The risk of global extinction of reef-building coral species is increasing. We evaluated extinction risk using a biological trait-based resiliency index that was compared with Caribbean extinction during the Plio-Pleistocene, and with extinction risk determined by the International Union for Conservation of Nature (IUCN). Through the Plio-Pleistocene, the Caribbean supported more diverse coral assemblages than today and shared considerable overlap with contemporary Indo-Pacific reefs. A clear association was found between extant Plio-Pleistocene coral genera and our positive resilience scores. Regional extinction in the past and vulnerability in the present suggests that Pocillopora, Stylophora and foliose Pavona are among the most susceptible taxa to local and regional isolation. These same taxa were among the most abundant corals in the Caribbean Pliocene. Therefore, a widespread distribution did not equate with immunity to regional extinction. The strong relationship between past and present vulnerability suggests that regional extinction events are trait-based and not merely random episodes. We found several inconsistencies between our data and the IUCN scores, which suggest a need to critically re-examine what constitutes coral vulnerability.
Climate change is increasing the severity and the intensity of thermal stress events on modern coral reefs [1–4]. One of the primary concerns of conservation biologists and ecosystem managers is assessing the vulnerability of ecologically important taxa and mitigating their risk. The risk of global extinction of coral species is currently assigned by the International Union for Conservation of Nature (IUCN) using population reduction thresholds. For the vast majority of coral species, IUCN conservation status is evaluated from a weighted average calculated by multiplying the area of reef, within a species' distribution range, by the per cent of total coral cover loss in 17 different geographical regions. Using this method, the IUCN has documented an increase in the elevated extinction risk status for coral species from 2 per cent of 704 species during pre-1998 to 33 per cent by 2008 . This approach is currently the only comprehensive assessment of coral extinction risk, but it is calculated with limited data on species-specific trends, and lacks a historical validation process [6,7].
The regional extinctions in the Plio-Pleistocene may shed light on which coral genera may be vulnerable or resistant to stress during contemporary ocean warming. During the Plio-Pleistocene turnover, 25 coral genera persisted, while 18 were lost from the Caribbean region (figure 1). During this period, some corals went regionally extinct with persisting congenerics in the present-day Indo-Pacific (Alveopora, Caulastraea, Galaxea, Gardinoseris, Goniopora, Isopora, Pavona, Pocillopora, Psammocora, Stylophora and Trachyphyllia), whereas others went globally extinct (Antillia, Antillophyllia, Hadrophyllia, Pironastrea, Placocyathus and Thysanus) . The rates of change towards regional and global extinction in the Caribbean varied considerably; some genera gradually declined, whereas others declined rapidly .
In the early Pliocene, large colonies of Pocillopora dominated the shallow reefs along with Stylophora [11,12]. Stylophora was highly vulnerable to the climate shifts of the Plio-Pleistocene, and went regionally extinct approximately 1.8 Ma from the Caribbean. By contrast, Pocillopora gradually diminished over millions of years, shifting from a ubiquitous distribution of predominantly large colonies, to a patchy distribution of small colonies, and eventually to regional extinction during the last interglacial. The foliose Pavona that went extinct in the Caribbean through the Plio-Pleistocene was never dominant but diminished over several million years through repeated isolation . The two Caribbean Acropora species (Acropora palmata and Acropora cervicornis), increased in relative abundance through the Pliocene and Pleistocene. The Montastraea annularis complex, Siderastrea and branched Porites changed little over this 5 Myr period .
If indeed biological traits lead to differences in fitness, then the march to regional extinction during the Plio-Pleistocene in the Caribbean should be reflected in the genealogy of contemporary species in the warming Indo-Pacific. By contrast, if individuals on coral reefs are responding to changing conditions equally and independently of species traits , species may march through time showing random episodes of extinction. Here, we test an alternative to the concept of random episodes of regional extinctions, and propose that biological traits could play a larger role than suggested by neutral theory . We test these ideas by comparing modern corals with coral populations that lived on ancient Plio-Pleistocene reefs of the Caribbean that experienced a protracted period of complex environmental change and extinction [9,14,15]. Our objective was to evaluate whether regional extinction events of the Plio-Pleistocene Caribbean region lend insight into future events in the Indo-Pacific. Our central hypothesis was that trends through the Plio-Pleistocene will reflect contemporary trends on Indo-Pacific reefs. A mismatch between past and present trends would lend support to neutral theory, which suggests that species extinctions are a random process. By contrast, a match between past and present trends would suggest a non-random, biological trait-based process.
2. Material and methods
(a) Modern genera resilience scores
We examined whether biological traits play a large role in the outcome of coral persistence through environmental change by first categorizing modern corals based on a resilience score. To determine the resilience score of modern corals, a group of experts compiled a suite of biological traits and processes that were considered as potentially subjected to selective pressure. These characteristics encompassed morphological, physiological and reproductive traits and processes of coral genera. Quorum responses and consensus decision-making  constituted that the traits and scoring were purposefully decided a priori (table 1), and that all traits were maintained throughout the analysis to avoid tautology. A consensus was reached among 10 coral-reef scientists on the general principle underlying each trait and the methodology of the scoring protocol (table 1) . For analysis, we selected 14 dominant (modern) coral genera for the Indo-Pacific Ocean and 15 for the Caribbean Sea. Two genera were subdivided based on morphology with Pavona classified as either foliose or encrusting/submassive and Porites as branching or massive. Montastraea in the Caribbean were evaluated as either Montastraea annularis complex or Montastraea cavernosa. Although both genera and subgenera groups were examined, we refer to all corals at the genus level. Each coral genus was then evaluated on their tolerance to a hypothetical two-month + 3°C thermal stress anomaly, during the solar insolation maximum. Similarly, four key process variables were scored for a suite of coral taxa, to address whether each process would affect recovery from the temperature anomaly (table 1). The scores for the traits and processes were summed for each coral genus and classified as a resilience score. Although contemporary species within each genus differs in terms of their geographical distribution—some are widespread and others are restricted—classic taxonomic texts  and recent molecular techniques show evidence that many genera (e.g. Acropora, Porites, Stylophora, Pocillopora and foliose Pavona) are distinct, and coherent monophyletic entities are closely related and often difficult to distinguish [18–22]. Therefore, for this investigation, we considered it appropriate to lump species within genera.
(b) Plio-Pleistocene reef data
The fossil data were based on samples extracted from outcrop exposures at 70 localities through four Plio-Pleistocene sequences: Curacao , Costa Rica , the Dominican Republic [12,24] and Jamaica . The collections comprised approximately 6528 specimens and 154 species, and were deposited at the US National Museum of Natural History (USNM), the University of Iowa (SUI) and the Natural History Museum in Basel, Switzerland (NMB). All of the specimens were identified to species using a standard set of morphological characters and character states, established in part by comparing morphological and molecular data and detailed in the Neogene Marine Biota of Tropical America (NMITA) taxonomic database . Localities were grouped into faunules, which were defined as a set of lithologically similar localities from a small geographical area (usually less than 1 km or approximately the size of a large rock quarry) and restricted stratigraphic interval (usually less than 20 m). Age dates for the faunules were obtained by integrating data using high-resolution chronostratigraphic methods, including nannofossils and planktonic foraminiferal biostratigraphy, palaeomagnetics and strontium isotope analyses [26,27], and generally range in accuracy from 0.5 to 2 Myr. The dataset consisted of counts of specimens belonging to species within each faunule, and is available on the NMITA website.
(c) Data analysis
We sought to answer the question: what was the probability that a coral taxon that went regionally extinct during the Plio-Pleistocene, in the Caribbean, was also vulnerable on modern reefs in the Indian and Pacific Oceans? The probability that a coral went extinct in the Plio-Pleistocene, Pp(x), was determined from the fossil dataset for the Caribbean region (see §2b). For the modern corals, we defined contemporary vulnerability as a coral that was among the three lowest-ranked corals in the experts' resilience list (outlined above). This contemporary vulnerability was defined as Mp(x). The product of the probabilities (Pp(x).Mp(x)) defined the chance that a coral would have gone extinct through the Plio-Pleistocene in the Caribbean, and also ranked among the three lowest (modern) resilience scores. Using Bayes's theorem, we iteratively calculated posterior probabilities for extinction using Markov chain Monte Carlo simulations and Gibbs sampling. We used a binomial distribution θ for extinction probability and a beta prior (1, 1) to calculate the posterior probabilities, p(θ|y) given the data y. All models were implemented using OpenBugs .
Using a penalized maximum-likelihood logistic regression [29,30], we estimated how the probability of persistence of coral genera in the Caribbean region, through the Plio-Pleistocene, varied with the resilience scores applied to the same genera in the modern Indo-Pacific. Included in the analysis were the eight coral genera that were present in the Plio-Pleistocene Caribbean region and were also present in the modern Indo-Pacific Ocean, which included Acropora, Indo-Pacific Montastraea, Pavona (branching), Pavona (foliose), Pocillopora, Porites (branching), Porites (massive) and Stylophora. The binary-dependent variable in the analysis described the status of each genera at the end of the Pliocene as either extant (with a value of 1) or extinct (0) with the resilience scores for the modern Indo-Pacific as the independent predictor variable.
We compared the resilience scores with the conservation status of reef-building corals determined by the IUCN Red List criteria. The IUCN has classified coral species using a set of criteria incorporating observed or anticipated population trends and levels of geographical isolation into the following categories: data-deficient (dd), least concern (lc), vulnerable (vu), endangered (en) or critically endangered (cr) . Following these methods, we categorized as ‘Threatened’, the coral species identified as vu, en or cr within the Indo-Pacific region (table 2). For the Indo-Pacific coral genera, we compared the rank correlation (Kendall's tau) of the percentage of threatened species within each genera and morphological type, where described, using classifications from Veron & Stafford-Smith  and their resilience scores. Statistical models were constructed in the R statistical software (http://www.r-project.org).
(a) Modern genera resilience scores
The capacity of coral taxa to persist regionally was defined as a resilience score, which was a combination of their biological trait and ecological processes. Three Indo-Pacific corals were ranked low and most vulnerable (less than or equal to −3; Stylophora, Pocillopora and foliose Pavona), and four were ranked high and least vulnerable (greater than or equal to 4; Goniastrea, Favia, massive Porites and Indo-Pacific Montastraea; tables 3 and 4). In the Caribbean, only Madracis and branching Porites ranked low and vulnerable (less than or equal to −3) and three ranked high and tolerant (greater than or equal to 4; Siderastrea, Montastraea annularis complex and Diploria; tables 3 and 4).
(b) Plio-Pleistocene and modern coral vulnerability
If extinction was a chance event, and biological traits and ecological processes were irrelevant to extinction probability, then the probability of a coral going extinct in the Plio-Pleistocene, Pp(x), was 0.42. The probability that a modern species was listed among the three least resilient corals in the Indo-Pacific, Mp(x), was 0.21. Therefore, there was approximately a 9 per cent chance (95% Bayesian credible intervals 2–16%), Pp(x).Mp(x), that a coral would have gone regionally extinct through the Plio-Pleistocene and that it was also at the bottom of the resilience list. Against the odds, this happened for three taxa: Stylophora, foliose Pavona and Pocillopora. With three vulnerable genera at risk today having also experienced Plio-Pleistocene extinction, this result strongly rejects a non-selective, random mechanism for extinction events and supports the proposition that biological traits play a critical role in determining vulnerability. Indeed, we identified a strong relationship between the fate of specific coral genera during the Plio-Pleistocene and their contemporary resilience score for the modern Indo-Pacific (figure 2). For the eight coral genera examined, the four that went regionally extinct during the Plio-Pleistocene all possessed negative resilience scores, while the four extant taxa all had positive scores. The significant association between extant Plio-Pleistocene coral genera and positive resilience scores (penalized logistic regression-likelihood ratio χ2, p = 0.01) supports the use of a biological trait-based framework for evaluating coral extinction risks. These associations also support the notion that coral extinction are trait-based, further rejecting neutral theory that suggests coral extinctions are a random process.
If both the IUCN risk of extinction scores  and our trait-based approach accurately predict the risk of extinction, we would expect a direct negative relationship between our trait-based resilient scores, where positive numbers suggest resilience, and positive IUCN scores suggest high extinction risk. We found this general relationship (Kendall's tau = −0.32, p = 0.09) among the rank correlation of the IUCN status and resilience scores, but it was not statistically significant. Most genera exhibited a concurrence between the trait-based score and IUCN status but Stylophora, and the Montastraea annularis complex were prominent outliers (figure 3). We suggest that Stylophora is more threatened than is predicted by the IUCN scores, whereas the Montastraea annularis complex is more resilient than is predicted by the IUCN scores (figure 3).
Understanding the vulnerability of corals to thermal stress is critical to management efforts in rapidly warming oceans. We compared contemporary vulnerability of modern corals with fossil data on extinctions from the Plio-Pleistocene, which experienced dynamic climate fluctuations. The strong relationship between past trajectories and modern vulnerability shows that inherent genealogies can guide our predictions of the possible future state of coral communities. Furthermore, our trait and historical approach may offer another method to evaluate vulnerability of coral taxa and augment existing methods.
Biological trait-based approaches have been recently applied to the evaluation of extinction risk among a variety of taxa including terrestrial plants , geometrid moths , desert fishes  and amphibians . Our study used a similar strategy but also benefited from, and was limited by, the inclusion of fossil data for historical validation of extinction vulnerabilities of corals. The primary caveats of working with fossil data were the limited taxonomic scope available [9–12] and implicit assumption of trait conservation for extant lineages. Furthermore, our expert-based approach, which qualified biological traits of coral species and processes, is in need of further quantification for all extant coral species. Nonetheless, general patterns observed across the breadth of taxa included in the analysis support the use of biological trait-based frameworks for the determination of extinction risk in corals.
The past episodes of regional extinction were strongly related to the vulnerability of modern corals. On modern reefs, Stylophora, Pocillopora and foliose Pavona are all highly sensitive to temperature anomalies [34,35], salinity changes  and irradiance extremes . Yet, encrusting and submassive Pavona (including Pavona varians, Pavona decussata and Pavona venosa) were not sensitive to warm temperate events in the 1998 warm anomaly . The percentage cover of Pavona and Pocillopora has increased on Kenyan reefs over the last 20 years . Nevertheless, in the unusually warm and variable regions of the Arabian Gulf, these taxa are either not present, or patchy and their numbers are too low to be measured by standard ecological transects [39–41]. If the Arabian Gulf is a good contemporary analogue for future reefs, then the patchy success of Pavona and Pocillopora on some Kenyan reefs may be short-lived in the warmer and more variable conditions predicted for the future. The consistent trends between the past and the present suggest that regional extinctions were not random episodes , but were instead a consequence of genealogies and associated traits that were vulnerable to climate fluctuations.
Estimates of the partial pressure of carbon dioxide (pCO2) in the early Pliocene are higher than any other part of the Neogene , with mean sea-surface temperatures in the Caribbean estimated to be 1–2°C warmer than present . Final closure of the Central American Seaway (4.2-3.4 Myr ago) isolated the Caribbean from the Indo-Pacific and corresponded with increased salinity, increased carbonate precipitation and more oligotrophic conditions in the Caribbean [44–46]. Interestingly, the main extinction peak for both molluscs and corals lags approximately 2 Myr behind final closure of the Central American Seaway [9,15,47,48]. The extinction peak occurred during a period of global cooling (glaciation) and an eustatic sea-level drop that changed oceanic circulation and increased oligotrophy .
Irrespective of the timing of the extinctions, and judging by the modern responses, stenothermal species, such as Stylophora and Pocillopora, were likely to be vulnerable to the highly fluctuating climate, whereas eurythermal species, such as Indo-Pacific Montastraea, Porites and Favia, may have simply tolerated the fluctuating conditions. Nevertheless, persistence can be a convergent trait, being dependent on local tolerance and regional ubiquity. Local thermal tolerance is attributed to the biological properties of the host and the symbionts, such as massive growth forms, thick tissue, low-metabolic rates and whether corals support thermally tolerant symbionts . Stylophora and Pocillopora are branched, with thin tissue , and heat-tolerant endosymbionts seldom dominant these taxa . However, regional persistence is also attributed to the life-history properties of corals, including their fecundity and growth rates. Weedy species with high fecundity and rapid growth are predisposed to regional success, especially in areas that support a variety of habitats that are frequently disturbed. Recent studies show that a number of species that were short-term losers under thermal stress, such as Acropora digitifera and Acropora gemmifera, turned out to be long-term winners, in part, because of their ubiquitous distribution .
The upward trajectories of Acropora through the Plio-Pleistocene do not match the declines of Acropora in the modern Caribbean. The two species of Acropora found on modern Caribbean reefs (Acropora palmata and Acropora cervicornis) are highly vulnerable to extinction, with no opportunity for redundancy. By contrast, there are over a hundred modern species of Acropora in the Indo-Pacific region, with varying levels of susceptibility to thermal events [51,52].
It could be argued that given the unprecedented rate of change in the modern climate, historical events in the Plio-Pleistocene may tell us little about future trajectories. Yet, our comparison across time shows consistent patterns, indicating that conserved genealogies allow a glimpse into near-future trends. Stylophora and Pocillopora may change from local dominance and a ubiquitous distribution in the Indo-Pacific region, to a more sparse distribution and local rarity. Isolated reef aggregations in the Pacific and Indian Oceans and those reefs with few neighbours may be particularly vulnerable to climate change, and may lose corals such as Stylophora and Pocillopora. By contrast, localities in the Pacific and Indian Oceans supporting diverse habitats and dense reef aggregations may continue to support climatically sensitive coral species.
Our capacity to successfully predict the risk of coral extinction will ideally lead to management policy that reduces such threats. Importantly, we would expect that coral species most at risk, with high IUCN scores, would show a strong negative relationship with our trait-based resilient scores. However, several inconsistencies were found when using the IUCN criteria. Therefore, predictions of extinction risk based on the IUCN criteria should be re-evaluated to incorporate biological traits and ecological processes. Progress towards this goal would be accomplished by a comprehensive accounting of the biological traits of coral species, quantifying changes in local and regional abundance patterns, and determining geographical range changes through time.
Many thanks extend to Sandra van Woesik and Peter Edmunds for editorial comments and Shirley Han for research assistance. Funding provided by the NSF (DEB-0102544, DEB-9705199, EAR-9219138, EAR-0445789 to A.F.B.; EF-0553768 to NCEAS for ‘Tropical coral reefs of the future: modelling ecological outcomes from the analyses of current and historical trends’ Working Group, awarded to Ruth Gates and Pete Edmunds; OCE-1041673 to M.J.D. for E.C.F., NOAA NMSP (MOA 2005-008/66882 for E.C.F.), the US EPA (FP917096 to E.C.F.). This is HIMB contribution no. 1483 and SOEST contribution no. 8550. This is contribution 61 from the Institute for Research on Global Climate Change at the Florida Institute of Technology.
- Received December 15, 2011.
- Accepted January 27, 2012.
- This Journal is © 2012 The Royal Society