One-third of the world's reef-building corals are facing heightened extinction risk from climate change and other anthropogenic impacts. Previous studies have shown that such threats are not distributed randomly across the coral tree of life, and future extinctions have the potential to disproportionately reduce the phylogenetic diversity of this group on a global scale. However, the impact of such losses on a regional scale remains poorly known. In this study, we use phylogenetic metrics in conjunction with geographical distributions of living reef coral species to model how extinctions are likely to affect evolutionary diversity across different ecoregions. Based on two measures—phylogenetic diversity and phylogenetic species variability—we highlight regions with the largest losses of evolutionary diversity and hence of potential conservation interest. Notably, the projected loss of evolutionary diversity is relatively low in the most species-rich areas such as the Coral Triangle, while many regions with fewer species stand to lose much larger shares of their diversity. We also suggest that for complex ecosystems like coral reefs it is important to consider changes in phylogenetic species variability; areas with disproportionate declines in this measure should be of concern even if phylogenetic diversity is not as impacted. These findings underscore the importance of integrating evolutionary history into conservation planning for safeguarding the future diversity of coral reefs.
The ongoing extinction crisis driven by anthropogenic impacts can potentially reach mass extinction levels in the future unless proper conservation measures are taken . Species richness has traditionally been the primary metric for quantifying biodiversity and setting conservation priorities , but it is increasingly apparent that given the limited resources available for conservation, not every species can be saved. This realization has prompted efforts to develop strategies for minimizing the loss of evolutionary diversity (ED) despite species loss [3–5]. Preserving ED is now seen by some as a goal in itself [6–10], and by others as a proxy for preserving functional diversity and ecosystem function [11–15]. Both of these goals require us to identify species and/or areas that are particularly important from an evolutionary standpoint using metrics that quantify ED.
A number of different ED metrics are currently available [16,17], each highlighting somewhat different aspects of the evolutionary process, but empirical analyses of how the loss of a given set of species affects different metrics remain scarce . This hampers our ability to understand the behaviour of these metrics in real-world situations and also impedes the development of guidelines for their use in risk assessment . Here we explore this issue using two different measures of ED in conjunction with geographical distributions of all 842 living reef-building coral species to predict regional diversity changes resulting from projected species loss according to the International Union for Conservation of Nature (IUCN) Red List of Threatened Species . Our goal is to compare the erosion of regional ED due to anthropogenic extinctions using the two measures, and to use the results to identify coral reef ecoregions whose ED may be most at risk in the future.
Coral reefs are among the most threatened ecosystems of the world [21–25]. One-third of all reef corals (Scleractinia) are facing heightened extinction risk from climate change and other impacts . In particular, the diversity and abundance of corals have been heavily impacted by mass bleaching and disease driven by increasing sea surface temperatures [26–29], as well as outbreaks of the predatory crown-of-thorns seastar, Acanthaster planci, caused by nutrient enrichment and overfishing of reefs [30–33]. These threats are not randomly distributed across the coral phylogeny [34,35]. Species susceptible to predation by the crown-of-thorns seastar are the most clustered on the tree, but projected extinctions due to bleaching and disease are also expected to disproportionately reduce the phylogenetic diversity of corals on a global scale . However, how such losses will affect coral ED on a regional scale, at which marine protected areas are being planned, implemented and enforced [37,38], remains poorly explored.
To address this knowledge gap, we model the loss of reef coral ED due to anthropogenic impacts using phylogenetic diversity (PD)  and phylogenetic species variability (PSV) . Of the two metrics, PD emphasizes the importance of deeper nodes on a phylogeny; loss of old lineages results in a much higher reduction of PD compared with the loss of younger taxa . Total PD of a group generally increases with species richness [4,40]. By contrast, PSV estimates the degree of phylogenetic disparity among species present in an area . PSV ranges from zero (representing maximum relatedness) to one (in which all species are unrelated, i.e. a star phylogeny) and is not expected to correlate with species richness [41,42]. Because the two metrics focus on different aspects of the phylogenetic tree, loss of the same number of species can lead to different levels of PD versus PSV loss (figure 1).
Phylogenetic diversity has been widely utilized in the conservation literature and is a useful metric if the goal is to minimize the loss of deeper evolutionary history [5,7,9,43]. However, PD may be a poor predictor of trait variation present in a region [44,45]. Because maintenance of traits and associated functional diversity is critical for the resilience of complex ecosystems such as coral reefs [24,46,47], we use the PSV metric, which might be a better proxy of functional variation under phylogenetic conservatism [48,49], to quantify the effects of species loss on evolutionary disparity of corals. Together these two metrics should provide a more comprehensive understanding of how different aspects of the ED of reef ecoregions are likely to change in the future.
2. Material and methods
(a) Phylogenetic reconstruction
We used the supertree method modified from two recent studies [34,36] to reconstruct the phylogeny of the scleractinian clade, comprising 842 reef and 705 non-reef species . The source trees were derived from a molecular phylogeny of 474 species (based on seven mitochondrial DNA markers), 13 morphological trees and a taxonomic tree [34,36]. These datasets were combined using two different methods, the SuperFine-boosted Matrix Representation with Parsimony  analysed in PAUP* 4.0b10 (10 000 iterations) , and the Matrix Representation with Likelihood (MRL)  executed in RAxML 7.2.8 (using the S2 + CAT model and 1000 replicates) [54,55]. We then computed the strict consensus of the resulting trees and used that as the basic supertree topology for subsequent analyses.
We used BEAST 1.8 [55–57] to fit the DNA sequence data mentioned above onto the supertree topology using fossil node calibrations detailed in Stolarski et al. . These calibrations date the origins of Caryophyllia to the Oxfordian (Late Jurassic), Dendrophylliidae to the Barremian (Early Cretaceous) and Flabellum to the Campanian (Late Cretaceous). These were each treated as a lognormal distribution with hard lower bound (age of fossil) and soft upper bound described by log mean and standard deviation of 2.0 and 0.85, respectively . We carried out 10 separate Markov chain Monte Carlo (MCMC) analyses each with 30 million generations and a sampling interval of 1000. Runs were combined following rejection of the first one-third of all posterior trees, with MCMC convergence checked in Tracer 1.5 . Polytomies were randomly resolved using PolytomyResolver  by constraining each node with a normal prior based on the 95% highest posterior density derived from the ultrametricized molecular tree . Five MCMC analyses of 3 million generations and a sampling interval of 10 000 were performed, with the first one-third of all posterior trees discarded and the remainder combined. The resultant 1000 fully resolved supertrees were then trimmed of all non-reef species for the analyses of evolutionary diversity.
(b) Geographical and conservation status data
We obtained geographical range information for each species from the Coral Geographic database [62,63], which defined 141 ecoregions using distributional data. Range data for species not covered by the Coral Geographic were derived from the IUCN Red List of Threatened Species  and the Global Biodiversity Information Facility (http://data.gbif.org). Recent updates by Hughes et al.  were also incorporated.
Conservation status of each species was obtained from the IUCN Red List. Of the 842 living reef-building scleractinian species, 688 were assessed in 2006 and 2007, with one-third being designated as threatened, including four Critically Endangered (CR), 23 Endangered (EN) and 198 Vulnerable (VU) species .
As an additional test, we also applied the vulnerabilities of species inferred by the trait-based approach of Darling et al. . Their analyses clustered 143 coral species into four groups representing competitive, weedy, stress-tolerant and generalist taxa. The 46 competitive species were predicted to be the most sensitive to anthropogenic impacts and environmental change , so we used this category as an analogue of the threatened species on the IUCN Red List. However, as this classification was only available for 17% of all living reef coral species, we focus primarily on results using the IUCN Red List classification.
(c) Extinction analyses
We used two different metrics to quantify the ED of each ecoregion before and after simulated extinctions. PD  was computed as the sum of all branch lengths of the phylogeny of species present in an ecoregion [44,66]. The second metric, PSV , estimates the degree of phylogenetic disparity (or the lack of evolutionary redundancy) among species by quantifying the variance of a hypothetical neutral trait that is shared by all species in an ecoregion (see also [41,42]). Note that this metric is half of the mean phylogenetic distance among species pairs  for an ultrametric tree with root-to-tip lengths scaled to one .
To determine how species loss affects tree shape, we quantified topological and branch length balance with the Colless  and γ  statistics, respectively. The Colless index sums each node's size disparity between the two daughter clades—a larger index value indicates greater imbalance. We used the standardized index based on the Yule (pure birth) model in order to compare trees with different sizes . For the γ statistic, a positive value indicates that the internal nodes are closer to the tips than expected under the Yule model; conversely, if γ is negative, branches near the tips are longer than those close to the root relative to the Yule expectation [70,72].
For each ecoregion, we simulated species loss by assuming that all threatened corals (with CR, EN or VU status) will go extinct in the future, a scenario referred to as ‘observed’. Reductions in PD and PSV, as well as associated changes in tree shape, were compared with a null model of the same level of random species loss (1000 replicates) [36,43,44,73]. To quantify the declines in PD and PSV specifically, we computed the excess loss of each metric under the ‘observed’ scenario relative to random, as a percentage of the random result . All calculations were performed for each of the 1000 posterior trees obtained from the phylogenetic analysis, and repeated for the 143 species classified according to life-history strategy  with the predicted extinction of ‘competitive’ corals.
To examine the effects of species richness and the level of extinction on both metrics, we fit a linear mixed-effects model with richness and proportion of threatened species as fixed effects on each of the response variables PD and PSV. As the two predictors were correlated (Pearson correlation coefficient, r = 0.827; electronic supplementary material, figure S1), we fit the models separately for each effect. Note that only one of them could be considered a valid predictor, though there was no a priori reason to choose either one. Phylogeny, represented by the 1000 posterior trees, was included as a random effect to account for phylogenetic uncertainty. We omitted the eight ecoregions containing less than 20 species because most of them were not projected to experience local extinction.
The PD of reef ecoregions scales positively with species richness (figure 2a,b; electronic supplementary material, figure S2), but the relationship is nonlinear with areas of low or moderate richness such as the central Pacific and Indian Ocean showing elevated PD. By contrast, phylogenetic species variability does not scale predictably with species richness (electronic supplementary material, figure S3). PSV is high in the Persian Gulf, Gulf of Oman and Gulf of Martaban in the Indian Ocean, as well as the central Pacific area (figure 2c). However, some of the areas with low species richness have extreme values (both high and low) of PSV, with most of the Caribbean ecoregions (n ≤ 63) recording the lowest PSV. The most species-rich ecoregions (i.e. the Coral Triangle) have moderate levels of PSV (figure 2c), reflecting the close phylogenetic relationships of species in this biodiversity hotspot.
Although ecoregions with the highest species richness also have the greatest proportions of threatened species (figure 3a; electronic supplementary material figure S1), simulated extinctions in these ecoregions reveal less-than-random excess loss of PD (figure 3b; electronic supplementary material, figure S4). Ecoregions with the highest excess PD loss are generally species poor, including Johnston Atoll, East Hawaii and Pacific Costa Rica and Panama (table 1; electronic supplementary material, table S1). Similarly, ecoregions with the highest richness have minimal excess loss of PSV, less than most ecoregions with fewer species (figure 3c; electronic supplementary material, figure S4). Spatial patterns of excess PD and PSV losses are not congruent, although some ecoregions rank high in both metrics, e.g. Pacific Costa Rica and Panama (figure 4a), Lakshadweep, Honshu, Gulf of Martaban and Clipperton Atoll (tables 1 and 2).
Results are generally consistent for species vulnerabilities predicted using life-history strategy by Darling et al.  (electronic supplementary material, table S2). Excess losses of both PD and PSV under the extinction of corals representing the competitive strategy are positively correlated with losses based on the IUCN Red List across ecoregions (r = 0.534 and 0.568 for excess PD and PSV loss, respectively, d.f. = 110, p < 0.001; electronic supplementary material, figures S5 and S6).
Interestingly, all of the Caribbean, Hawaiian and Johnston Atoll ecoregions are projected to suffer from large excess PD loss but with gains in PSV. For example, the threatened species at Johnston Atoll are represented by the deep lineages of Alveopora (one species) and Cyphastrea (two species), which account for the high PD loss (figure 4b). However, there are several other losses of very short branches that are spread across the phylogenetic tree, giving rise to relatively small reduction in γ (slight increase in ratio of terminal versus internal branch lengths) and large decline in the Colless index (greater balance in tree topology; table 1), which overall lead to a net positive change in PSV.
Ecoregions with high excess PSV loss but low PD loss are generally characterized by extinctions of species from one part of an initially balanced tree. In particular, seven of the 32 species in southwest Western Australia are threatened (table 2). These are mostly medium-length branches concentrated within the ‘complex’ clade (figure 4c), resulting in average projected loss of γ but large gain in the Colless index.
Linear regressions of excess PD and PSV losses on the fixed effects of richness and threatened proportion, accounting for phylogeny as a random effect, all reveal significant negative relationships (figure 5; p < 0.001 for non-zero slope). These trends hold even for the small dataset of 143 species classified according to life-history strategy  (p < 0.001 for non-zero slope).
Our results show that the two metrics of evolutionary diversity used here provide different pictures of how anthropogenic extinctions are likely to affect the ED of reef corals on a regional scale. Some of this difference is to be expected given that the metrics emphasize different attributes (age versus disparity), but the lack of congruence highlights the danger of using any one biodiversity metric for risk assessment.
Interestingly, areas with the most coral species that are current targets of conservation efforts are also characterized by high levels of evolutionary redundancy, whereby species tend to be very closely related to one another . In particular, the Coral Triangle region, long recognized as a biodiversity hotspot for corals and other taxa [62,82–84], has the lowest PD relative to the random expectation (electronic supplementary material, figure S2). The evolutionary redundancy in species-rich ecoregions also accounts for their low phylogenetic disparity, as quantified by the PSV metric (figure 2c). These ecoregions are therefore better buffered against losses of ED even at high rates of species loss, a result well corroborated by the extinction simulations (see below). However, many ecoregions with much lower species richness exhibit higher levels of ED than expected. The Persian Gulf, Gulf of Oman and Gulf of Martaban stand out as ecoregions that possess as much or greater PD than random assemblies of species across the phylogeny (figure 2b; electronic supplementary material, figure S2). Their PSV values are also the highest among all ecoregions, indicating large phylogenetic disparity among species. Thus even a few species lost from these areas can greatly affect their ED (electronic supplementary material, table S1).
Both PD and PSV show an overall negative relationship between projected excess loss of ED and extinction magnitude (figure 5). This combined with a strong positive correlation between the latter and species richness (electronic supplementary material, figure S1) leads to the following generalizations: (i) ecoregions with high species richness of reef corals (and thus high PD) have excess PD losses that are less than random expectations (figure 5a); (ii) ecoregions with high species richness have excess PSV losses that also tend to be low but are generally higher than random expectations (figure 5c); and (iii) ecoregions with low richness have highly variable excess PD and PSV losses ranging from very high to very low (figure 5a,c).
One of the more surprising findings here is that while ecoregions with very high coral species richness could lose large numbers of species due to anthropogenic extinctions, they are unlikely to lose a disproportionate amount of their ED. This is particularly true for the Coral Triangle region where estimated PD losses are lower than the random expectation and PSV losses are also low, despite the highest projected rates of species loss of any ecoregion (figures 3 and 5; electronic supplementary material, figure S4). In contrast to the high level of evolutionary redundancy seen in the richest areas, ecoregions with lower species richness often harbour older and species-poor lineages of corals that are threatened with extinction. For example, along the Pacific coast of Costa Rica and Panama, no close relatives of the Critically Endangered species Siderastrea glynni (Siderastreidae) are present (figure 4a) . Consequently, the extinction of this species spells the loss of an extremely long branch (median height = 170 Myr, 95% highest posterior density (HPD): 120–252 Myr), accounting for the ecoregion's disproportionately large loss of PD (11.1% of original PD). Because S. glynni is endemic to this region it has also been characterized as an Evolutionarily Distinct and Globally Endangered (EDGE) species . Other areas that are in danger of losing a disproportionate amount of their PD include the Hawaiian archipelago, Johnston Atoll and the entire Caribbean region (table 1).
Ecoregions with high excess loss of PD that are also projected to lose a large proportion of their PSV include Pacific Costa Rica and Panama, Lakshadweep, Honshu, Gulf of Martaban and Clipperton Atoll (figure 3; tables 1 and 2). For these areas, the extirpation of entire long branches are responsible for the high PD loss, but removal of short branches mainly from one section of the ecoregion phylogeny upsets the balance of tree topology leading to a disproportionate loss of PSV. It is worth noting that reefs in these ecoregions have been predicted to suffer from annual coral bleaching episodes at or earlier than the median year for reefs globally . Aside from the probable loss of Siderastrea glynni in Costa Rica and Panama, four other species with relatively short tip branches—all from the ‘robust’ clade—are in danger of extinction (figure 4a). The largest decline of PSV in our data is evident in southwest Western Australia resulting from a high extinction magnitude (21.9%) exclusively targeting species in the ‘complex’ clade (figure 4c).
The 10 ecoregions with the highest excess loss of PSV all show gains in the Colless index (table 2, cf. table 1). Substantial increases in phylogenetic imbalance have the potential to put reefs at further risk, due to the difference in structural diversity of corals in the ‘complex’ and ‘robust’ clades. The ‘robust’ corals comprise primarily species that produce heavily calcified skeletons forming massive or submassive colonies, while ‘complex’ corals are less heavily calcified and occur mainly as branching, columnar or laminar colonies . Although there are some exceptions to this generalization , differential extinctions of species in these clades can potentially compromise the future structural and functional diversity of reefs. For example, in southwest Western Australia, loss of ‘complex’ Acropora and Turbinaria species (figure 4c) with branching and laminar morphologies could reduce the complex three-dimensional architecture of reefs. This has negative consequences for the abundance, richness and functional diversity of reef inhabitants, including fish [88–90] and invertebrates [91–93]. Conversely, the preferential loss of ‘robust’ corals as projected for Pacific Costa Rica and Panama (figure 4a) may result in surviving assemblages dominated by Acropora valida, Gardineroseris planulata, Pavona and Porites spp., which are more susceptible to outbreaks of bleaching, disease and the crown-of-thorns seastar [20,34,36,94,95]. These results suggest that extinction risk examined in the context of phylogenetic disparity and associated changes in shape of the coral phylogeny can provide useful clues to the future ecological functioning of reefs.
The putative connection between changes in PSV and functional attributes of the reef system suggests the potential for an extinction debt scenario  in areas with high excess loss of PSV even where PD declines are expected to be low (table 2). In these areas, the initial loss of species could lead to substantial changes in habitat complexity and ecosystem function, which in turn could affect the long-term viability of the survivors. This hypothesis cannot currently be tested since we lack the data to directly quantify how PSV relates to the functional diversity of corals. Trait information for the vast majority of corals remains sparse [65,97,98] and predominantly focuses on life-history characteristics [99,100]. Our results indicate that a concerted effort to collect functional trait data spanning the breadth of the coral phylogeny will be extremely valuable for refining phylogeny-based predictions of the future functioning of coral reefs.
While some ecoregions rank high regardless of which metric of ED is used, highlighting their importance from a conservation perspective, risk assessment of ecoregions that score high on one metric but not on the other is more difficult and involves qualitative decisions about which aspect of the diversification process we value more . On one hand, if the goal is to preserve evolutionary heritage, a key component of biodiversity [5,7,10,102], then focusing our efforts on reef ecoregions with high-standing PD (e.g. Persian Gulf) or more-than-expected PD loss due to anthropogenic extinctions (e.g. Caribbean) makes sense. On the other hand, PD has been shown to poorly capture trait or functional diversity [19,103] especially on a regional scale [44,45]. More importantly, the most recent common ancestor of modern reef corals probably lived about 400 million years ago in the Palaeozoic , and many major coral lineages are relatively old . Given their long history it is likely that some of the old lineages (e.g. Catalaphyllia jardinei and Moseleya latistellata) whose loss would disproportionately affect future PD represent a ‘dead clade walking’ scenario . Whether preserving such lineages are effective from a functional and evolutionary standpoint is an open question. Finally, metrics such as PSV, which emphasize phylogenetic disparity among species, may be useful predictors of the future structural and functional diversity of coral reefs, although this needs to be further tested with appropriate trait data.
In summary, using two different metrics and comparing projected and random extinctions, we have identified key ecoregions where the ED of corals may be most at risk from anthropogenic impacts. Regardless of the metric used, our results suggest that these areas are not necessarily ecoregions with the highest species richness. In fact, the projected loss of ED in the Coral Triangle, a well-known marine biodiversity hotspot, is lower than that of most other regions. While much remains to be done to better understand how phylogenetic metrics like those used here relate to ecological and functional diversity of corals, our results clearly indicate that many species-poor areas are of potential conservation interest. We remain optimistic that regional analyses such as these will help better allocate conservation resources that are needed to safeguard the future of coral reefs.
All datasets and R scripts are available at the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.178n3.
This work is made possible by the Ah Meng Memorial Conservation Fund R-154-000-507-720 (Wildlife Reserves Singapore), and D.H. is supported by US National Science Foundation grant DEB-1145043-1145408.
We thank Gregory Rouse, Loke Ming Chou and Nancy Budd for support and advice, as well as Keith Crandall and two anonymous reviewers for constructive comments that improved the manuscript. Participants of the Royal Society scientific meeting on Phylogeny, Extinction Risks and Conservation contributed valuable ideas.
One contribution of 17 to a discussion meeting issue ‘Phylogeny, extinction and conservation’.
- © 2015 The Author(s) Published by the Royal Society. All rights reserved.