Temporal evolution of coral reef fishes: global patterns and disparity in isolated locations

Authors

  • Jennifer R. Hodge,

    Corresponding author
    1. School of Marine and Tropical Biology, James Cook University, Townsville, QLD, Australia
    2. Australian Research Council Centre of Excellence for Coral Reef Studies, James Cook University, Townsville, QLD, Australia
    3. Centre for Sustainable Tropical Fisheries and Aquaculture, James Cook University, Townsville, QLD, Australia
    • Correspondence: Jennifer R. Hodge, School of Marine and Tropical Biology, James Cook University, Townsville, QLD 4811, Australia.

      E-mail: jennifer.hodge@my.jcu.edu.au

    Search for more papers by this author
  • Lynne van Herwerden,

    1. School of Marine and Tropical Biology, James Cook University, Townsville, QLD, Australia
    2. Centre for Sustainable Tropical Fisheries and Aquaculture, James Cook University, Townsville, QLD, Australia
    Search for more papers by this author
  • David R. Bellwood

    1. School of Marine and Tropical Biology, James Cook University, Townsville, QLD, Australia
    2. Australian Research Council Centre of Excellence for Coral Reef Studies, James Cook University, Townsville, QLD, Australia
    Search for more papers by this author

Abstract

Aim

Many coral reef regions have a history of isolation and extinction. Our aim was to test whether the disparate evolutionary and biogeographical histories of the world's coral reef regions have significantly impacted temporal patterns of speciation within regions. In essence, do assemblages in peripheral locations contain the youngest coral reef fish species?

Location

Pan-tropical coral reef systems.

Methods

Molecular data (mitochondrial 16S rRNA, 12S rRNA, CO1, cytochrome b; nuclear TMO-4C4, S7 intron 1) were assembled for genera with near-complete taxon sampling (minimum 70% nominal species) from four major coral reef fish families (Chaetodontidae, Labridae, Pomacanthidae and Serranidae). This was combined with fossil data to simultaneously infer the phylogeny and estimate species' divergence times. species' distributions were quantified using IUCN maps and the ages of species with different biogeographical extents were compared. Model fitting was used to compare the distribution of species' ages across the whole phylogeny with age distributions of species restricted to the Red Sea and Hawaiian Islands.

Results

Temporal patterns of coral reef fish divergence were similar among major marine realms and regions. However, notable differences were recorded between the Red Sea and Hawaiian Islands. Red Sea endemics have diverged consistently throughout the last 16 Myr, whereas endemic species colonized the Hawaiian Islands in two distinct waves (0–3 Ma and 8–12 Ma). Differences in the proportions of allopatric and sympatric sister-species between Red Sea and Hawaiian endemics were also detected.

Main conclusions

Despite differing geological histories, marine realms and regions have all experienced comparable and relatively recent divergences of extant coral reef fish species. Differences in age distributions and spatial relationships of endemic species in the Red Sea and Hawaiian Islands suggest that markedly different processes have shaped patterns of diversification in these peripherally isolated locations.

Introduction

Coral reef fishes are exceptionally diverse throughout the world's tropical oceans. To understand the foundation of such high biodiversity, and to help predict the response of contemporary patterns of diversity to future environmental change, we study the distribution of organisms through space and time. Over the past 150 years, many authors have described spatial patterns of marine diversity (reviewed by Bellwood et al., 2012; Briggs & Bowen, 2012; Kulbicki et al., 2013). However, the molecular techniques and analytical tools used to assess distributional patterns through evolutionary time have only recently been developed. These tools allow us to examine the impacts of differing geological histories on the age structure of fauna from distinct biogeographical areas, or within isolated locations.

Spatial patterns of diversity are generally described by the delineation of areas with unique biotas. Biogeographical divisions of marine environments have been based on the concentration of endemic species (Ekman, 1953; Briggs, 1974; Briggs & Bowen, 2012), shared biotic and environmental characteristics (Longhurst, 1998; Spalding et al., 2007), or quantitative assessments of community composition (Floeter et al., 2008; Kulbicki et al., 2013). These methods have produced varying degrees of division and identified areas that often differ in boundary placement and scale. However, most workers have recognized the importance of historical isolation in separating two major biogeographical realms: the Indo-Pacific and the Atlantic. Each realm has experienced a distinctive history that has shaped its constituent fauna – from ongoing speciation as part of the expanding biodiversity hotspot that has developed in the central Indo-Pacific since the Miocene (Renema et al., 2008; Cowman & Bellwood, 2013a), to substantial loss of many marine taxa in the Atlantic during a period of faunal turnover in the Plio-Pleistocene (Bellwood, 1997; O'Dea et al., 2007). Given such different evolutionary histories, we may expect to find distinct realm-specific temporal patterns in the evolution of coral reef fishes. Our first aim was to investigate these realm-specific patterns.

With increased taxon sampling and increasingly congruent molecular sequence data, we are also able to evaluate the predictions of the four main models of diversification (centre of origin, overlap, accumulation or survival) used to explain the hotspot of biodiversity in the central Indo-Pacific [referred to as the Indo-Australian Archipelago (IAA) or Coral Triangle] (the ‘centre of’ hypotheses are reviewed briefly in Palumbi, 1997; Barber & Bellwood, 2005; and in detail in Bellwood et al., 2012). There are two contrasting scenarios. If the IAA were a centre of accumulation or overlap of species' ranges (Gaither & Rocha, 2013), we would expect to find younger species in peripheral locations outside the IAA. These hypotheses suggest that species arise in relatively isolated, peripheral locations (such as the island arcs in the West Pacific) and remain there as endemics until they disperse to the IAA. Conversely, if the IAA were the centre of origin, we would expect to find the youngest species within this region. The centre of survival hypothesis suggests that species can arise anywhere, but they survive better in the IAA.

If species arise at the same rate across all biogeographical regions, then there should be no pattern in the distribution of endemics. But if species arise in proportion to regional species richness, then most endemics would be located in the IAA. Consequently, the geographical distribution of young endemics can help evaluate these alternative hypotheses. Our second aim, therefore, was to examine the extent of peripheral speciation as a mode of diversification in reef fishes. To specifically examine peripheral locations, we compared the ages of endemic species from two disparate locations with high coral reef fish endemism, the Red Sea and the Hawaiian Islands (Allen, 2008). If either location has been a significant area of recent species origination, we would expect endemic species to be younger than species in the larger adjacent regions (the Western Indian and Central Pacific regions, respectively).

To address our aims, we constructed the most comprehensive assemblage of coral reef fish genera with near-complete taxon sampling to date, and combined it with available fossil data to simultaneously infer the phylogeny and estimate divergence times of species from four major coral reef fish families. We used this phylogenetic hypothesis in conjunction with recent distributional data to investigate how the age structure of coral reef fish species varies in response to differing geological histories among marine biogeographical areas, and test the predictions of the ‘centre of’ hypotheses. Our specific questions were: (1) How old are coral reef fish species and do their ages differ among major marine realms or regions? (2) Do peripheral, isolated locations of high endemism support greater numbers of young endemics; are peripheral locations a source of ‘new’ species?

Materials and Methods

Data selection

We obtained sequence data from GenBank for those coral reef fish families with distribution maps available from the IUCN Red List spatial database (IUCN, 2011). At the time of data collection this included genera belonging to the families Acanthuridae, Chaetodontidae, Labridae, Pomacanthidae and Serranidae. Species designations were based on the IUCN Red List (IUCN, 2011) and FishBase (http://www.fishbase.org/). Only loci with the most coverage across all families were considered. Among the mitochondrial loci, these included 16S rRNA, 12S rRNA, cytochrome c oxidase subunit I (CO1) and cytochrome b (CYTB). Nuclear loci included TMO-4C4 and S7 intron 1. Sequence coverage was also assessed within each of the reef fish families to maximize taxon sampling within genera and minimize missing sequence data within the alignment. Only those genera with a minimum of 70% of constituent species with sequences available for any of the four mitochondrial loci, and at least 50% sequence coverage across all six loci were considered. This resulted in the inclusion of 53 genera within four coral reef fish families: Chaetodontidae, Labridae, Pomacanthidae and Serranidae (see Appendix S1: Table S1 in the Supporting Information). Sequences were also included for taxa used to root the phylogeny (Opsanus pardus and Porichthys notatus) and to provide additional nodes for fossil calibration (Appendix S1: Tables S1 & S2).

Sequences were aligned in Geneious Pro 6.1.2 (Biomatters, Auckland, New Zealand; http://www.geneious.com/) using default settings for each locus. All alignments were manually adjusted through the insertion or deletion of gaps, and trimmed to minimize the amount of missing data. PartitionFinder 1.0.1 (Lanfear et al., 2012) was used to simultaneously select an appropriate partitioning scheme for the concatenated alignment and the best fitting models of molecular evolution for each locus based on the Bayesian information criterion (BIC) (Schwarz, 1978) (Table 1).

Table 1. Partition scheme and models of evolution selected by PartitionFinder under the Bayesian information criterion (BIC). The data set consisted of molecular sequences for 312 percomorph fishes found throughout the tropical Atlantic and Indo-Pacific. Γ represents the gamma shape parameter
Data setBase pairsPartitionBIC model
16S 5691TVM+I+Γ
12S308
CO1 477
CYTB 92
TMO-4C4 4342TrN+I+Γ
S7 6113TrN+I+Γ
Total2491

Age estimation

A time-calibrated phylogeny was constructed based on partitioned Bayesian analyses in beast 1.7.5 (Drummond & Rambaut, 2007) using nine fossil calibrations (Table S2). Partitioning and models of molecular evolution were specified according to those identified by PartitionFinder (Lanfear et al., 2012). Divergence times were estimated under a relaxed uncorrelated lognormal clock model (Drummond et al., 2006) following preliminary analyses that rejected a strict molecular clock for all partitions. The birth–death process was specified as the tree prior to account for speciation and extinction (Gernhard, 2008), and starting trees were randomly generated for each run. Time calibrations were made based on fossil evidence used in previous phylogenetic studies of fishes (Cowman et al., 2009; Near et al., 2012) (Table S2). Posterior samples from six independent Markov chain Monte Carlo (MCMC) analyses, each with 40 million generations and sampling every 2000th generation, were assessed for convergence and appropriate burn-in using Tracer 1.5 (Rambaut & Drummond, 2007). Tree files were combined using LogCombiner 1.6.1 (Drummond & Rambaut, 2007) following the removal of 20% burn-in and resampling every 4000 states. A maximum clade credibility tree was constructed using TreeAnnotator 1.6.1 (Drummond & Rambaut, 2007) to display median ages and 95% highest posterior density (HPD) intervals for each node.

Biogeographical analysis

Contemporary spatial data were compiled for all species included in the phylogenetic analysis (except the ingroup species Aprops bilinearis, Grammistes sexlineatus, Niphon spinosus and Zalanthias kelloggi, which had not been assessed by the IUCN Red List at the time of sampling; IUCN, 2011). Spatial analyses were conducted in grass (GRASS Development Team, 2011) and qgis (QGIS Development Team, 2012). We followed Kulbicki et al. (2013) to classify species' distributions by realm: Indo-Pacific and Atlantic; by region: Western Indian (WI), Central Indo-Pacific (CIP), Central Pacific (CP), Eastern Tropical Pacific (ETP), Western Atlantic (WA) and Eastern Atlantic (EA); and to identify those species with distributions restricted to the Hawaiian Islands and the Red Sea (Table 2). Kulbicki et al. (2013) quantitatively delineated biogeographical patterns based on species composition. They described the ETP as both a realm and region; however, we consider it only as a region outside of the Indo-Pacific and Atlantic realms.

Table 2. Sample sizes and resulting mean species age for the two approaches used to compare the ages of coral reef fish species between marine biogeographical realms and regions. (a) The full-phylogeny approach that considered the ages of all extant species, excluding fossil calibration and outgroup species. (b) The sister-species approach that considered the age of only the species with the smallest geographical range (per sister-species pair, excluding monotypic genera). Contemporary spatial data was used to classify species by realm and region, and to identify those species endemic to the Red Sea and Hawaiian Islands (IUCN, 2011) following the biogeographical delineations of Kulbicki et al. (2013). Regions included the: Western Indian (WI), Central-Indo Pacific (CIP), Central Pacific (CP), Eastern Tropical Pacific (ETP), Western Atlantic (WA) and Eastern Atlantic (EA). Numbers in parentheses corresponding to the WI and CP indicate the number of species from that region endemic to the Red Sea and Hawaiian Island provinces, respectively. Region-restricted species combined with widespread species (WS) in the Indo-Pacific and Atlantic realms to give the total number of species for each realm. Bootstrap resampling was performed for all biogeographical areas with more than two constituent species. Asterisks indicate a significant difference (< 0.05) in the mean age of species compared with 1000 resampled permutations
RealmNo. extant speciesMean (Ma)RegionNo. extant speciesMean (Ma)
Age95% HPDAge95% HPD
(a) Full-phylogeny approach
Indo-Pacific2045.53.1–8.4WI29 (16)3.9 (4.2)2.0–6.2 (2.2–6.8)
CIP62.91.4–4.9
CP18 (9)5.3 (4.8)2.9–8.6 (2.7–7.4)
WS1515.93.4–9.0
ETP224.62.7–6.9
Atlantic575.63.3–8.5WA346.33.8–9.2
EA215.02.6–7.8
WS21.90.6–3.9
(b) Sister-species approach
Indo-Pacific663.01.3–5.3WI15 (9)2.6 (3.0)1.1–4.7 (1.2–5.3)
CIP20.50.1–1.2
CP9 (4)2.6 (1.2)0.9–4.9 (0.4–2.2)
WS403.3*1.4*–5.9*
ETP82.10.9–3.8
Atlantic172.91.3–5.0WA102.20.9–4.0
EA64.42.1–7.0
WS11.20.3–2.4

Age comparison

Phylogenetic reconstruction is based on a bifurcating process that produces an exponential distribution of age estimates, we therefore log-transformed the data to normalize the distribution. We used two approaches for defining species age with the aim of comparing ages between realms and regions. First, we considered the estimated divergence time for all nodes in the phylogeny subtended by extant species as reflective of species age, excluding fossil calibration and outgroup species. We refer to this as the full-phylogeny approach where the advantage was achieving a large sample size for comparison among biogeographical areas. There were several limitations to this approach including the covariance of ages shared among sister-species, which may have a levelling effect on any potential differences in age structure if sister-species are found in adjacent, or different, areas.

Another limitation of the full-phylogeny approach was the impact of missing taxa on the estimation of species' ages. Phylogenetic methods are only able to estimate the age of the most recent common ancestor between two species if both species are present in the analysis. Therefore, taxa missing from the phylogeny, whether extinct or extant, will cause the ages of sister taxa to be overestimated. Genera were selected with the aim of minimizing the amount of missing extant species; however, the proportion of missing extant species in our analysis varied by genus (see Appendix S1: Table S3). Extinct species will have a greater impact on the age estimation of older lineages if the rate of extinction through time has remained relatively constant (Rabosky & Lovette, 2008).

To minimize the effects of missing taxa on species age estimation our second approach considered only sister-species pairs, excluding monotypic genera. We refer to this as the sister-species approach where we assigned estimated divergence times to the species with the smallest geographical range and did not include its sister in our analysis. This approach eliminated the levelling effects of covarying sister-species' ages and minimized the likelihood of underestimating the ages of widespread species that may have persisted through successive peripheral divergence events (i.e. peripheral budding; Hodge et al., 2012).

For both approaches, we used bootstrap resampling to test whether the mean age of species from each biogeographical area differed from a random distribution of mean ages. Bootstrap resampling was used because the data did not meet the assumptions of independence required by parametric analyses. Furthermore, the bootstrap analyses could be applied to both age-estimation approaches. For each realm, region and province we sampled the corresponding number of species (for all biogeographical areas with more than two constituent species) randomly without replacement from the full phylogeny and from the full pool of resultant species following the sister-species approach. We also compared the mean ages of Red Sea and Hawaiian Island endemic species with corresponding random samples of species restricted to the WI and CP regions, respectively, to resolve whether endemic species are young relative to non-endemic species from the same region and thus potentially mark locations of species origination. Random sampling was repeated 1000 times for each case. We performed a two-tailed significance test where the observed mean age was considered significantly different from random if it was outside of the central 95% of the resampled distribution. We repeated the same randomizations using the mean upper and lower 95% HPD intervals of species age.

Peripheral speciation

To further explore the frequency of peripheral speciation events we fitted a model to the distribution of untransformed species' ages across the entire phylogeny and compared it with models fitted to distributions of untransformed species' ages for both Red Sea and Hawaiian Island endemics. All models were fitted to density distributions of 2-Myr intervals. Finally, to gain insight into potential processes underlying endemism and establish patterns of geographical range distribution among peripherally isolated species, we determined whether endemic species are contemporarily allopatric or sympatric with regard to their respective sister-species (Barraclough & Vogler, 2000) or clade (using independent contrasts sensu Felsenstein, 1985). All statistical analyses were performed using R (R Development Core Team, 2011).

Results

Phylogenetic analysis

Six independent MCMC analyses resulted in a well-resolved chronogram of species from the coral reef fish families Chaetodontidae, Labridae, Pomacanthidae and Serranidae (Fig. 1, Appendix S1: Table S4). The maximum clade credibility chronogram was compiled from 48,003 post-burn-in trees (48,003,000 generations). Log files showed high effective sample sizes (posterior ESS values > 200 for six combined analyses), which indicated valid estimates based on independent samples from the posterior distribution of the Markov chain.

Figure 1.

Time-calibrated phylogeny of 312 percomorph fishes compiled from post-burn-in topologies of six independent Bayesian Markov chain Monte Carlo (MCMC) analyses (40 × 106 generations per run) using nine fossil calibration points implemented in beast. Molecular data included four mitochondrial loci (16S rRNA, 12S rRNA, CO1, cytochrome b) and two nuclear loci (TMO-4C4, S7 intron 1); see Table 1 for partitioning scheme. Nodes represent median ages from a maximum clade-credibility tree [values provided in Appendix S1: Table S4 along with corresponding 95% highest posterior density (HPD) intervals]. Branch colours subtending nodes indicate posterior probability for the node: black > 75%; blue 50–74%; and red < 49%. • indicates nodes calibrated with priors based on fossil data (Table S2). Time scale is in millions of years before present.

Age comparisons

Species with distributions confined to the Indo-Pacific and Atlantic realms had strikingly similar mean ages and variances (using both the full-phylogeny and sister-species approach; Fig. 2, Table 2). Mean species age was expectedly younger under the sister-species approach (2.9 Ma vs. 5.4 Ma for species sampled using the full-phylogeny approach), but with comparably little variability between realms. We found no significant differences in the mean ages of species from the Indo-Pacific and Atlantic realms when compared with distributions of means from 1000 resampled permutations, for both the full-phylogeny and sister-species approach.

Figure 2.

Biogeographical variation in the ages of coral reef fish species. (a) Map showing two marine realms, the Indo-Pacific and Atlantic, and the regions within each realm differentiated by colour: Western Indian (WI) (red), Central Indo-Pacific (CIP) (orange), Central Pacific (CP) (yellow), Eastern Tropical Pacific (ETP) (green), Western Atlantic (WA) (light blue) and Eastern Atlantic (EA) (dark blue). Areas follow those outlined by Kulbicki et al. (2013). Two peripherally isolated provinces within the Indo-Pacific realm, the Hawaiian Islands and the Red Sea, are outlined in black. The map uses a Behrmann projection. Below the map are the results of age comparisons of coral reef fish species among marine realms and regions using (b) a full-phylogeny approach, and (c) a sister-species approach. Mean species age (dark grey horizontal bars) and 95% highest posterior density (HPD) intervals (light grey shading) are displayed for species in the Indo-Pacific and Atlantic realms, as well as for species restricted to regions (corresponding coloured circles). Black circles indicate mean ages (and corresponding 95% HPD intervals) of non-region-restricted species for both the Indo-Pacific and Atlantic realms. Corresponding coloured squares represent median species age. Asterisks indicate a significant difference (< 0.05) in the mean age of species compared with 1000 resampled permutations. Sample sizes for each biogeographical area are listed in Table 2.

We also detected little variation in mean species age among regions under both approaches (Fig. 2, Table 2). We found no significant differences in mean species age compared with the randomly resampled distributions for all six regions using the full-phylogeny approach, and for all five bootstrapped regions using the sister-species approach (note that we did not perform bootstrapping for the CIP region using the sister-species approach because it did not have more than two constituent species). Overall, both approaches produced similar patterns (Fig. 2, Table 2). The most notable distinction between the two approaches was observed in the mean age of widespread species in the Indo-Pacific realm. Under the full-phylogeny approach, the mean age did not differ significantly from the bootstrapped distribution, whereas the mean age was significantly older than the central 95% of the resampled distribution using the sister-species approach. We also detected a shift in the age structure of species in the Atlantic realm along with a slight increase in regional variation of mean species age using the sister-species approach. In general, the sister-species approach produced ages 0.6–2.7 Myr younger than the full-phylogeny approach, except for the WA region where the mean age was 4.1 Myr younger using the sister-species approach.

Species endemic to the Red Sea and Hawaiian Islands had similar mean ages to those restricted to the WI and CP regions, respectively (Table 2). The two oldest species in the WI region are endemic to the Red Sea (oldest species: Larabicus quadrilineatus; median age: 14.7 Ma; 95% HPD: 7.4–23.8 Ma). However, Hawaiian endemic species did not include the oldest species in the region. We found no significant difference in the mean age of species endemic to the Red Sea or Hawaiian Islands when compared with the bootstrapped distributions of species across the full phylogeny, or when compared with the bootstrapped distributions of species restricted to the corresponding regions. These results persisted under the sister-species approach.

Peripheral speciation

Phylogeny-wide age estimates were randomly distributed under a negative binomial model (= 0.2, prob = 3.2; = 293; d.f. = 1, 292; χ2 = 2.72; = 0.10) (Fig. 3a). However, Red Sea endemic species have diverged steadily through time and fit a Poisson model (λ = 1; = 16; d.f. = 1, 15; χ2 = 0.09; = 0.76) (Fig. 3b), while Hawaiian endemic divergences do not fit either a negative binomial or a Poisson model. Instead, the age estimates of Hawaiian endemic species fit a bimodal distribution with distinct peaks of 0–3 Ma and 8–12 Ma (Fig. 3c), which conforms to a mixture of two normal distributions (= 20.58; > 2 required for clear separation; Ashman et al., 1994). Furthermore, the mean 95% HPD intervals of both peaks do not overlap (95% HPD early peak: 5.7–13.9 Ma; late peak: 0.4–2.3 Ma) (Fig. 4). The separation of Hawaiian Island endemic species' divergences is apparent despite ongoing divergence within the CP throughout this period. It also holds true for multiple species within genera. For example, two of the three endemic Chaetodon species arose in the late peak (0–3 Ma), while the third endemic Chaetodon species arose in the early peak (8–12 Ma) (Fig. 4).

Figure 3.

Frequency distributions of age estimates for coral reef fish species plotted as the proportion of total species on the y-axis (left side) and as a frequency (right side). (a) Age estimates for all species sampled in the phylogeny (full-phylogeny approach), which are found throughout the tropical Atlantic and Indo-Pacific. This age distribution fits a negative binomial model (= 0.2, prob = 3.2; = 293; d.f. = 1, 292; χ2 = 2.72; = 0.10), represented by the black line. (b) The age distribution of species endemic to the Red Sea (grey bars) fits a Poisson model (λ = 1; = 16; d.f. = 1, 15; χ2 = 0.09; = 0.76), represented by the black line; asterisks denote the distribution of species restricted to the Western Indian (WI) region. (c) The age distribution of species endemic to the Hawaiian Islands (grey bars) is bimodal (Ashman's = 20.58); asterisks denote the distribution of species restricted to the Central Pacific (CP) region. Note that the x-axis scale is comparable for (b) and (c), but differs for (a).

Figure 4.

The ages of Red Sea and Hawaiian endemic coral reef fish species [circles and squares; 95% highest posterior density (HPD) intervals indicated by horizontal black lines]. Circles represent endemic species sister to a single species; squares represent endemic species sister to a clade. White shapes indicate endemic species that have allopatric distributions with their closest sister-species or clade; black shapes represent endemic species that have sympatric distributions (i.e. degree of sympatry above zero) with their closest sister-species or clade. Degree of sympatry was calculated following Barraclough & Vogler (2000) and using independent contrasts (Felsenstein, 1985) for endemic species sister to a clade. The ages of Hawaiian endemics conforms to a mixture of two normal distributions according to Ashman's D statistic (= 20.58; > 2 required for clear separation; Ashman et al., 1994). Dashed vertical lines show the mean of the two separate periods of divergence and the grey shaded area indicates the mean 95% HPD intervals, which do not overlap (95% HPD early peak: 5.7–13.9 Ma; late peak: 0.4–2.3 Ma).

To explore peripheral speciation processes we determined whether species endemic to the Red Sea and Hawaiian Islands are contemporarily allopatric or sympatric relative to their sister taxon. Endemics in these two regions differed in the proportion of allopatrically and sympatrically distributed sister-species (Fig. 4). Seven of the nine Hawaiian endemic species (78%) have allopatric distributions with their sister-species or clade, while the remaining two species have sympatric distributions (22%). Six of the nine endemic species are sister to a single species and the remaining three are sister to a clade. Interestingly, both sympatric species are sister to a clade and were among the early wave of speciation in the Hawaiian Islands (8–12 Ma) with an average degree of overlap less than 10%. The average degree of sympatry for Hawaiian endemic species was 1%, compared with an average of 19% sympatry among species restricted to the CP region (where four of nine species were sympatric). In contrast, half of the Red Sea endemic species (eight of 16 species) have allopatric distributions, and half have sympatric distributions. Of the 16 endemic species, 11 are sister to a single species, and the remaining five are sister to a clade. Species that are sister to a clade are distributed throughout the age range of endemic Red Sea species, as are allopatric and sympatric species. The average degree of sympatry for Red Sea endemics was 23%, similar to the average degree of sympatry recorded for species in the WI region (19%, eight of 13 species were sympatric), but substantially larger than the average degree of sympatry among Hawaiian endemics.

Discussion

Despite differing geological and evolutionary histories between the Indo-Pacific and Atlantic biogeographical realms, our results show limited variation in the ages of their constituent coral reef fish faunas. The CIP and its peripheral regions have all experienced recent divergence events, with no detectable difference in the mean ages of reef fish species among regions in the Indo-Pacific. Atlantic regions have also experienced recent divergence events, with mean ages similar to those of the Indo-Pacific. Furthermore, ages of endemic species do not differ from more widespread species. Therefore, our results do not support the indiscriminate use of endemic species as markers of species origination. Interesting patterns of diversification in isolated locations provide insights into the processes underlying endemism and peripheral speciation. Specifically, we report notable differences between the Red Sea and the Hawaiian Islands in the timing of divergence events and in patterns of contemporary geographical distribution of endemic species.

Age comparisons

The Indo-Pacific and Atlantic realms are undoubtedly characterized by unique assemblages of coral reef fishes (Briggs & Bowen, 2012; Kulbicki et al., 2013). While some species have attained circumtropical distributions, most species are restricted to a particular realm and share the influence of its historical periods of isolation (Kulbicki et al., 2013). Vicariance events such as the closure of the Tethys seaway (Steininger & Rögl, 1984) and the rise of the Isthmus of Panama (Coates & Obando, 1996) have produced diffuse signals of vicariance among reef fish lineages, while soft hydrological barriers to dispersal can result in tightly concordant vicariance (Cowman & Bellwood, 2013b). Furthermore, each region has a markedly different amount of space for reef fishes to occupy, which has varied through time and may have influenced the rate of species divergence or extinction (Bellwood & Wainwright, 2002; Renema et al., 2008). Despite the potential for temporally diffuse or concentrated vicariance events, and variable habitable area within each realm, we found few significant differences among the ages of constituent species using two approaches for age comparison. On average, recent divergence of extant species has occurred throughout the past 1–5 Myr in the Indo-Pacific and Atlantic realms, suggesting similar timing in the divergence of reef fishes, despite different geological histories and different contemporary patterns of biodiversity.

Differences between the two approaches in the ages of widespread species in the Indo-Pacific realm suggest that extinction may have resulted in substantial overestimation of species' ages using the full-phylogeny approach, whereas the larger magnitude of change in mean species age in the WA region, relative to other regions, suggests that extinction may have had the greatest influence on the ages of extant species in this region. The Atlantic has experienced a substantial loss of marine taxa during a period of faunal turnover in the Plio-Pleistocene (Bellwood, 1997; O'Dea et al., 2007), which may explain the differences we observed when comparing full-phylogeny and sister-species approaches.

In the Atlantic, widespread species are younger than their region-restricted counterparts. However, we found just two species with distributions spanning the Eastern and Western Atlantic regions. The low number of widespread species in the Atlantic may be due in part to the relatively low overall coral reef fish biodiversity of this realm (Bellwood, 1997; Kulbicki et al., 2013), but it may also indicate that regional spatial structure is an important characteristic of Atlantic species (Bender et al., 2013). In comparison to the Indo-Pacific, the Atlantic has fewer centrally located islands available to facilitate range expansion, which may produce more defined spatial structure among its faunal constituents.

We detected more variation in the ages of reef fishes between regions than among realms, with younger ages in the CIP than in the WI and CP. Younger ages of reef fish species in the CIP, although not significantly different from random permutations, lends some support to the centre of origin or centre of survival hypotheses. Recent models of coral reef fish evolution and dispersal over the last 65 Myr demonstrate that the IAA (located within the greater CIP region) has played a number of different roles, supporting the accumulation, survival, origination and export of species (Cowman & Bellwood, 2013a). More specifically, models suggest that since the Miocene (23 Ma) the IAA has been characterized by exceptionally high rates of species origination (Cowman & Bellwood, 2013a). This may explain the slightly younger species' ages we obtained for the CIP relative to adjacent regions that, according to the models, were colonized by lineages that originated in the IAA during this time (Cowman & Bellwood, 2013a).

If vicariance has played a major role in recent species diversification, it would have a neutralizing effect on any potential patterns between vicariant regions when comparing species' ages using the full-phylogeny approach. For example, the rise of the Isthmus of Panama produced geminate species pairs with a shared age that are now distributed either side of the barrier. Therefore, when comparing the ages of species from the WA and the ETP, the ages of geminate species pairs are considered for both regions and neutralize any differences in age structure between them. However, patterns of relative age between realms and regions remained constant, with the exception of species in the WA region, when we accounted for such covariance of ages using the sister-species approach.

Our methods considered only extant species and their lineages, and therefore were not capable of resolving historical evolutionary differences that may have distinguished biogeographical areas. The protocols we applied to maximize sampling of extant species meant that generic sampling achieved within families was limited and we were not able to resolve deeper splits in the phylogeny with confidence, precluding an assessment of historical differences among lineages. Instead, we have focused on recent speciation events because it is unlikely that many older species have survived to the present day. If older species have survived and also given rise to other species through time (i.e. peripheral budding sensu Hodge et al., 2012), their true age may be masked by recent speciation events. Thus, historical signals are largely overwhelmed by recent speciation. Different geological and evolutionary histories among regions have probably shaped lineages, but speciation has largely been shaped by events in the past 1–5 Myr.

Peripheral speciation

Endemic species in the Red Sea and Hawaii displayed different age distributions when compared with each other and with the underlying age distribution of the full phylogeny, suggesting that distinctive processes of diversification have operated at these peripherally isolated locations. Our topological and chronological hypotheses of Red Sea and Hawaiian endemic species agree with previous phylogenetic hypotheses for Chaetodon (Fessler & Westneat, 2007; Bellwood et al., 2010; Craig et al., 2010), Anampses (Hodge et al., 2012), Chlorurus (Choat et al., 2012), Larabicus (Westneat & Alfaro, 2005; Cowman et al., 2009; Kazancıoğlu et al., 2009), Scarus (Choat et al., 2012) and Thalassoma (Bernardi et al., 2004), offering additional confidence in the chronogram.

Red Sea endemics appear to have arisen steadily throughout the last 16 Myr, roughly 10 Myr after the sea first appeared (Bosworth et al., 2005). They include the oldest extant lineage in the WI region, Larabicus quadrilineatus, which is estimated to have diverged 14.7 Ma, around the time when the Red Sea was becoming increasingly isolated from the Mediterranean (Bosworth et al., 2005) and the Arabian hotspot was dwindling (Renema et al., 2008). In the recent geological past, the Red Sea has experienced volatile changes in temperature and salinity (Biton et al., 2008). The effects of these environmental fluctuations have reportedly caused mass extirpation of marine organisms including planktonic foraminifera (Hemleben et al., 1996). Our results, as well as other phylogeographical studies of reef fishes (DiBattista et al., 2013), suggest that the Red Sea, or the adjacent Gulf of Aden, has sustained coral reef fish lineages (and presumably coral reefs) throughout these environmental fluctuations. Older endemic lineages, such as L. quadrilineatus, are likely to have survived outside of the Red Sea during extreme environmental periods and subsequently re-invaded when the Red Sea opened up to the WI region c. 5 Ma and the environment became suitable for the maintenance of coral reefs (Siddall et al., 2003; Bosworth et al., 2005). The majority of Red Sea endemics (75% of those studied herein) diverged after this time. The continuity of the continental shelf to the east and south of the Red Sea, in combination with its relative close proximity to the IAA, is likely to have facilitated the ongoing divergence of lineages throughout the last 16 Myr.

In contrast to the Red Sea, the Hawaiian Archipelago is located in the central Pacific Ocean and is part of the larger Hawaiian–Emperor seamount chain, a series of volcanic islands and atolls separated by oceanic channels. Our data suggest that colonization of the Hawaiian Islands has occurred independently for multiple species belonging to the genera Chaetodon, Anampses and Thalassoma. This is consistent with previous phylogenetic hypotheses (Bernardi et al., 2004; Fessler & Westneat, 2007; Craig et al., 2010; Hodge et al., 2012). Multiple models of diversification have probably led to the divergence of Hawaiian endemics including, but not limited to, successive colonization/division and peripheral budding (sensu Hodge et al., 2012). For example, Craig et al. (2010) proposed that a closely related Chaetodon species complex containing C. punctatofasciatus, C. pelewensis and C. multicinctus, originated in the western Pacific, spread through the South Pacific Islands and finally colonized the Hawaiian Islands. With our inclusion of the closely related WI species, C. guttatissimus, we find that the diversification of species in this clade could fit a successive division or colonization model, with initial separation between the Indian Ocean and the CIP/CP, followed by separation between the CIP/CP and the Hawaiian Islands. However, the region of origination of this lineage remains unclear.

In contrast to the ongoing cladogenesis in the Red Sea, we found evidence for two distinct waves of divergence among Hawaiian endemics. The two waves of divergence (0–3 Ma and 8–12 Ma) occurred either side of a broad period of increased primary productivity (Dickens & Owen, 1999) that coincided with increased cladogenesis for a wide range of taxa in the Indo-Pacific, including reef fishes, between 3.5 and 9.0 Ma (Renema et al., 2008; Reid et al., 2010; Cowman & Bellwood, 2013a,b). The first wave coincides with the late Miocene–Pliocene (9–12 Ma) when deep-water circulation reorganization occurred as a result of the reduction in deep-water exchange between the Atlantic and Pacific Oceans through the Panamanian gateway prior to the emergence of the isthmus (Lyle et al., 1995). These changes in deep-water circulation caused disruption to large-scale ocean circulation patterns (Butzin et al., 2011). Further evidence suggests that atmospheric and oceanic circulation intensified about 10 Ma (Rea & Bloomstine, 1986) and during the glacial periods of the past 1.2 Myr (Hall et al., 2001). These changes in ocean circulation and intensity may have produced more favourable conditions for founder populations of reef fishes to reach the Hawaiian Islands and establish themselves, ultimately leading to the divergence of peripatric populations and the formation of endemic species.

Different proportions of allopatric and sympatric sister-species among Red Sea and Hawaiian endemics provides additional support for the operation of distinctive processes of diversification at these peripherally isolated locations. Red Sea endemics have equal proportions of allopatrically and sympatrically distributed sister-species, and approximately one-third appeared as sister to a clade. Secondary endemism, where a primary endemic gives rise to one or more subsequent endemics (Rotondo et al., 1981), appears likely to have operated in the Red Sea province given the sympatric distribution of the well-supported sister-species pair: Scarus persicus and Scarus ferrugineus, both Red Sea/Persian Gulf endemics (Choat et al., 2012). Our results, in combination with knowledge of the dynamic geological and environmental past of the Red Sea, suggest that a number of speciation modes may have operated through time and that both allopatric and sympatric speciation are likely to have played a role in generating Red Sea endemics.

While a number of speciation modes may also have led to the diversification of Hawaiian endemics, our results reveal potential key differences between the two isolated provinces. As in the Red Sea, one-third of Hawaiian endemic species are sister to a clade. However, allopatric distributions constitute the bulk (78%) of endemic species' distributions in the Hawaiian Islands. With such a low level of sympatry it seems unlikely that either sympatric speciation or secondary endemism has been important in the evolution of Hawaiian endemic reef fishes. Rather, allopatric speciation, probably in the form of peripatric speciation, appears to have been the dominant mode in generating Hawaiian reef fish endemics. The topological relationships and estimated ages of sympatric endemic species (between 8 and 12 Ma), combined with the low levels of average overlap (< 10%), suggest that peripheral budding (sensu Hodge et al., 2012) may be a key model under which Hawaiian endemic reef fishes have diverged.

Conclusions

Both marine realms and all six biogeographical regions show similar patterns of recent species divergence. Two peripherally isolated locations with high levels of endemism, the Hawaiian Islands and the Red Sea, show contrasting patterns. The age structure of Red Sea endemics is comparable to that of the larger WI region, with a steady increase in species over the last 16 Myr. In contrast, the Hawaiian Islands showed two distinct periods of divergence: 0–3 Ma and 8–12 Ma. Spatial distributions of endemic species relative to their closest sister taxon reveal that allopatric and sympatric speciation are likely to be important processes in the divergence of endemic species, with peripatry especially important in the divergence of Hawaiian Island endemics. Taken together, the differences in age structure and spatial patterns between Red Sea and Hawaiian Island endemic species point to markedly different processes of diversification at two peripherally isolated locations.

Acknowledgements

We thank G. Bernardi and K. Ma for the provision of additional sequence data; P. Cowman, J. Welsh and S. Staples for comments on previous versions of the manuscript; H. Choat for stimulating discussion; members of the Reef Fish lab for support and encouragement; and M. Dawson and three anonymous referees for thoughtful comments and suggestions. This work was funded by the Australian Research Council (D.R.B.).

Biosketches

Jennifer Hodge is a PhD candidate in the School of Marine and Tropical Biology at James Cook University, Townsville. She is broadly interested in the evolution and biogeography of coral reef fishes. This paper forms part of her thesis.

Lynne van Herwerden is a Senior Lecturer of Marine Biology at James Cook University. Her research focuses on the evolution and adaptation of coral reef fishes. The aims are to understand connectivity among populations and species and to explore how that structure relates to environmental change.

David Bellwood is a Professor of Marine Biology at James Cook University. His research focuses on the evolution and ecology of coral reef fishes. The aims are to understand the processes that generate and maintain coral reef fish biodiversity and to explore the consequences of variation in biodiversity for ecosystem function.

Author contributions: J.R.H and D.R.B. conceived the ideas; J.R.H. compiled and analysed the data; and all authors contributed to the writing and intellectual discussion.

Ancillary