Community assembly and diversification in Indo-Pacific coral reef fishes

Theories of species coexistence have played a central role in ecology and evolutionary studies of the origin and maintenance of biodiversity in highly diverse communities. The concept of niche and associated theories predict that competition for available ecological space leads to a ceiling in species richness that influences further diversification patterns. By contrast, the neutral theory supports that speciation is stochastic and diversity independent. We examined the phylogenetic community structure and diversification rates in three families and 14 sites within coral reef fish communities from the Indian and Pacific oceans. Using the phylogenetic relationships among 157 species estimated with 2300 bp of mitochondrial DNA, we tested predictions in terms of species coexistence from the neutral and niche theories. At the regional scale, our findings suggest that phylogenetic community structure shifts during community assembly to a pattern of dispersion as a consequence of allopatric speciation in recent times but overall, variations in diversification rates did not relate with sea level changes. At the local scale, the phylogenetic community structure is consistent with a neutral model of community assembly since no departure from a random sorting of species was observed. The present results support a neutral model of community assembly as a consequence of the stochastic and unpredictable nature of coral reefs favoring generalist and sedentary species competing for living space rather than trophic resources. As a consequence, the observed decrease in diversification rates may be seen as the result of a limited supply of living space as expected in a finite island model.


Introduction
The heterogeneous distribution of biodiversity on Earth has called considerable attention in ecology and evolutionary biology (MacArthur and Wilson 1967;Nelson and Platnick 1981). Because ecosystems are dynamic and species permanently adapt to changing landscapes, understanding the evolutionary mechanisms structuring communities is a challenge (Ricklefs and Schluter 1993;Leibold et al. 2004). Species that live together in a local community do so because they are present in the larger regional species pool and have characteristics that permit their existence at a given locality and their coexistence with other species in the community . In this context, species richness results from the inter-action of biotic and abiotic factors determined at both local and regional scales (Ricklefs 1987;Ricklefs and Schluter 1993;Cornell 1993;Holt 1993;Caley and Schluter 1997;Losos and Schluter 2000;Whittaker et al. 2001;Ricklefs 2004).
According to the niche theory of species coexistence (Hutchinson 1959;Cornell and Lawton 1992;Leibold 1998), species occupy parts of the ecological space available and coexist locally if they are able to partition it in a way that minimizes ecological overlap otherwise they mutually exclude (limiting similarity hypothesis, Hutchinson 1959; Saturation hypothesis, Terborgh and Faaborg 1980; regional similarity hypothesis, Mouquet and Loreau 2002). In this context, one may expect that factors influencing the breadth of the ecological space available locally will influence species richness (species-energy hypothesis and primary production, Currie 1991;habitat heterogeneity, Fraser 1998;Guéguan et al. 1998).
Until recently, community assembly was seen as the result of species immigration from elsewhere in the regional pool followed by habitat filtering and biotic interactions at the local scale (Loreau and Mouquet 1999;Mouquet and Loreau 2003;Leibold et al. 2004;Holt 2009). By contrast, the ecological filling of niches through diversification has been traditionally considered to act on much larger scales than those involved in community assembly and to concern the evolution of the regional pools of species (Qian and Ricklefs 2000;Chase 2003;Wiens and Donoghue 2004;Weir and Schluter 2007). Worth noting, the scarcity of documented cases of community assembly through adaptive radiation supported the hypothesis that speciation contributes to community assembly only in particular ecosystems (Losos et al. 1998;Ricklefs and Bermingham 2001;Gillespie 2004;Duponchelle et al. 2008). The neutral theory of biodiversity challenged this view by proposing a model that explicitly assumes both an equivalent per capita fitness of species and a significant contribution of speciation in increasing species richness at the local scale (ecological drift model, Hubbell 2001). According to the ecological drift model, immigration and speciation counterbalance species loss by extinction at local scales and niche partitioning has no stabilizing effect on species richness (Mouquet and Loreau 2002;Kraft et al. 2008;Levine and HilleRisLambers 2009).
Clades diversify in an ecological context, yet most macroevolutionary models do not directly relate the dynamic of community assembly to large-scale evolutionary patterns (Kraft et al. 2007;McPeek 2008;Ackerly 2010;Ricklefs 2010). The neutral and niche theories of coexistence, however, are likely to affect patterns of diversification differently (Table 1). In a niche context, a lineage may initially diversify in a relatively empty ecological space and speciation rate should slow down as more niches become occupied (Schluter 2000). By contrast, the neutral model predicts no such decrease in net diversification since speciation and extinction, as stochastic and diversity independent processes, will be constant (Hubbell 2001;Ricklefs 2003). Yet, studies that explored the dynamic of community assembly and its impact on diversification gathered results either in silico (Chave et al. 2002;Mouquet and Loreau 2003;Allouche and Kadmon 2009;Chisholm and Lichstein 2009;Vergnon et al. 2009) or from communities of sessile organisms with limited dispersal abilities (Webb 2000;Cavender-Bares et al. 2004;Ackerly et al. 2006;Hardy and Senterre 2007;Kraft et al. 2007; Levine and HilleRisLambers 2009;review in Vamosi et al. 2009).
Coral reefs are among the most diverse and structurally complex ecosystems (Sale 1991). Currently exhibiting more than 5000 species (Froese and Pauly 2011), that is, 15% of the world's ichthyofauna, that coexist and aggregate in some of the largest vertebrate communities, the Indo-Pacific coral reef fish remarkably reflects this complexity (Bellwood and Hughes 2001;Mora et al. 2003;Froese and Pauly 2011). Community assembly in coral reef fish has been subject of intensive debate for decades (Sale 1977(Sale , 1978Anderson et al. 1981;Fraser and Currie 1996;Sale 1998;Almany 2004). Since coral reef fish tends to be generalists forming heterospecific feeding schools, several authors opposed to the conventional niche theory, a stationary model of coexistence, which emphasizes the role of stochastic recruitment and mortality in determining local coexistence of species (lottery model, Sale 1977Sale , 1978Chesson and Warner 1981;Warner and Chesson 1985). Although not explicitly assuming equivalent individual fitness among species, the lottery model makes predictions about community assembly that are very similar to the neutral theory since ecological overlap has no influence on coexistence (Table 1). By contrast, large-scale patterns of species richness in coral reefs are reliably predicted by ecological factors including primary productivity and disturbance (Fraser and Currie 1996), a pattern generally interpreted as a consequence of a niche-based model of community assembly (Hutchinson 1959;Terborgh and Faaborg 1980;Mouquet and Loreau 2002).
In this context, we focused on several predictions from the neutral and niche theories related with community assembly and diversification pattern in coral reef fish from the Indian and Pacific oceans (Table 1). We inferred phylogenetic relationships among 157 species from the families Chaetodontidae, Pomacentridae, and Labridae, being among the most diverse tropical reef fish lineages (Bellwood and Hughes 2001;Mora et al. 2003), to assess the phylogenetic structure of communities and pattern of diversification underlying those communities. The results are discussed in light of the predictions drawn from the neutral versus niche-based theory to unravel mechanisms underlying community assembly for coral reef fish.

Sampling design, specimens storage, and species identification
We focused our study on the phylogenetic diversity of a limited set of abundant families in order to avoid potential bias due to the inclusion of rare taxa with long branches that can lead to spurious phylogenetic structure (Kembel and Hubbell 2006;Hardy and Senterre 2007;Vamosi et al. 2009). We selected the families Pomacentridae, Chaetodontidae, and Labridae that are among the most speciose and whose taxonomy is regularly updated (Froese and Pauly 2011). Several phylogenies have been published during the last decade for these families (Westneat and Alfaro 2005;Fessler and Wesneat 2007;Cooper et al. 2009). Nevertheless, these phylogenies 230 Table 1. Summary of the predictions tested in the present study.

Theory Assumptions Predictions References
Community assembly Niche (1) Species have ecological preferences and differential fitness (2) Habitat filtering and/or competition drive community assembly Phylogenetic relatedness depart from expected under random species sorting Hutchinson (1959); Pianka (1976); Anderson et al. (1981); Webb (2000); Webb et al. (2002); Hardy and Senterre (2007); Holt (2009) Neutral (1) Species have equivalent individual fitness (2) Community assembly is a random process Phylogenetic relatedness does not depart from expected under random species sorting Sale (1977Sale ( , 1978; Hubbell (2001); Webb (2000); Webb et al. (2002); Hardy and Senterre (2007) Diversification Niche (1) The ecological space is limited in nature (2) Species richness is ceiled by competition Diversification rates are diversity dependent Ricklefs (1987); Schluter (2000); Rüber and Zardoya (2005); McPeek (2008); Phillimore and Price (2008); Rabosky (2009); Pigot et al. (2010); Ricklefs (2010) Neutral (1) The ecological space does not limit species coexistence (2) Species accumulation is a stochastic process Diversification rates are diversity independent Hubbell (2001); McPeek (2008) were inferred from different sets of molecular markers and samplings were focussed on taxonomic rather than geographical coverage. In this context, we gathered new phylogenetic data based on the same set of molecular markers for all families and covering all the species in the sampled communities in order to limit bias in our phylogenetic inferences. The sampling was conducted in sites across both inner reefs and outer slopes within Madagascar, Réunion, and French Polynesia ( Fig. 1; Appendixes 1 and 2). At each site, specimens were caught by rotenone poisoning performed in sampling plot measuring 20 × 20 m 2 . Poisoning was conducted with a constant amount of rotenone and duration. All species were collected; fresh specimens were labeled and photographed. Typically, a sample of white muscle was fixed in 90% ethanol and fresh specimens were conserved in a 10% formaldehyde solution. Identifications were done independently by three of us and further confirmed through examination of morphological characters (color, meristic counts). Recent molecular studies regularly emphasized the presence of cryptic diversity in coral reef fish including in the three selected families (McCafferty et al. 2002;Kuriiwa et al. 2007;Drew and Barber 2009;Steinke et al. 2009;Leray et al. 2010). In this context, several specimens from different sites and islands were sampled for each species and the three mitochondrial markers were sequenced in order to consider biological units and preclude taxonomic bias in the community analyses. The presence of the 157 sampled species was recorded in each of the 14 geographical and ecological sampling units and habitat characteristics such as depth and location on the inner reef or outer slope were registered (Appendix 1). Worth noting, specimens in French Polynesia were collected during the Moorea biocode project between 2006 and 2008 across more than 30 sites (http://mooreabiocode.org/). This project was designed to collect all the species from the island but not all the species for each site were recorded as in Reunion and Madagascar. Since habitat type and depth were recorded for each specimen, however, we listed the species from each kind of habitat (inner reef and outer slope) and treated each as a single site in our analyses.
All sequences generated for this publication have been deposited in GenBank (Appendix 1) and BOLD (project IPCOM). GenBank accession numbers for the outgroup sequences obtained from published mitochondrial genomes are also provided Appendix 2.

Phylogenetic reconstructions and divergence time estimates
Protein-coding regions were aligned manually while 16S sequences were first aligned through multiple alignments using Clustal W (Thompson et al. 1994) and manually refined. Aligned datasets ranged between 2319 bp for the Chaetodontidae and 2334 bp for the Pomacentridae and reached 2333 bp in the Labridae. Phylogenetic relationships within each family were assessed using the BIONJ algorithm (Gascuel 1997) as implemented in PAUP * 4.0b10 (Swofford 2002) and statistical support assessed with bootstrap proportion (BP) (Felsenstein 1985) through 2000 replicates. Once a tree was computed for each family, a final cladogram based on the topologies inferred for each family was build and a new alignment including a single, randomly picked individual for each 157 species was used for branch length optimization of the final cladogram in maximum likelihood (ML) using PhyML (Guindon and Gascuel 2003). ML optimizations were conducted under the GTR+I+ model with parameters optimized simultaneously during branch length computations. All trees generated in the present study have been submitted to Treebase and are available at http://purl.org/phylo/treebase/phylows/study/TB2:S11274.
The heterogeneity in substitution rate across lineages was first assessed through the likelihood procedure implemented in r8s (Sanderson 2003). Divergence times were reconstructed under the assumption of a molecular clock following the likelihood approach implemented in the Langley-Fitch method (LF method, Langley and Finch 1974) and then by relaxing the assumption of constant rate across the tree in using k separate rate parameters. Likelihood scores computed for the clock-like and the nonclock models were then compared with a likelihood ratio test (LRT). Since a model with two rate parameters (k = 2; nonclock) was compared with a single rate parameter (k = 1; clock-like) model, the LRT had one degree of freedom.
The heterogeneity in absolute substitution rate among lineages was further explored through the penalized likelihood model developed by Sanderson (2002) and also implemented in r8s. This approach combines a parametric model having different substitution rates on each branch with a nonparametric roughness penalty that limit the heterogeneity of the substitution rates across the tree. Thus, it assumes that substitution rates tend to be correlated on contiguous branches and the optimality criterion becomes the log likelihood of the parametric model minus the roughness penalty (Sanderson 2002). The relative contribution of each component is determined by a smoothing parameter (S) whose optimal values are determined through a cross-validation procedure that measures the fit between observed and predicted numbers of substitution in terminal branches according to the model. The S value maximizing the performance of the predictions is determined by a cross-validation score (CV ) that corresponds to the squared deviations between observed and inferred values and standardized by the observed ones. This method requires several reference ages across the phylogeny to produce robust inferences (Britton et al. 2007), therefore several age estimates based on published molecular phylogenies of Chaetodontidae, Pomacentridae, and Labridae were used as calibration points (Table 2).

Diversification rates and phylogenetic community structure
We first checked for potential departures from a stationary model of diversification through a lineage through time (LTT) plot analyses ) using APE 2.6-3 (Paradis et al. 2004) and based on the chronogram with the optimal smoothing parameter. The LTT plot corresponds to a lineage accumulation curve showing time versus number of lineages that we used to check the plausibility of mod-  Read et al. (2006) els of macroevolution by comparing observed and expected LTT plot under a given speciation model .
The goodness-of-fit of several alternative time-dependent birth-death models of diversification (Nee 2006) was determined through the Akaike Information Criterion (AIC) as implemented in APE. We first fitted to each family a birth-death model ) in order to estimate speciation and extinction rates separately. In all three cases, the estimated extinction rate was equal to zero. Giving this result and the notorious difficulties to estimate extinctions with molecular phylogenies (Kubo and Iwasa 1995;Stadler 2009;Paradis 2011), we focused on a Yule model considering speciation as a proxy for diversification. We also considered a time-dependent extension of the Yule model as implemented in APE that allows the user to specify any arbitrary model. We analyzed the phylogenetic structure of communities through an additive partitioning of phylogenetic diversity within and between sites following the model proposed by Hardy and Senterre (2007) and implemented in Spacodi (Hardy 2007). Providing that a "community" is defined as any assemblage of species spatially localized (e.g., all the fish species in a transect of 100 m in a reef or 20 × 20 m 2 plots as here) and picked from a regional pool (e.g., an archipelago or an ocean), partitioning the species phylogenetic distances  among communities from a regional pool may document the processes ruling community assembly and species coexistence (Webb 2000;Webb et al. 2002;Hardy and Senterre 2007). This hypothesis relies on the assumption that phy-logenetic distance mirrors ecological divergence (i.e., niche conservatism), a prerequisite supported in marine fish by recent phylogenetic studies (Streelman et al. 2002;Rüber et al. 2003;Rüber and Zardoya 2005;Fessler and Westneat 2007 According to Hardy and Senterre (2007), the measure of the mean phylogenetic distance between distinct species ( P ) can be evaluated at different levels with P S , the average within sites, and P T , the average among all sites, and both estimates can be used to estimate the gain of phylogenetic dis-tance between species occurring in different sites compared with species occurring in the same site with ST = ( P T -P S )/ P T . Then, ST can be computed after randomizing the species in the tree in order to test for significant departure from a model of random association of species according to their phylogenetic relatedness. If partial randomizations are performed according to threshold of phylogenetic divergence (e.g., less than 2% or 5 Millions years [Myr]), changes in community phylogenetic structure through time may be detected.
The ST index was computed at two spatial scales, among islands (i.e., regional scale influenced by geographic isolation among islands and oceans) or between inner reef and outer slope within island (i.e., local scale influenced by individual interactions, competition, and habitat choice), through two distinct analyses. In the first analysis, we assessed the phylogenetic community structure among islands by calculating ST using the entire species pools (i.e., all the plots) from Madagascar, Reunion, and French Polynesia (regional scale). This analysis aimed at estimating the influence of large-scale geographic isolation on community assembly likely influenced 236 by historical factors. In the second analysis, we estimated the phylogenetic community structure between contiguous habitats within islands by calculating ST using species list from the inner reef and outer slope following occurrence of species in the sampled 20 × 20 m 2 plots (local scale). This second analysis aimed at detecting departures from a random model of community assembly as a consequence of biotic interactions (e.g., ST < 0 as a consequence of competitive exclusion) or ecological constraints (e.g., ST > 0 as a consequence of habitat filtering). In both analyses, the ST index was plotted against time and computed every 5 Myr at the regional scale and each 10 Myr at the local scale.

Results
The 2300 bp from the two mitochondrial genes and ribosomal subunit provided well-supported phylogenies (Figs. 2,3,4). The topologies obtained for the three families were grouped into a single cladogram and a single individual for each of the 157 species was randomly picked from each of the three datasets. This composite cladogram including the 157 species and its mirrored dataset was further used for ML branchlength optimization using the GTR+I+ model (General Time-Reversible model; logL = -78462.14; base frequencies A = 0.249, C = 0.291, G = 0.186, T = 0.274; GTR relative rate parameters A-C = 1.83, A-G = 10.00, A-T = 2.11, C-G = 0.74, C-T = 13.35, G-T = 1.00; proportion of invariant site I = 0.474; gamma distribution shape parameter γ = 0.758). Together with previous estimates of clade ages (Table 2), the 157 species phylogram obtained from the ML analysis was used to explore the heterogeneity of rates of substitution. The LRT was significant for all the lineages examined except the Chaetodontidae (Table 3). In this context, we used a relaxed molecular clock model estimated through penalized likelihood analyses. The cross-validation procedure indicated that the CV scores were lower and best for intermediate values of the smoothing parameter (S = 1, CV = 3793; S = 3.2, CV = 3694; S = 10, CV = 3525; S = 32, CV = 3325; S = 100, CV = 3166; S = 1000, CV = 3040), while a large number of failures (more than 30) were observed in iterations involving smoothing parameter value of S = 10,000 and above. This result is consistent with previous observations that intermediate values of S between 100 and 1000 often provide the best fit (Sanderson 2002). Thus, we used the chronogram estimated with a smoothing parameter S = 1000 for the analyses of community phylogenetic structure and diversification (Fig. 5).
The three families provided very similar net diversification rates, overlapping with the estimate from the pooled data, suggesting common diversification regimes (Table 4; Fig. 6). We then tested several alternative models of diversification fitted by ML. Sea levels dramatically fluctuated during the last 50 Myr leading to alternative phases of fragmentation and connectivity (Haq et al. 1987;Fig. 6). Surprisingly, no effect was detected as diversification rates (λ) did not differ between periods of low or high sea levels and a logistic timedependent model of diversification provided a better fit. Some variations were detected, however, and models incorporating breakpoints provided the best fit, the one including three rates of diversification and two breakpoints being the most likely following AIC scores (Table 4). According to this model, the community diversification rate slowed down twice through time at 22.1 Ma and 6.4 Ma (Fig. 6).
The analysis of phylogenetic community structure supported a pattern of phylogenetic clustering when considering the 157 species as shown by a low, but significantly different from zero, positive ST among islands ( ST [50] = 0.004; Table 5; Fig. 6). This result is consistent with a long-term effect of geographic isolation on communities' composition. However, the very low ST value indicates that most of the phylogenetic diversity was shared among islands. Nevertheless, the use of partial randomization with respect to absolute divergence thresholds detected a nested but marked pattern of phylogenetic dispersion for the most recently formed species ( ST [5] = -0.214; Table 5; Fig. 6). Species diverged on average by 3.467 Myr within islands but only by 2.854 Myr on average among islands. This result is consistent with a model of allopatric speciation in recent times as the fragmentation of a species range distribution through vicariance leads to sister-species with nonoverlapping distribution, each sisterspecies co-occurring with phylogenetically more distantly related species (e.g., Barraclough et al. 1998;Barraclough and Vogler 2000). This model recently received support from the study of the evolution of species range distribution arguing that allopatric speciation followed by range change as a consequence of major shifts in species range distribution seems to be the rule in coral reef fish (Quenouille et al. 2011). Worth noting, this model predicts that range overlap is a function of divergence time between species (i.e., higher range c 2011 The Authors. Published by Blackwell Publishing Ltd.  Table 2 and a smoothing parameter S = 1000 for the penalized likelihood function (Sanderson 2002). overlap between species with higher divergence times), a prediction consistent with the pattern of phylogenetic dispersion detected only in recently formed species. Interestingly, this time-dependent model of community phylogenetic structure matches the shifts in diversification rates since a pattern of phylogenetic dispersion appeared at 22.1 Myr and further clearly establish at 6.4 Myr (Fig. 6).
We further explored phylogenetic community structure among communities sampled in the 20 × 20 m 2 plots in the inner reef and outer slope of Réunion, Madagascar and French Polynesia. Overall, the communities sampled harbored very low similarity since plots distant by a few hundred meters within the same habitat differed to the same extent as those from distinct habitats (Table 6). This pattern 238 c 2011 The Authors. Published by Blackwell Publishing Ltd. Table 4. Parameters estimates for distinct time-dependent models of diversification. AIC = -2log L + 2k, L the likelihood score, k the number of parameter of the model; Sea levels 50, model of diversification including periods of sea levels 50 m higher than present (λ >50 , according diversification rate) and remaining periods considered as low sea levels times (λ <50 , according diversification rate). was consistently repeated in Réunion and Madagascar and was not unexpected since it has been repeatedly described in the past (Sale 1977(Sale , 1978Sale and Dybdhal 1978;Anderson et al. 1981;Sale and Williams 1982;Sale et al. 1994). We were not able to estimate variability within habitat in French Polynesia, however, according to the sampling designed (see Materials and Methods section). By contrast with the first analysis of phylogenetic community structure among islands, all the communities sampled did not depart from the expectation of random associations at the local scale except in Réunion ( ST [50]; Table 7; Fig. 7). Nevertheless, this pattern was not stable in Réunion and only observed for the two larger divergence thresholds (Table 7). Globally, the ST estimates are consistent with a random assembly of species at the local scale in each island. Worth noting, this pattern was remarkably stable across the three islands.

Discussion
Phylogenetic community structure and spatial scales Initially proposed by Webb (2000), the analysis of phylogenetic community structure is based on the assumption that, because organisms interact via their phenotypes, and because phenotypes are not randomly distributed with respect to phylogeny, we should expect that the phylogenetic composition of a community is partially the product of species interactions (Webb et al. 2002;Emerson and Gillespie 2008;Vamosi et al. 2009;Ackerly 2010). In a niche-based view (Table 1), fluctuations in species richness within communities is likely to affect species interactions since diversity influences the probability of having better competitors in the community (Schluter and McPhail 1992;Gillespie 2004;Ackerly et al. 2006;Ackerly 2010). In this context, ecological mechanisms behind species interactions including habitat filtering, competitive exclusion, mutualism, or facilitation might themselves apply differently across descendant clades (McPeek 2008;Vamosi et al. 2009). Here, the use of a threshold-based approach confirmed that phylogenetic community structure might not be constant and helped detect nested patterns. Shifts in phylogenetic community structure, however, raise the question of the ecological mechanisms driving community assembly, and previous authors pointed to the importance of spatial scale in detecting phylogenetic community structure (Cavender-Barres et al. 2006;Swenson et al. 2006;Vamosi et al. 2009). The main challenge of considering several geographic scales consists in disentangling the drivers preventing species co-occurrence ranging from geographic isolation resulting from historical factors at the regional scale (e.g., allopatric speciation) to ecological mechanisms at local scale (e.g., habitat filtering, competitive exclusion). As a consequence, both competitive exclusion and allopatric speciation can lead to the confinement of species in alternative sites depending on scale and influence phylogenetic community structure (Vamosi et al. 2009).
Several observations support that the present pattern of phylogenetic dispersion among islands for the most recently derived species in the phylogeny result from allopatric speciation and not from competitive exclusion. First, evolutionary imprints of ecological mechanisms derived from individual interactions such as competitive exclusion are supposed to be effective among contiguous patches of alternative  Observed values for each diversification regime (λ 1 , λ 2 , λ 3 ) and 10 Myr windows in light and black lines, respectively. Best-fitted model of diversification for the entire community including three diversification rates (λ 1 , λ 2 , λ 3 ) and two breakpoints (τ 1 , τ 2 ) in red. (C) Cenozoic long-and short-term eustatic curves (modified from Haq et al. 1987). Times of sea levels 50 m higher than present are considered as high sea levels periods (bars on the time axis) in the sea level model of diversification in Table 4. Changes before 30 Myr are not considered due to the low diversity in extant lineages between 50 Myr and 30 Myr.
habitats (Losos et al. 1998;Ricklefs and Bermingham 2001;Gillespie 2004;Gavrilets and Vose 2005). The lack of significant structure, here, at the local scale argues for a distinct origin of the regional structure. Second, Indian and Pacific oceans have been repeatedly isolated during the last 5 Myr and particularly during the climatic fluctuations of the Pleistocene as sea levels frequently dropped by more than 100 m during glacial times leading to the formation of a land bridge across the Sunda shelf (Indonesia, Malaysia and Philippines archipelago) and isolating marine communities on each side (Randall 1998;Barber et al. 2000;Rocha and Bowen 2008). Finally, recent studies on dispersal and recruitment in coral reef fish argue for much restricted dispersal than previously thought despite the long-lasting pelagic larval stages making long-distance dispersal unlikely in coral reef fish (e.g., Doherty et al. 1995;Jones et al. 1999;Almany et al. 2007).
The shift in phylogenetic community structure among islands toward phylogenetic dispersion emphasizes that marine barriers to dispersal fluctuate in effectiveness through times (Rocha and Bowen 2008). A relationship between species range overlap and species divergence times has been recently detected in Pomacentrid fish and Labrid fish as expected following a model of allopatric speciation and evolution of species ranges through major shifts (Quenouille et al. 2011). The present pattern is consistent with this model since most of the lineages exhibiting alternative distribution (i.e., nonoverlapping ranges) at the regional scale are younger than 5 Myr and this pattern of phylogenetic dispersion abruptly disappears when considering older lineages. Worth noting, Quenouille et al. (2011) pointed to a large shift in species range overlap between 4 Myr and 5 Myr leading to an increased cooccurrence of lineages regionally, as described here. Sea levels fluctuations, particularly the last episode of marine highstand between 4 Myr and 5 Myr (Fig. 6), proceeding to variations in the extent of emerged lands and land bridges constitute a good example of such fluctuations that can influence the effectiveness of barrier to dispersal.
The present study confirms the importance of scales when considering communities phylogenetic structure since alternative patterns may be nested on distinct spatial and temporal scales (Vamosi et al., 2009). The pattern detected at the local scale confirmed a scale dependency of the phylogenetic structure since all three islands supported independently a model of random community assembly, with no phylogenetic structure between the inner reef and the outer slope, by contrast with the pattern of phylogenetic dispersion among islands. Overall, the assembly of species in communities across habitats departs from a deterministic model of species assembly through phylogenetic relatedness and competitive interactions, a result consistent with a model of random association (Sale 1977(Sale , 1978Chesson and Warner 1981;Warner and Chesson 1985).   (Réunion, Madagascar, and French Polynesia). ST estimates among islands using the chronogram from Figure 5 and species occurrence from Appendix 1. P S = mean divergence between distinct species from different sites (here, among species pools of each islands), P T = mean divergence between distinct species from the total pool of species (here, all the 157 species sampled in the three islands). Partial randomizations were conducted according to 10 thresholds (each 5 Myr). Obs = mean observed from data, Exp = mean expected after partial randomization, SD = standard deviation, CIinf = inferior 95% confidence interval, CIsup = superior 95% confidence interval, <exp = P-value of the one-sided test of Obs < Exp, >exp = P-value of the one-sided test of Obs > Exp, * significant one-sided test. Habitat filtering relies on ecological trade-offs for traits associated with the use of alternative resources differentially distributed in landscapes (Schluter and McPhail 1992). The lack of significant departure of the community phylogenetic structure from a model of random assembly may either mean that life-history strategies are not correlated with phylogenetic distance (i.e., phylogenetic conservatism of traits does not hold for coral reef fish), or that mechanisms promoting species coexistence are not based on niche specialization. The first hypothesis seems very unlikely since recent phylogenetic studies suggest that traits such as trophic strategies are generally conserved and major trophic shifts in the phylogeny correspond to large evolutionary steps seldom crossed during lineages' evolution (Streelman et al. 2002;Rüber et al. 2003;Alfaro et al. 2007;Fessler and Westneat 2007). By contrast, ecosystems with benign but frequent and unpre-dictable perturbations may be expected to favor species with the broader ecological requirements since ecological specialization reduces the opportunity to find suitable sites that are randomly available (Sale 1977(Sale , 1978. Thus, the present pattern of phylogenetic structure may be seen as an ecological consequence of a random model of community assembly promoting communities of generalist species rather than the lack of imprint of ecological mechanisms on community assembly and phylogenetic structure.

Community assembly and diversification
Communities are the product of biotic and landscapes fluctuations over evolutionary times (Ricklefs 1987;Ricklefs and Schluter 1993). Lineages diversify and supply communities with species that arrange themselves in assemblies as c 2011 The Authors. Published by Blackwell Publishing Ltd. Table 6. Summary statistics of community structure from inner reefs and outer slopes sites in Reunion, Madagascar, and French Polynesia according to Appendix 1. Similarity given by the Sørensen index defined as with 2c /(A + B), c being the number of common species between two samples, A and B being the number of species in samples a and b, respectively (Sørensen 1957 a consequence of ecological constraints (i.e., habitat preferences, competitive exclusion, stochastic perturbations) and macroevolutionary processes (i.e., speciation, extinction) proceeding to the sorting of species in communities (Johnson and Stichcombe 2007;Emerson and Gillespie 2008). Ecological and evolutionary processes were thought to act on distinct spatial scale; however, due to large differences in the timescales involved in speciation (i.e., Million years) and biotic interactions (i.e., a few tens of generations).
However, several examples highlighted that the ecological dynamics of community assembly might contribute to determine the evolutionary processes that drive diversification (Harvey and Pagel 1991;Ricklefs and Schluter 1993;Barraclough and Nee 2001;McPeek 2008;Rabosky 2009). First, a lineage may become extinct as a consequence of physical perturbations (e.g., climate change, volcanic activity, sea level fluctuations) or interactions with other lineages (McPeek 2008), and the extinction of a lineage may open new ecological opportunities and foster the diversification of other lineages (Schluter 2000;Gavrilets and Vose 2005). Second, a clade may initially diversify in relatively empty ecological space as a consequence of the colonization of novel habitats or the evolution of key innovations and foster community assembly through adaptive radiation (Sanderson and Donoghue 1994;Losos et al. 1998;Schluter 2000;Ricklefs and Bermingham 2001;Gillespie 2004;Gavrilets and Vose 2005;Ackerly et al. 2006;Ackerly 2010).
Sea level changes are such major physical perturbations able to foster communities rearrangements through macroevolutionary processes (i.e., speciation and extinction). The phylogenetic community structure detected here among islands is consistent with an important step of diversification through allopatric speciation during the last 5 Myr. Sea level changes have been hypothesized to promote allopatric speciation through vicariance on each side of the Sunda shelf during glacial times with low sea levels (Barber et al. 2000;Rocha and Bowen 2008), a hypothesis that predicts temporal changes in diversification rates through increased rates of speciation related to sea level changes. Likewise, sea level rises may be expected to increase the extinction rate, as a 150-m highstand will submerge most oceanic islands from the Indian and Pacific oceans and rarefy suitable habitats for coral reef fish in open areas (Briggs 1974). Slowdowns in diversification rate may result from either a decreasing of speciation or increasing of extinction through time, both scenarios being difficult to distinguish in practice (Kubo and Iwasa 1995;Rabosky and Lovette 2008;Stadler 2009). The likelihood analysis of alternative diversification patterns, however, found no evidence for an influence of long-lasting sea level changes on diversification (Table 4).
By contrast, the present pattern of diversification provided an intriguing insight to the interplay between community assembly and diversification. In a niche-based view (Table 1), diversification can be seen as the filling of landscapes ecological space through the production of new species that insert in the ecological system (Webb et al. 2002). Following this hypothesis, speciation rate can be high as lineages ecologically diversify and fill the available ecological space but speciation rate should slowdown as more niches become occupied (Schluter 2000;Gavrilets and Vose 2005;McPeek 2008;Phillimore and Price 2008;Rabosky 2009). Decreases in diversification rate have been recently described, including in marine fish, and have been systematically described as a consequence of the saturation of the ecological space and increased biotic interactions including competitive exclusion (Johns and Avise 1998;Rüber et al. 2003;Barber and Bellwood 2005;Rüber and Zardoya 2005;Read et al. 2006;Alfaro et al. 2007;Philimore and Price 2008;reviewed in McPeek 2008 andRabosky 2009). In our study, a decrease in diversification rate, consistent with a niche-based view of diversification, was found in agreement with previous studies (Table 1). According to the niche theory, competitive exclusion promote resource partitioning and ecological specialization (Schluter 2000), a trend that may be expected to produce frequent ecological shifts during lineage evolution to foster the filling of the ecological space (Losos et al. 1998;Ricklefs and Bermingham 2001;Gillespie 2004;Gavrilets and Vose 2005;Duponchelle et al. 2008). Nevertheless, this expectation is challenged by recent phylogenetic studies on marine fish supporting that ecological shifts have been rather rare during the diversification of several lineages of coral reef fish as exemplified by the scarcity of trophic shifts (Streelman  Table 7. Phylogenetic community structure between inner reef and outer slope sites within island (Réunion, Madagascar, and French Polynesia).
ST estimates using the chronogram from Figure 5 and species occurence from Appendix 1. P S = mean divergence between distinct species from different sites (here, among communities from the inner reef and outer slope), P T = mean divergence between distinct species from the total pool of species (here, all the species sampled in the island). Partial randomizations were conducted according to five thresholds (each 10 Myr). Obs = mean observed from data, Exp = mean expected after partial randomization, SD = standard deviation, CIinf = inferior 95% confidence interval, CIsup = superior 95% confidence interval, <exp = P-value of the one-sided test of Obs < Exp, >exp = P-value of the one-sided test of Obs > Exp, * significant one-sided test.

Obs
Exp   Rüber et al. 2003;Alfaro et al. 2007;Fessler and Westneat 2007). Similarly, our results on phylogenetic community structure at the local scale challenge this view since the lack of significant departure from a random community assembly model rather supports a neutral than a niche model of species coexistence. The lottery model of community as-sembly predicts that ecological specialization is not favored in a stochastic environment. By contrast, the unpredictability of the supply of living space favors sedentary species, which breed often, produce numerous clutches of dispersive eggs or larvae, and have broad ecological requirements (Sale 1977(Sale , 1978. In this context, the supply of suitable living space is more restrictive than the amount of trophic resources by itself (Almany 2004). In this case, competition may occur without leaving an evolutionary imprint of competitive exclusion in community phylogenetic structure, since competition occurs for living space and not for resources. Nevertheless, this model may conduct to a pattern of decreasing diversification still, providing that the supply of living space is limited as expected in a finite island model (MacArthur and Wilson 1967).

Conclusion
Overall, the phylogenetic community structure at local scale is consistent with the prediction from the neutral model of community assembly as proposed initially by Sale (1977) for coral reef fish and Hubbell (2001) for tropical forest trees. Nevertheless, Sale's lottery model of community assembly predicts that the stochastic and unpredictable nature of coral reefs prevents species loss in communities by favoring generalist and sedentary species with restricted individual living space that minimize interactions among adult individuals. Likewise, stochastic perturbations will likely prevent communities to reach equilibrium between extinction and immigration in interacting species assemblage, and competitive exclusion may take tens of generations to complete dominance in such a system (Sale 1977). By contrast, Hubbell's neutral model does not advocate such stabilizing effect of stochastic perturbations on coexistence. According to both the lottery and the ecological drift model, species-area relationships are a consequence of spatially limited supply of living space. Nevertheless, the implications for speciation and extinction rates are not trivial. The present pattern suggests that available living space is not only of importance for coexistence but also has evolutionary consequences on diversification. The decrease in diversification rates in noninteracting species assemblages may result from the saturation of space without consideration for the available ecological space. In this context, species richness may increase through diversification up to a stationary state in communities as a function of species-area relationships. According to this hypothesis, the heterogeneous distribution of species richness in Indo-Pacific coral reef fish communities may be a simple consequence of species-area relationships.
BIOCODE project for support and funding. The sampling was conducted according to the following permits: Collection permits 630/AM/07 and 91/AM/08, "Affaires Maritimes," Réunion; Collection permit 248/08-MAEP/SG/DGAEP/DPRH/SPP, "Direction de la pêche et des ressources halieutiques," Madagascar; Permanent agreement, "Délégationà la Recherche," French Polynesia. The authors thank D. Strasberg, F. Guérin, L. Humeau, and D. Da Silva in the "UMR PVBMT" at the "Université de la Réunion" for providing facilities in sample storage, DNA extraction, and support for sampling in Réunion. NH and SP thank E. Rochelle in the "Université de Perpignan" and C. Meyer in the Smithsonian Institution, Washington, for providing sequencing facilities. NH wishes to thank J. Eyraud and B. Warren for early insights into questions of community phylogenetic structure. All the authors thank C. Thébaud for his longstanding support in the BIOTAS project as well as G. Bernardi and the two anonymous reviewers for their constructive comments on a previous version of the article. This is publication ISEM 2011-097.
Appendix 1. Sampling records for the 157 species of Chaetodontidae, Pomacentridae, and Labridae included in the present study. Data include habitat, depth, geo-references, and date. Depth provided in meters.