Genetic seascape of the threatened Caribbean elkhorn coral, Acropora palmata, on the Puerto Rico Shelf

Sampling

Twenty-four reefs were located in Puerto Rico (including Mona, Culebra and Vieques) (Fig. 1, Table 1). Special effort was dedicated to (i) alternate small geographic distances (few kilometers) with moderate distances (tens of kilometers) between reefs and (ii) select reefs from areas all over the Puerto Rican archipelago in a comprehensive design. The other six reefs represented samples from the eastern (Curaçao, Guadeloupe) and western populations (the Bahamas). All samples were taken between 2006 and 2009 and were collected opportunistically (non-randomized pattern). Particular efforts were made to sample both potential clone colonies (various ramets of a same genet) and potentially different genotypes. Hence, for each reef we sampled tissue from colonies within a 5-m radius (likely to be clones) as well as colonies separated by tens to hundreds of meters (unlikely to be clones). Whenever possible, 20–50 colonies per reef were collected, preferentially by snapping the tip of branches. Samples from 412 colonies were obtained, including 86 from Garcia Reyes & Schizas (2010).

Molecular techniques

From each sample, 5–10 polyps were cut off and total genomic DNA was extracted with the DNeasy Blood & Tissue Kit (Qiagen, Valencia, CA, USA) following the manufacturer's animal tissue protocol. Each sample was then screened for five polymorphic microsatellite markers, following a modified protocol from Baums et al. (2005a). The selected markers (#166, #181, #182, #192 and #207) were the same as those used in Baums et al. (2005a, 2006a,b). PCR amplifications were done in 10-μl reactions, containing 1 μl genomic DNA (5–15 ng·μl⁻¹), 0.8 mm dNTPs, 0.1 μm of forward primer with M13 tail, 0.1 μm of M13 fluorescently labeled with FAM (markers #166 and #182) or HEX (markers #181, #192 and #207), 0.2 μm of reverse primer, MgCl₂ (2 mm), 0.3 μl of 1 U·μl⁻¹ Taq DNA polymerase (Fermentas, Vilnius, Lithuania), and 1× of the PCR buffer. Temperature cycling was performed by denaturing 1 min at 94 °C, followed by 20 cycles of 20 s at 94 °C, 35 s at 56 °C and 30 s at 72 °C, 15 cycles of 20 s at 94 °C, 35 s at 50 °C and 30 s at 72 °C and a 10-min extension step at 72 °C. Amplicons were diluted up to 50× to approach 10–20 ng·μl⁻¹, pooled whenever possible (#166 with #207 and #182 with #192) and were run on an ABI3130xl Genetic Analyzer with ROX-labeled size standards. Microsatellite alleles were scored using the software GENEMAPPER^® 4.0 (Applied Biosystems, Carlsbad, CA, USA).

Genetic diversity and structure

The probability of identity (PI) is the probability that two genetically different samples have identical multilocus genotypes given a set of genetic markers. Computation of PI was performed in GENALEX 6.1 (Peakall & Smouse 2006). Identical multilocus genotypes were then considered ramets of the same genet (clones of a same genotype) with a confidence probability PI. All subsequent analyses were performed by reducing the dataset to the number of unique genets. Because ramets were represented by a single genet, the final dataset had no further information on genotype frequency. Genetic indices of diversity, tests of linkage disequilibrium (50,000 permutations), pairwise F_ST between localities (50,000 permutations) and Hardy–Weinberg disequilibrium (HWE, 10,000 burn-in, 1,000,000 permutations) were estimated in ARLEQUIN 3.5 (Excoffier & Lischer 2010) and P-values were adjusted to control for false discovery rate (FDR; Benjamini & Hochberg 1995) with the stats package in R (R Development Core Team 2010). F_ST was preferred to R_ST because it is a more suitable measure of genetic distance between populations when the number of markers is <20 (Gaggiotti et al. 1999). The number of K populations was estimated in STRUCTURE 2.3.3 (Pritchard et al. 2000; Hubisz et al. 2009) and repeated 10 times for each value of K, ranging from K = 1 to K = 5. Using the FullSearch algorithm in CLUMPP 1.1.2 we permuted the independent replicate runs (Jakobsson & Rosenberg 2007). The mean of the permuted matrices across replicates was visualized in DISTRUCT 1.1 (Rosenberg 2004). To find the number of discrete populations in the dataset, we used STRUCTURE under three different models. The model ADM used only allelic information to find population structure. In contrast, the model POPINFO added a priori population assignments for some individuals to assign the remaining individuals to the a priori populations. Finally, the LOCPRIOR model enabled the incorporation of the sampling locations as a priori information. While these analyses were performed assuming 14 sampling localities (see Table 1), analyses were also repeated using the alternate ‘reef’, ‘region’ and ‘island’ grouping for the LOCPRIOR model. STRUCTURE analyses, including the different methods to select the number of populations K that best describe the dataset, are further detailed in the Supporting Information (Data S1 and accompanying Figs S1 and S2). A matrix of genetic distances F_ST/(1−F_ST) was generated and tested against the geographic distance matrix to explore IBD patterns. Reflecting the short pelagic life of their larvae, Acropora palmata populations seem to be largely self-recruiting, and connectivity across large distances seems to approximate a rather one-dimensional path following shallow water habitats (Baums et al. 2006b), in particular along the Lesser Antilles. Hence we generated a geographic distance matrix based on the shortest shallow nautical (SSN) distances in Google EARTH 6.2.1.6014 at the 1,000,000:1 scale. Alternatively, another matrix based on the shortest nautical (SN) distances was constructed to explore IBD patterns involving direct connectivity, in particular for distant reefs such as the Bahamas, Guadeloupe and Curaçao. Matrices are appended in Table S1. Significance of IBD patterns was tested with Mantel tests (30,000 permutations) and reduced major axis (RMA) regressions were used to estimate the strength of genetic differentiation (RMA slope, jackknife 95% CI). Because the data are only known with error, RMA was favored over ordinary least-square regression (Hellberg 1994; Jensen et al. 2005). Mantel tests and RMA calculations were performed in IBDWS (Jensen et al. 2005) by comparing the matrix of pairwise Slatkin's F_ST (F_ST/1−F_ST) with the geographic distance matrices after setting genetic distances of <0–0. A log₁₀ transformation was applied to the SN matrix to account for bi-dimensionality (Slatkin 1993; Rousset 1997). On the other hand, connectivity between distant reefs in the SSN matrix was rather one-dimensional since it followed shallow water habitats (from one island to the next) and thus no logarithmic transformation was applied to the SSN matrix (Rousset 1997). Comparisons were first made with all studied localities but because some populations had fewer individuals (n < 15), part of the observed patterns were not highly supported (Cornuet et al. 1999). As a balance between improved statistical confidence and data loss, the results were compared again by including only localities with a certain minimum of different genotypes (n = 15, n = 20 and n = 30). When the data could not be shaped into a usable matrix for IBDWS, RMA for JAVA (Bohonak & Van Der Linde 2004) was used instead (one-delete jackknife with 30,000 bootstraps for 95% CI), as was the case for RMA calculations excluding within-Puerto Rico pair-wise comparisons or to obtain the regression slopes between a specific location (Mona or Culebra) and mainland Puerto Rico.

Clonality and genetic diversity

Based on our dataset, the probability for two genetically different samples to have identical multilocus genotypes by chance (PI) using the five microsatellite markers was approximately 1.48 × 10⁻⁷, which closely matched the approximately 1.5 × 10⁻⁷ estimate in Baums et al. (2005b, 2006a). Therefore, it was a reasonable assumption that identical genotypes represented biological clones. We identified 309 unique microsatellite multilocus genotypes (genets; 75%) and an additional 103 clones (25.0%) from a total of 412 colonies (Table 1). To avoid artificial, misleading signals resulting from the presence of those clones in our dataset, the remaining analyses, including estimates of genetic diversity and population structure, were based on the 309 unique genets identified in this study.

Mean genetic diversity across localities and loci was high (H_E = 0.761) and consistent across localities (0.685–0.844). Mean allelic diversity per locality across loci was 9.7 (Table 2). Allelic diversity ranged from 8 to 23, from the least (#181) to the most (#166) polymorphic locus. Mean allelic range per location across loci was 11.3, ranging from 13 to 24 depending on the marker (#181 and #166, respectively). After correcting for multiple comparisons with the FDR, no test of pair-wise linkage disequilibrium was significant (140 pair-wise tests comparing all loci at each of the 14 localities). Therefore, the five loci were assumed to be unlinked. FDR-corrected P-values for tests of HWE at each locus were not significant (70 tests, five loci for 14 localities), indicating that the sampled populations were at equilibrium.

Number of populations with STRUCTURE

Simulations performed under the ADM model (without a priori information on sampling locations) had a higher probability L(K) for K = 1 (Fig. 2A and B). Furthermore, the summary statistics α and F did not stabilize, indicating that STRUCTURE did not detect population structure with this algorithm (Pritchard et al. 2000), as visually confirmed in Fig. 3A. This contrasted with the findings of Baums et al. (2005b) who found an optimal number of population K = 2 using the same algorithm. In the POPINFO model, which assigns some genotypes to user-specified a priori populations, the results were similar (Figs 3B and S2D and E). Again, the summary statistics did not stabilize, indicating that STRUCTURE could not find a likely assignment for the given genotypes. Under the LOCPRIOR model, which uses sampling locations as a priori information, the optimum burn-in period was found to be relatively long, with r (the contribution of the predefined sampling locations to the end probabilities assigned to the individuals), α and F generally stabilizing after 5–10 million iterations. These control steps were necessary to have confidence in the resulting probability for K (Hubisz et al. 2009). Also, the variance of L(K) within and between runs was relatively high when K > 2, requiring a high number of iterations and the use of replicate runs. Based on the arithmetic mean of L(K) across replicate runs, the optimal number of K found by STRUCTURE in all models was K = 2 (Fig. 2H). This was also true for other recommended methods, such as using only the best L(K) among replicate runs (Fig. 2G) or the ∆K parameter described in Evanno et al. (2005) (Fig. 2I). When performing alternate LOCPRIOR analyses, i.e. by assigning individuals to a priori locations corresponding to ‘reefs’, ‘regions’ or ‘islands’ (Fig. 2G and H), the original ‘localities’ model was generally found to reach superior probabilities across the range of tested K (1–5), although at K = 2, the ‘reef’ model had a slightly higher L(K) than the ‘localities’ model (mean L(K)_reefs = −6412 and mean L(K)_localities = −6421, respectively). Following the population model at K = 2, the Bahamas, Mona, Rincón, Lajas, Guánica, Ponce, Isabela and Vega Baja were clustered into one population and Curaçao, Guadeloupe, Culebra, Fajardo clustered into another. Finally, Vieques and the San Juan area had admixed origins of populations (Fig. 3C).

Genetic structure and isolation-by-distance

The genetic distances between localities were small (F_ST ranging from −0.011 to 0.047) but often significant, even after correcting for multiple comparisons (Table 3). Conversely, they were not significant for those locations with the lowest numbers of genets, such as Guánica or Isabela (n = 3 and n = 5, respectively). In general, genetic distances were higher between the most distant localities, in conformity with a model of IBD. In the west, pair-wise F_STs indicated that despite a relatively low number of genotypes (n = 11) the Bahamas showed significant divergence with all localities except Mona, Rincón and Vega Baja. Interestingly, Mona showed no significant difference from the closest localities in Eastern Puerto Rico (e.g. Rincón, Lajas). In the east, Culebra (n = 19), Guadeloupe (n = 47) and Curaçao (n = 10) had non-significant F_STs among them. On the other hand, these locations exhibited significant pair-wise differences with most other localities.

IBD plots showed consistent patterns of genetic differentiation via IBD, regardless of the type of geographic distance matrix or of the minimum sample size for each population (Fig. 4). Mantel tests were high and significant (P < 0.05) except when minimum sample size per population was set to 15 different genotypes. In such cases, removing the pair-wise comparisons involving localities outside Puerto Rico effectively resulted in significant Mantel test results within Puerto Rico (Fig. 5, dashed line r = 0.53, P = 0.0064), suggesting different patterns and strengths of differentiation within and outside Puerto Rico. This was supported by marginally non-overlapping confidence intervals ranges (95% CI) between the slopes of their respective RMA regression, respectively 6.6 × 10⁻⁵ to 1.7 × 10⁻⁴ for within Puerto Rico pair-wise comparisons and 1.9 × 10⁻⁵ to 6.5 × 10⁻⁵ for the remaining pair-wise comparisons (Fig. 5). However, there was no clear, separate clustering pattern characteristic of discrete populations (Figs 4 and 5). Ideally, population structure in discrete populations would show separate clustering of within-population genetic distances from clustering of between-population genetic distances. However, an intermediate clustering pattern can be observed: for near-identical geographic distances, pair-wise comparisons between a priori western and eastern populations in Figs 4 and 5 (orange symbols) were characterized by higher genetic distances than pair-wise comparisons within each of these groups (light blue and blue symbols, respectively). Although stronger statistical support could be achieved by increasing the low number of genets in the Bahamas and Curaçao (respectively n = 11 and n = 10), pair-wise F_STs per 1000-km unit were higher within Puerto Rico than when locations thousands of kilometers apart were compared, supporting the presence of a zone of further differentiation in Puerto Rico (Table 3). In this regard, Culebra was characterized by the highest F_ST per 1000 km. In fact, the RMA regression slope of pair-wise comparisons between Culebra and mainland Puerto Rico (s = 1.2 × 10⁻⁴, SE = 2.5 × 10⁻⁵) was superior to the slope for Mona-mainland Puerto Rico (s = 7.0 × 10⁻⁵, SE = 2.7 × 10⁻⁵), although their respective 95% CI overlapped (8.3 × 10⁻⁵ to 1.5 × 10⁻⁴, and −4.6 × 10⁻⁵ to 1.9 × 10⁻⁴, respectively).

Sampling strategy and clone proportions

Clone distribution in the three-dimensional reef space depends on a variety of factors (Coffroth & Lasker 1998). For example, the genetic disposition of individual genets is likely to be important for the successful settlement of new ramets. Environmental factors such as hurricane disturbance, reef orientation and inclination, current dynamics and competition for space with other reef organisms will be responsible for part of the observed clone distribution, frequency and density (Highsmith et al. 1980). Because we wished to avoid overrepresentation of clones for the benefit of genetic structure analyses, we favored a non-random, opportunistic sampling strategy. Hence, population dynamics implications based on the frequency and density of ramets in this study should be interpreted with caution. We found that unique genets represented 75% of the samples (Ng/N = 0.75, mean Ng/N per locality = 0.75). In contrast, Baums et al. (2006a) used randomized circle plots and opportunistic sampling and found that both sampling strategies yielded similar results (mean Ng/N per reef = 0.52 and 0.51, respectively). Since genotypes were scored with the same markers in both studies, differences at the molecular level are unlikely to explain the difference in the proportion of clones. Because clonality varies greatly among reefs, the choice of sampling localities might explain some amount of discordance. The rest of the differences can probably be explained by a sampler effect (personal preferences for certain coral colonies during sampling) and/or other uncontrolled factors, e.g. the depth of sampling. A lower value of unique haplotypes (42%) was estimated from reefs of West and Southwest Puerto Rico, where 46 unique mitochondrial haplotypes were detected from 110 distinct colonies (Garcia Reyes & Schizas 2010). The difference with the current results is not surprising because microsatellites are usually more variable than mitochondrial markers, detecting more unique genets (Baums et al. 2005a).

The frequency of clones in each locality was unrelated to a purely geographic division between western and eastern provinces. Rather, difference in clonal structure is more likely explained by differences in environmental conditions such the size and depth of the shelf area (Baums et al. 2006a). The areas with the most clones were found in shallow, extensive shelves, such as Vega Baja and Escambrón. The back reefs of these locations varied in depth between <1 and 2 m, providing asexual recruits with space to settle and shelter from waves. Tres Palmas was also well represented by clonal genets, although its shallow shelf is short and rapidly reaches unsuitable depths for dense Acropora palmata stands. In contrast, reefs with no or few clones were often representative of areas with less suitable habitat. For example, the Anse-Bertrand reef is positioned against a small cliff, and elkhorn coral colonies were found at depths where they were growing in an encrusting fashion, limiting their ability to produce asexual recruits via fragmentation. Reefs at Piñones and Sun beach simply had no space for settlement of asexual recruits. Reefs at Shack, Pointe-des-Châteaux and Grand-Cul-de-Sac-Marin were positioned in shallow, extensive shelves, but numerous dead stands of A. palmata separated the sampled colonies, most likely the ghosts of disease or bleaching in the past. In the event of disease outbreaks, sensitive genets will disappear first (Reusch et al. 2005). Hence, because of their low genetic diversity, highly clonal reefs are more likely to lose extensive coral cover compared with heterogeneous reefs.

High genetic diversity

An important step in the genetic diversity and structure analyses is that the frequency of each genotype was removed from the dataset prior to analyses. This precaution avoided unreasonable assumptions for the estimation of gene flow and genetic structure (Baums et al. 2005b). In A. palmata, asexual reproduction happens by fragmentation when broken branches rise into new, identical clones near the original colony (Bruckner 2002; Fong & Lirman 2005; Reusch et al. 2005). Thus, the genetic fingerprint of clones within a reef will mostly be the result of environmental hazards, not of sexual reproduction and gene flow. On the other hand, the probability of inbreeding and gene flow depends on the effective population size, and genets having many ramets will contribute more to sexual reproduction (Coffroth & Lasker 1998). Hence, some amount of information is inevitably lost in the analysis. The genetic diversity found in this study (H_E = 0.761) was higher than the genetic diversity in Acropora nasuta (MacKenzie et al. 2004) or Acropora cytherea (Concepcion et al. 2010) but surprisingly lower than for Acropora muricata and Acropora digitifera using the same microsatellite markers used in this study (Tang et al. 2010). On the other hand, our results were consistent with those of Baums et al. (2005b) for A. palmata. Interestingly, genetic diversity was also consistent across all localities in our study and was higher than the genetic diversity found in a study of 14 rare and common species of Indo-Pacific Acropora using nine neutral microsatellite markers (including the five markers used in the present study; Richards & van Oppen 2012). With respect to conservation efforts, these results suggest that there is high genetic diversity in A. palmata despite the dramatic losses during the last decades, in particular due to White Band disease (Bruckner 2002). However, the high genetic diversity estimated with microsatellites is not reflected in the mitochondrial nucleotide diversity (π = 0.00075) of A. palmata in Puerto Rico, with among the lowest reported values for scleractinian corals (Garcia Reyes & Schizas 2010). Part of the discrepancy can be explained by the low levels of genetic variability observed in mitochondrial DNA of corals compared with nuclear genes (Hellberg 2006).

Clustering of western and eastern populations in Puerto Rico

In the study of Baums et al. (2005b), two populations, roughly divided east and west of Puerto Rico, were detected using STRUCTURE. Sampling around the Mona Passage, the proposed region of population admixture, was limited to 36 unique genets from Mona and 90 unique genets along the west and southwest coast of Puerto Rico (Rincón, Lajas, Bajo Gallardo). To improve our understanding of population differentiation in the region, further sampling efforts in the south of Puerto Rico and in the Dominican Republic were conducted in Baums et al. (2006b). In their study, Western and Southwestern Puerto Rican reefs were pre-assigned to the eastern population via the POPINFO option in STRUCTURE, in disagreement with the initial STRUCTURE results of Baums et al. (2005b), where the same Puerto Rican reefs clustered with the western population when no a priori assignments were made. The added samples from the Dominican Republic clustered with the western population (Baums et al. 2006b) but the a priori assignment of Western and Southwestern Puerto Rican samples to the eastern cluster did little to improve our understanding of population differentiation around Puerto Rico. By including new samples and locations around the proposed region of population admixture (Baums et al. 2005b), in particular where sampling was missing (Northern and Eastern Puerto Rico), the present study further details the population structure of elkhorn coral in this region of particular interest.

In the present study, in accordance with high genetic diversity and low estimates of population structure (F_ST, AMOVA analyses presented in Supporting Information in Data S2 and accompanying Table S2), genetic differentiation was detected with STRUCTURE but was weak: the basic admixture model (ADM) in STRUCTURE was not able to detect population structure in the data, and a high number of iterations were needed to reach statistical stability and determine population structure when using prior information on locations via the LOCPRIOR model, recommended in case of weak population structure (Hubisz et al. 2009). In contrast, Baums et al. (2005b) were able to find population structure separating a western from an eastern population using the same markers and admixture model without a priori information, and thus suggested that two populations of Acropora palmata meet at Puerto Rico. Differences between the two studies might be due to the more extensive geographic coverage and larger number of unique genotypes in Baums et al. (2005b) than in this study (n = 709 and n = 309, respectively) or user-related differences in the assessment of marker states.

Baums et al. (2005b) performed additional runs in STRUCTURE, using the built-in POPINFO option to define a priori assignments of non-Puerto Rican localities to their respective eastern or western cluster, allowing the software to estimate population admixture of the unassigned Puerto Rican genotypes. The resulting assignments of evenly admixed Puerto Rican colonies were then interpreted as hybrid genotypes between the eastern and the western populations. Similarly, we used the POPINFO option to define a priori assignments of selected locations. Various configurations combining western and eastern a priori assignments invariably resulted in (i) lack of stabilization of control parameters during the runs and (ii) similar patterns of totally admixed genotypes, in equal proportions between K populations, for all individuals that were not assigned to a priori cluster. This pattern suggests that those admixed assignments resulted from a lack of discriminative power of the algorithm, and represented undecided assignments rather than perfectly admixed genotypes (with 1:1 proportions of each population of origin in the case of K = 2; Pritchard et al. 2000).

Using the LOCPRIOR option in STRUCTURE, which was not available when Baums et al. (2005b, 2006b) published, we further detailed the population structure of A. palmata in Puerto Rico. Our results agree with Baums et al. (2005b), with K = 2 having the highest probability to explain the data, even when reefs outside Puerto Rico were not included. In Puerto Rico, the western population would include Mona Island and the west coast of Puerto Rico, as was described in Baums et al. (2005b). The western population most likely extends (i) northwest and north of Puerto Rico reaching past Vega Baja; (ii) southwest and south past Lajas and Guánica, reaching past Ponce. In contrast, the eastern population probably includes Culebra and reaches past Fajardo. A transitional area occurs somewhere between Vega Baja and Fajardo in the north, explaining the admixed profiles of the San Juan area. In the south/southeast of the island, the transitional area is located between Ponce, Fajardo and Vieques. Keeping in mind that the low number of genets for some of these localities might affect our results (Cornuet et al. 1999), we also find significant F_ST and AMOVA results supporting the genetic structure evidenced by our STRUCTURE results.

Patterns of IBD versus discrete populations

STRUCTURE explicitly allocates individuals into an a priori number of groups that are discrete and whose members minimally violate the assumptions of HWE (Pritchard et al. 2000). In instances where genetic differentiation is correlated with geographic distance, individuals separated by sufficiently large geographic distances may violate the assumptions of a population at HWE when placed into one group, and therefore two or more groups may better explain the clustering of these individuals. However, this does not mean that these groups are discrete, and that individuals with intermediate genotypes represent admixed individuals between two discrete populations, but rather indicates that the program is forcing continuous variation into discrete and discontinuous clusters. Because Bayesian clustering programs such as STRUCTURE can overestimate the number of clusters in datasets characterized by IBD, IBD should also be tested and interpretations based on the results of all analyses (Frantz et al. 2009).

STRUCTURE analyses indicate two clusters with genetically intermediate individuals and localities occurring in Western Puerto Rico (Fig. 3). This result is broadly comparable to that of Baums et al. (2005b, 2006a). However, our interpretation, based mainly on the addition of localities and samples in the proposed area of admixture, differs slightly. Analyses of pair-wise F_ST and of correlations between genetic and geographic distances suggested a pattern of IBD, which normally characterizes populations with limited connectivity, such as in a stepping stone model (Hellberg 2007). This is consistent with what is now known of limited larvae dispersal in corals and other marine species (Palumbi 2003; Cowen et al. 2006; Galindo et al. 2006; Hellberg 2007; Andras et al. 2013). Furthermore, the choice of a different geographic distance matrix or limiting analyses to localities with a certain minimum sample size (no minimum, 15, 20 or 30 genotypes) did not affect the interpretation of genetic differentiation by IBD. Our data suggests that the strength of IBD measured as the slope of the correlation between genetic and geographic distances differs along the studied seascape inhabited by A. palmata. The slope was steeper in comparisons involving the small geographic distances around Puerto Rico than Caribbean-wide comparisons, suggesting that IBD is much stronger within the region of Puerto Rico than outside this region. This could reflect the stronger genetic changes associating with the hybridization of the western and eastern populations in the Puerto Rican region (Baums et al. 2005b). While STRUCTURE consistently placed the Bahamas within the putative western population along several Puerto Rican localities, the Bahamas were found to be significantly differentiated from almost all other locations in this study, a result reminiscent of the genetic separation of the Bahamas from other Caribbean locations in A. cervicornis (Galindo et al. 2006; Vollmer & Palumbi 2007; Garcia Reyes & Schizas 2010; Hemond & Vollmer 2010) or Orbicella (previously Montastraea) annularis (Foster et al. 2012). East of Puerto Rico (Culebra, Guadeloupe and Curaçao) there was little genetic divergence between localities, suggesting near-panmixia within the eastern region (Table 3), a result largely consistent with the eastern cluster described by Baums et al. (2005b) but also with the strong separation of eastern locations from the west in both experimental data and dispersal models in O. annularis (Foster et al. 2012).

Although our findings within Puerto Rico were well supported, further studies should be conducted to confirm our observations regarding non-Puerto Rican localities because of the low number of sites (n = 3) and genets per site (n = 11 in the Bahamas and n = 10 in Curaçao) analyzed outside of Puerto Rico in this study.

Re-assessing population structure in the elkhorn coral

Genetic structure in this study was characterized by a non-uniform IBD across the seascape inhabited by A. palmata, suggesting that connectivity, at least among some A. palmata reefs, is limited. IBD seemed to be stronger among localities from Puerto Rico, in particular in Eastern Puerto Rico. IBD was weaker to insignificant outside of the Puerto Rico Shelf, as suggested by mostly non-significant genetic structuring within the eastern population. When the distribution of the genetic diversity is interpreted in the framework of a discrete population structure as implemented in STRUCTURE, one then observes a zone of transition in Eastern Puerto Rico between two apparently discrete genetic groups. This zone of transition did not seem to associate directly with the Mona Passage, west of Puerto Rico, which was suggested to be a significant barrier to gene flow in A. palmata (Baums et al. 2006b). Foster et al. (2012) also found an east–west break in the genetic structure of O. annularis but noted that its location was ambiguous, since their results pointed to a separation between the British Virgin Islands and Dominica. Understanding why the genetic transition should be more pronounced in the east of Puerto Rico, as suggested by the slopes of RMA regressions between mainland Puerto Rico and Mona versus mainland Puerto Rico and Culebra, is an intriguing challenge. The filtering effect of the Mona Passage, coupled with asymmetric net gene flow in the easterly direction, as is generally the case for surface currents during the spawning season on the north coast of Puerto Rico, could provide an explanation. Admittedly, the genetic structure in the south of the island would then be expected to respond to gene flow following opposite currents in the westerly direction, which does not seem to be the case, unless gene flow in the southeast is hindered by another, unidentified barrier. Residual genetic structure inherited from historical shifts in the geographical ranges of Caribbean species during the late Quaternary (Lighty et al. 1982; Toscano & Macintyre 2003) could explain some of the present genetic patterns, as did the last glaciation event on Indo-Pacific population dynamics of marine species (Benzie 1999). Fewer than 15 genotypes were recovered from two of the three non-Puerto Rican localities in this study (Curaçao and the Bahamas). Because STRUCTURE and other analyses of population structuring are context-dependent, adding molecular markers, new locations in the larger Caribbean and increasing sample size per population seem essential for further improved analyses (e.g. when assigning individuals to populations of origin, see Cornuet et al. 1999). These additions will likely provide us with a better understanding of possible IBD and other genetic structuring of A. palmata populations outside of Puerto Rico. New sampling efforts concentrated in the northeast, southeast and east of Puerto Rico are also needed to identify more accurately the location(s) of the strongest genetic shift(s) between eastern and western populations. Meanwhile, building on the knowledge that most of the extant genetic diversity of A. palmata is present on the Puerto Rico Shelf, we exhort local decision-makers to implement new conservation measures for the elkhorn coral, as effective protection at a local scale could be beneficial for preserving the global genetic diversity of A. palmata.