwater node was dated to 14.59 ma. Over half of the 37 freshwater species included in our study date to less than 3.5 ma, and nine are members of a clade from the central Mexican plateau region that date to only 0.52 ma. Together this indicates that marine Menidiinae are at least 7 million years older than the earliest freshwater lineage and the majority of freshwater silverside lineages are relatively young.

SPECIATION, EXTINCTION, AND TRANSITION RATES

Our BiSSE analyses showed that pure birth models without extinction were poorly supported (Akaike weight < 0.0006; Table 2). Similarly, models constraining speciation and/or extinction rates to be equal for marine and freshwater species provided a much worse fit (Akaike weights < 0.0004; Table 2) than models that allowed these rates to vary. Our best-fit model was one in which speciation and extinction rates differed greatly between marine and freshwater biomes, but character transition rates were symmetrical (Akaike weight = 0.73; Tables 2). The next best-fit model was the full BiSSE model in which all parameters were asymmetrical (Akaike weight = 0.27; Table 2 and 3).

We explored the posterior distribution of parameter values along our MC tree under the full BiSSE model (Fig. 2). These distributions show that speciation and extinction rates were much greater in freshwater than marine lineages. For speciation and extinction rates, the 95% credible intervals did not overlap between freshwater and marine lineages (indicating a high degree of certainty in these estimates), but overlap did occur for net diversification rates and transitions rates between marine and freshwater.

Figure 2.

Posterior distribution of speciation rates (A), extinction rates (B), net diversification rates (C), and character transition rates (D) for marine (red) and freshwater (blue) silverside lineages from our BiSSE Bayesian analysis.

To determine whether model parameters varied significantly between marine and freshwater while simultaneously accounting for topology and branch length uncertainty, we estimated posterior distributions of parameter values across a sample of 97 posterior trees for the full model. For each Bayesian MCMC sample we subtracted parameter estimates in marine environments from those in freshwater. If rates are significantly different between marine and freshwater, the resulting posterior distributions will not overlap in their 95% credible intervals. These intervals were used to obtain 2-tailed P-values. Results of this analysis indicate that speciation and extinction rates were both significantly higher in freshwater than marine biomes (speciation: P = 0.0016; extinction: P = 0.0176; Fig. 2). Net diversification rates and transition rates between marine and freshwater were not significantly different (net rates: P = 0.6956; transition rates: P = 0.7297).

Maximum likelihood estimates of speciation and extinction rates under GeoSSE were qualitatively the same as under BiSSE (faster speciation, extinction, and net diversification rates in freshwater vs. marine). Given the similarity in estimates, we report parameter estimates only for BiSSE (Table 3).

Table 3. Speciation rate, extinction rate, and transition rate estimates from the fully unconstrained model (λ0 ≠ λ1, μ0 ≠ μ1, q01 ≠ q10) in the BiSSE analyses. See Table 2 for symbol definitions
λ0λ1μ0μ1q01q10
0.056840380.812134330.017716470.760058440.018351870.02166593

Our power analysis using 1700 simulated phylogenies and character datasets strongly support both the appropriateness of the best-fit BiSSE model and the utility of that model to accurately estimate parameter values. First, using AIC, the best-fit model for 76% of simulated trees was the same model that trees were simulated under, demonstrating that the various models we tested can be accurately discriminated. When using likelihood ratio tests, the model under which data were simulated rejected simpler models for 89–99.9% of simulations (depending on which models were being compared). Second, the distribution of parameter estimates under the best-fit model obtained from the simulated datasets overlapped closely with the actual parameter estimates we obtained for our phylogeny (Fig. S3). Importantly, speciation and extinction rates were significantly higher in freshwater versus marine species in our power analysis. These results indicate that not only does our silverside tree of 50 species have sufficient power to estimate a realistic model (despite lacking 24 species), but it also provides reasonable parameter estimates of both speciation and extinction.

LTT ANALYSES

Our LTT plots (Fig. 3) show that the slope of the LTT is steeper in freshwater lineages than in marine lineages, which is consistent with the faster rates of freshwater speciation estimated in our BiSSE analysis. Freshwater lineages also show a significant upturn near the present (γ = 2.937, P = 0.0033) that is characteristic of either high extinction rates or rapid, recent speciation rates (Nee et al. 1994; Rabosky 2006a). BiSSE estimated high background extinction in freshwater lineages suggesting that the upturn near the recent is due to extinction, and not a recent increase in speciation rates. The LTT for marine lineages shows only a slight (and nonsignificant; γ = −0.577, P = 0.564) downturn through time, suggesting almost constant lineage accumulation over time.

Figure 3.

Lineage-through time plots (LTT) for silversides from marine (red) and freshwater (blue) habitats.

Discussion

Theory on species-area relationships (MacArthur and Wilson 1967) suggest that oceans should be more species rich than continents, and yet oceans harbor only 5–15% of all species (Vermeij and Grosberg 2010). Here we provide explicit estimates of speciation and extinction parameters using state-dependent diversification analysis to test the hypothesis that freshwater (continental) lineages have higher diversification rates than marine lineages. Our data suggest that freshwater silversides had both higher speciation rates than marine lineages, but also higher extinction rates (Table 3). The resulting net diversification rates were slightly higher in freshwater versus marine lineages, but the differences was not statistically significant. Importantly, marine silverside lineages are generally older than freshwater lineages, suggesting that freshwater lineages have had less time to generate species. We also found that freshwater silverside diversity is not due to asymmetrical transition rates between habitats, because transition rates between habitats were roughly equal and a model with transition rates constrained to be equal had almost identical likelihood and much stronger Akaike weight (Table 2). Together, these results support the idea that differences in habitats can drive macroevolutionary processes (Ribera et al. 2001; Hughes and Eastwood 2006; Alfaro et al. 2007; Moore and Donoghue 2007, 2009; Kozak and Wiens 2010) and are an important determinant of broad-scale patterns of diversity (Wiens et al. 2011). Below we discuss the evidence and possible causes for the differences in speciation and extinction rates between aquatic habitats.

SPECIATION RATES ELEVATED IN FRESHWATER

Our results confirm that speciation rates are higher in freshwater than marine silverside lineages, showing a 15-fold difference in rate (Fig. 2; Tables 2 and 3). We propose that the most probable cause of this discrepancy is differences in the abundance of physical (vicariant) barriers that limit gene flow and promote genetic divergence. Most speciation events in fishes are likely a result of allopatric speciation via vicariance (Cracraft 1985; Lynch 1989; Coyne and Orr 2004), and the frequency of vicariant events is likely higher in freshwater than marine habitats (Strathmann 1990; May 1994; but see Paulay and Meyer 2002; Dawson and Hamner 2008). Rivers in particular are strongly influenced by geological events that cause stream capture or drainage subdivision, which are widely recognized causes of vicariant events in freshwater fishes (Rosen 1978; Mayden 1988; Waters et al. 2001; Burridge et al. 2006; Albert and Carvalho 2011). Freshwater habitats are often highly fragmented even on small spatial scales (i.e., within drainages, or between adjacent drainages; Burridge et al. 2008), resulting in high levels of microendemism and population structure (Hollingsworth and Near 2009; Keck and Near 2010), which may lead to elevated speciation rates. In addition, the high degree of habitat complexity in freshwater systems likely facilitates local adaptation, a process that has been demonstrated to play an important role in the diversification of freshwater fishes (Fuller et al. 2007; Tobler et al. 2008; Plath et al. 2010; Tobler et al. 2011). Many freshwater silverside lineages show biogeographic patterns that are consistent with allopatric speciation via river basin isolation.

In contrast, oceans are more “open” and contiguously connected ecosystems (Rapoport 1994; Carr et al. 2003). Many marine fishes, including silversides (Watson 1996), have pelagic planktonic larvae that disperse long distances, and accordingly, marine species are known to have high levels of gene flow and population connectivity (Waples 1987; Palumbi 1994; Bohonak 1999; Bierne et al. 2003; Hellberg 2009; Puebla 2009). As a result, marine fishes tend to have lower levels of population structure than freshwater fishes (Ward et al. 1994; MaKinen et al. 2006). Together, high population connectivity and gene flow in marine ecosystems dampen the effects of local adaptation and impede speciation (Bierne et al. 2003). There are notable exceptions where marine species show signatures of restricted gene flow over small spatial scales (e.g., Taylor and Hellberg 2003, 2005), but many of these taxa are reef associated and might have diversification rates on par with freshwater lineages (Bellwood and Wainwright 2002; Rocha et al. 2005; Alfaro et al. 2007; Rocha and Bowen 2008; Price et al. 2010, 2011). Marine silversides in our study are not reef associated, precluding our ability to compare diversification rates between reef and freshwater lineages. Rather most marine silversides have large, coastal distributions that likely contribute to low speciation rates.

Freshwater silversides from lakes of the Central Mexican Plateau (Chirostoma and Poblana) may represent a “species flock” that resulted (in part) from intra-lacustrine sympatric speciation (Echelle 1984). It is possible that the estimate of overall speciation rate in freshwater silversides is strongly influenced by rapid speciation in this clade. Although some studies suggest that allopatric speciation may be slower than sympatric speciation (McCune and Lovejoy 1998), others have demonstrated that allopatric speciation can occur just as rapidly, even on par with clades thought to be classic examples of sympatric speciation (Near and Benard 2004). If the latter is true, then the prevalence of a particular mode of speciation on continents or in oceans may not be a good predictor of disparity in species richness. To our knowledge, there is no study comparing the frequency of speciation modes between closely related marine and continental lineages. If sympatric speciation is more common in continental lineages, then the inclusion of a candidate clade for sympatric speciation in our study is informative and a critical component of a complete explanation for why continental species richness is high.

EXTINCTION RATES ELEVATED IN FRESHWATER

Extinction rates in freshwater silverside lineages were estimated to be almost 60 times higher than in marine lineages (Fig. 2b and Table 3). However, despite high extinction rates in freshwater lineages, net diversification (speciation rate minus extinction rate) is still 1.2 times higher in freshwater lineages, suggesting that high rates of faunal turnover in freshwater have not suppressed net diversification. Extinction rates estimated from molecular phylogenies are problematic (Rabosky 2010) and must be interpreted with caution. Nonetheless, our results for higher extinction rates in freshwater lineages are both biologically plausible and intriguing.

We propose that habitat connectivity and the ability to move in response to environmental disturbance may be key parameters that result in differences in extinction rates between marine and freshwater lineages (Jablonski 2008, and references therein). Marine habitats are stable, long lasting, and have high levels of connectivity (Lee and Bell 1999). In contrast, freshwater habitats are highly compartmentalized and spatially fragmented, with connectivity among drainages (and populations) dictated by the geomorphology of the region (Carr et al. 2003). Rivers and streams in particular have unique spatial structuring and ecosystem dynamics because they are dendritic networks, with drainages separated by uninhabitable (terrestrial and marine) areas (Grant et al. 2007). In the event of environmental disturbance (e.g., temperature change), marine fishes may be able to track suitable habitat because there are fewer barriers and physical restrictions to geographic species range shifts. For instance, during cooler periods tropical species occurring at higher latitudes might shift their ranges, moving to warmer latitudes near the equator (assuming biotic interactions allow it). Meanwhile, the fragmented nature of freshwater habitats reduces the possibility of range shifts in response to environmental disturbance such as climate change and marine incursions. Freshwater species may not have the necessary inter-drainage connections for dispersal, putting them at a higher risk of extinction (Fagan 2002; Fagan et al. 2002; Carr et al. 2003).

Both species range size and population size may also affect extinction rate. An inverse relationship between extinction rate and range size is widely supported by empirical and theoretical evidence (Jablonski 1987, 2007, 2008; Jablonski and Hunt 2006; Eastman and Storfer 2012), indeed range size is frequently cited as a likely cause of differential survival at the species level (i.e., species selection; Rabosky and McCune 2010). Marine silverside species tend to have large ranges whereas freshwater species are narrowly distributed, often limited to a single river drainage (Barbour 1973a; Chernoff 1986b); this pattern is likely general among teleost fishes. Numerically large populations are also expected to have lower extinction rates (Jablonksi 2008). Although there are no data comparing population sizes in silversides, marine taxa likely have larger populations, which may further reduce extinction risk. Our finding of higher extinction rates in freshwater compared to marine lineages may be a widespread pattern among aquatic taxa. If so, this finding has significant implications for both macroevolutionary patterns and conservation.

GEOGRAPHIC PATTERNS OF DIVERSITY AND LINEAGE ACCUMULATION

When resources are plentiful and competitors are limited, lineages are expected to accumulate exponentially through time (e.g., constant rates). As diversification continues and niche space fills, resources become limited and competition increases, resulting in diversity-dependent diversification and a slowdown (downturn) in lineage accumulation (Weir 2006; Rabosky and Lovette 2008b; Rabosky 2009a,b). Benton (2001, 2009) Because marine habitat is ancestral in silversides, it might be expected that diversity-dependent processes have regulated diversity here, whereas the more recently derived continental diversity may not yet have been subjected to ecological constraints on clade growth. However, our results showed that neither continental nor marine silversides showed a significant pattern of diversity-dependent lineage accumulation (Fig. 3). Instead marine lineages fit a constant growth (exponential) pattern of lineage accumulation, whereas continental lineages showed a significant upturn near the present in LTT plots. The increase near the present in the freshwater LTT plot (and γ value) is consistent with our estimated high background extinction rates. This further suggests that our estimated extinction rates are plausible despite the difficulty associated with estimating extinction rates.

The nearly straight (marine) or upturned (freshwater) LTT slopes suggest that silversides have not diversified to the point of saturating niche space in either environment (although diversification seems to be more limited in eastern North America as discussed later). We find that marine lineage accumulation plots have a shallower slope than freshwater lineages, which is consistent with our hypothesis that marine lineages accumulate at a slower rate due to lower diversification rates. The few studies to investigate patterns of lineage accumulation in marine fishes have shown mixed results, but have focused on clades that are reef associated and thought to be examples of marine adaptive radiations (Ruber et al. 2003; Ruber and Zardoya 2005; Alfaro et al. 2007; Cowman and Bellwood 2011). As discussed earlier, it is likely that reef-associated clades have different diversification dynamics than nonreef-associated clades. We suspect that the processes determining marine silverside diversification are likely shared with other near-shore nonreef marine fishes, and that ecological limits are not controlling clade growth in these taxa.

The geography of marine to freshwater transitions has played an important role in shaping patterns of diversity of silversides, as well as other fishes including anchovies (Bloom and Lovejoy 2012), needlefishes (Lovejoy and Collette 2001), and sea catfishes (Betancur-R et al. 2012). Silversides independently invaded freshwaters in eastern North America/Central Mexican Plateau, southern Mexico/Central America (twice), and the Pacific coast of Colombia. However, freshwater silverside diversity is not evenly distributed among these regions (Fig. 1). Diversity is low in Colombia (one species) and eastern North America (four species, of which only Labidesthes sicculus has invaded far beyond lowland coastal rivers), and high in the Central Mexican Plateau (>19 species) and southern Mexico/Central America (>13 species). Overall, the fish fauna of eastern North America is very diverse (more so than southern Mexico/Central America and the Central Mexican Plateau region), and probably has been for a long time (Cavender 1986; Wilson and Williams 1992; Near et al. 2003). Both the initial invasion and subsequent diversification of lineages that have transitioned from marine to freshwater habitats may depend on the amount of competition with the incumbent freshwater community (Vermeij and Dudley 2000; Lovejoy et al. 2006; Betancur-R 2010; Yoder et al. 2010; Bloom and Lovejoy 2011, 2012; Betancur-R et al. 2012). We propose that competition with older, more diverse groups has prevented silversides from extensive diversification in North America. In contrast, the Central Mexican Plateau region has a relatively depauperate fish fauna and lacks large predatory fishes (Miller et al. 2005). Our time-calibrated tree also suggests that silversides were present in southern Mexico and northern Central America prior to the formation of the Panamanian Isthmus; thus, silversides were able to diversify before many South America fishes colonized Central America (Albert and Reis 2011). We suggest that in both the Central Mexican Plateau and southern Mexico/Central American there was ample ecological opportunity for diversification due to reduced competition with incumbents, which explains the greater freshwater silverside diversity in these regions (Betancur-R et al. 2012; Bloom and Lovejoy 2012). However, we acknowledge that testing for past competition is difficult, and future studies on past competition in silversides would benefit from explicit tests (i.e., Betancur-R et al. (2012). Our results highlight the usefulness of marine/freshwater sister lineage comparisons for understanding the effect of habitat on lineage diversity.

Conclusion

Our study demonstrates that clades of fishes that include both marine and freshwater members are excellent systems for studying macroevolutionary processes that determine patterns of species richness across oceans and continents. We have shown that transitions from marine to freshwater result in accelerated speciation and extinction rates (and to a lesser extent, net diversification rates), and that these rate differences may help explain the remarkable disparity in species richness between continents and oceans. The greater number of barriers in freshwater habitats relative to marine habitats likely results in more frequent allopatric speciation events. The higher extinction rates in freshwater habitats is likely due to variation in habitat stability and dynamics of living in an open (marine) versus restricted (freshwater) system. We suggest that elevated diversification rates of continental lineages might be a widespread phenomenon that contributing to the uneven distribution of species richness between continents and oceans.

ACKNOWLEDGMENTS

For contributing specimens/tissues or help collecting specimens in the field we thank J. Lyons, N. Mercado-Silva, H. Lopez-Fernandez, D. Phillip, D. Taphorn, E. Holm, E. Lewallen, W. Starnes, M. Sabaj, P. Gesundheidt, G. Orti, M. Cashner, and H. J. Walker. The Smithsonian Tropical Research Institute provided logistical support and help with permits for collecting in Panama. M. Davis offered helpful advice on BEAST analyses. We thank members of the Lovejoy Lab and Lopez-Fernandez lab for constructive comments on this article. This work was supported by AMNH Lerner-Gray and the University of Toronto Centre for Global Change Studies grants to DDB, an NSERC Discovery Grant to NRL, and National Science Foundation DEB 0918073 to KRP.

Ancillary