Seascape genomics uncovers contrasting population genetic structures in reef corals

Summary

Abiotic conditions influence biodiversity and population connectivity, the assessment of which is critical for predicting climate change impacts and guiding conservation efforts. Coral reefs face severe losses, making effective conservation strategies increasingly urgent. Here, we use a seascape genomics approach to examine the population genetic structure of three non-model scleractinian corals around the Arabian Peninsula, a region with distinctive oceanographic features. We reveal contrasting patterns across species, which we integrate with geographic distances, larval dispersal models, and environmental data. Population structure appears shaped both by isolation by distance (IBD) and isolation by resistance (IBR), with environmental gradients often covarying with spatial separation. Our findings clarify coral diversification and evolutionary responses to conditions around the region. This work also provides a methodological framework to advance our understanding of processes structuring reef populations, thereby informing efforts to design robust marine protected area networks in the face of climate change and biodiversity loss.

Graphical abstract

Highlights

• •Seascape genomics to explore reef coral population differentiation in a unique region
• •Different species show contrasting genetic patterns across the Arabian Peninsula
• •Genetic breaks align with geography, ocean currents, and spawning timing
• •Isolated coral populations highlight the need for targeted conservation efforts

Ecology; Genomics; Evolutionary biology

Introduction

Tropical shallow coral reefs support extraordinary biodiversity^1,2 and provide services essential to many societies.³ These ecosystems, however, are among the most threatened by climate change^4,5: about 30% of 1980s reefs have already disappeared, and up to 90% will disappear by 2050 without adequate mitigation.⁶ Effective conservation strategies, based on accurate scientific knowledge of ecosystem functioning, have become necessary to secure a future for coral reefs.⁷

One of the key structural and functional components of coral reef ecosystems are reef-building corals. Estimating their adaptive potential and connectivity, through the characterization of genetic diversity and structure,⁸ is crucial for designing strong marine protected area networks capable of counterbalancing biodiversity loss.^7,9 Additionally, understanding how abiotic conditions influence gene flow, and ultimately shape biodiversity,¹⁰ is crucial for predicting large-scale connectivity¹¹ and future¹² species distributions under climate change, thereby enhancing the long-term success of conservation efforts.

Despite growing interest in coral connectivity, its underlying determinants remain poorly understood. Connectivity can be influenced by both environment and transport, and when these vary, their relative importance can be non-intuitive,¹³ but has rarely been assessed. The Arabian Peninsula (AP; Figure 1A) is a region bordered by a complex circulation and different water bodies (counterclockwise: Gulf of Aqaba, Red Sea, Gulf of Tadjoura, Gulf of Aden, Arabian Sea, Gulf of Oman, and Arabian/Persian Gulf; Figure 1C), each with unique, sometimes extreme, oceanographic conditions,^14,15,16 that can also serve as model to understand reef response under future climate scenarios. For example, coral communities of the Red Sea¹⁷ and Arabian/Persian Gulf¹⁸ in summer are facing the world’s warmest conditions, exceeding those projected by climate models elsewhere. Meanwhile, upwellings along the northwestern Arabian Sea coast locally reduce surface temperatures below 20°C in summer,¹⁹ influencing water conditions as far as the Gulf of Aden.^20,21 Winter surface temperatures in the Arabian/Persian Gulf can drop by 10°C due to strong northwesterly winds (the shamals).^22,23 Whether these peculiar conditions modulate gene flow and shape local or regional coral adaptation remains poorly documented.

Figure 1.

Environmental classification of sampling localities and surface waters around the Arabian Peninsula

Analyses are based on 65 surface environmental variables downloaded from Bio-Oracle v2.2²⁴ (Table S3).

(A) Position of the Arabian Peninsula (AP; polygon) and the Maldives (MAL; turquoise dot).

(B) Dendrogram of the 17 sampling localities (codes as shown in A and C) based on pairwise Euclidean distances calculated from the first four environmental principal components (ePC1-4; representing 84.5% of the total variability) of the principal component analysis (PCA) on localities' environmental variables (see Figures S1 and S2). Localities are colored according to the best cluster partition (k = 8).

(C–E) Heatmaps of individual scores for the first three environmental principal components (ePC1-3, respectively) of the PCA on surface water environmental variables (see Figure S3). Dots indicate sampling localities and are colored following (B).

Previous studies have assessed the population genetic structure of sponges,²⁵ corals,^26,27,28 and fishes^29,30,31 in different parts of the AP, but none in its entirety, providing only a partial overview of regional patterns. Together with larval dispersal simulations,^32,33,34 they revealed genetic breaks aligning with environmental discontinuities and water mass boundaries. To further explore the role of abiotic conditions in shaping coral diversity, distribution, and connectivity around the AP, we assessed for the first time the population genetic structure of three scleractinian species – Diploastrea heliopora (Lamarck, 1816), Coscinaraea monile (Forskål, 1775), and Psammocora profundacella Gardiner, 1898 – from the Gulf of Aqaba to Kuwait. These zooxanthellate reef-building species exhibit different life history traits (including spawning, reported in May for D. heliopora in the Red Sea,³⁵ April-July for C. monile in the Arabian/Persian Gulf (P. Range, pers. comm.), and September-November for P. profundacella in the Red Sea³⁶) and distributions (from widespread throughout the AP for C. monile and P. profundacella to more restricted, likely temperature-limited, for D. heliopora),³⁷ making them complementary for assessing regional coral populations’ structure. We used target capture of ultraconserved elements (UCEs) and exons^38,39 to collect single-nucleotide polymorphisms (SNPs) and conduct population genomic analyses. We then evaluated correlations between F_ST and geographic distances, seasonal currents through larval dispersal modeling, and environmental conditions, revealing several barriers in the region. This study underscores the pivotal roles of geography and oceanographic features in structuring coral diversity around the AP and across reef systems in general, providing critical insights to guide stakeholders in implementing effective long-term conservation efforts.

Results

A total of 36, 109, and 130 colonies of D. heliopora, C. monile, and P. profundacella, respectively, were sampled across 16 localities of the Arabian Peninsula (AP), spanning from the Gulf of Aqaba in the northwest to Kuwait in the Arabian/Persian Gulf in the northeast (except D. heliopora, sampled only from the central Red Sea to Djibouti, due to its more restricted distribution). Additionally, 13, 5, and 13 colonies, respectively, were collected in the Maldives (Figures 1 and 2; Tables S1 and S2). Thus, the sampling encompassed multiple water bodies with distinct oceanographic conditions, so we first characterized this diversity. Using genome-wide single-nucleotide polymorphisms (SNPs) harvested from the target capture of ultraconserved elements (UCEs) and exons,^38,39 we then assessed, for each species, the population genetic structure and its correlation with distances, currents, and environment.

Figure 2.

Population genetic structure for each species

(A) Diploastrea heliopora, (B) Coscinaraea monile, and (C) Psammocora profundacella. For each population, pie charts (radius proportional to sample size) represent ancestral lineage proportions inferred by sNMF⁴⁰ at the optimal k (individual bar plot shown at the bottom left of each panel; locality codes as in Figure 1C; sample sizes in parentheses). Arrows indicate mean relative migrations above 0.5 (width proportional to relative migration) estimated with divMigrate⁴¹ over 100 bootstraps of five (10 for D. heliopora) randomly sampled individuals per population, to account for unequal population sizes (populations with less than five individuals are not included). Dashed lines in (A) show D. heliopora distribution.

Environmental classification of sampling localities and surface waters

Marine environment at the 17 sampling localities was characterized by a set of 65 variables downloaded from Bio-Oracle v2.2²⁴ (Table S3), representing all available present-day surface variables (except those related to ice) averaged over the 2000–2014 period with a spatial resolution of 5 arcmin (∼ 9 km for the study area). Collinearity among these variables was reduced by a principal component analysis (PCA; Figure S1). The first four environmental principal components (ePC1-4) accounted for 84.5% of the total environmental variability, with variables related to photosynthetically available radiation, dissolved oxygen, and silicate concentrations, and salinity being the main contributors across these components. A matrix of pairwise Euclidean distances among sampling localities was then calculated from these four ePCs (Figure S2), and a dendrogram was reconstructed to cluster localities by environmental similarity (Figure 1B). Geographically close localities often clustered together, with Kuwait being the most environmentally distinct. However, Djibouti and the Maldives clustered together at a relatively high hierarchical distance (5.8), before grouping with all other localities (except Kuwait, at 7.1), suggesting few environmental similarities between them while differing markedly from the remaining localities. Of the 23 indices tested to automatically determine the number of environmental clusters, seven suggested eight clusters, followed by five for five clusters (same partitions but lumping all Red Sea localities with Muscat). These eight clusters were consistent with sampling regions (Figures 1B, 1C, and S1C; Table S2): 1 – Gulf of Aqaba and Northern Red Sea (Sindala and Duba); 2 – Central Red Sea (Al-Wajh, Yanbu, and KAUST reefs); 3 – Southern Red Sea (Al-Lith and Farasan Islands); 4 – Gulf of Tadjoura (Djibouti); 5 – Gulf of Aden (Aden, Bal’haf, and Al-Mukalla) and Arabian Sea (Dalkhut and Mirbat); 6 – Gulf of Oman (Muscat); 7 – Arabian/Persian Gulf (Kuwait); 8 – Maldives.

The PCA with the same set of environmental variables for each 5 × 5 arcmin (∼ 9 × 9 km) surface water cell around the AP (Figure S3) supported the eight environmental clusters. The heatmaps based on individual (i.e., surface water cell) scores of ePC1-3, explaining 34.2%, 21.3%, and 10.3% of the surface water environmental variability, respectively, highlighted several environmental discontinuities (Figures 1C–1E). For instance, ePC1 emphasized environmental breaks between the Red Sea and the Gulf of Aden (Strait of Bab-el-Mandeb), the Arabian Sea and the Gulf of Oman, and the Gulf of Oman and the Arabian/Persian Gulf (Strait of Hormuz). The other two ePCs pointed out additional environmental discontinuities.

Clonal and genetic diversity

Targeted sequencing of UCEs and exons for the 306 colonies (plus five sequencing replicates) yielded a total of 1.9 × 10⁹ reads (2.9 × 10¹¹ bp), with between 1.4 × 10⁶ and 1.8 × 10⁷ reads per individual library (mean ± s.e. = 6.2 ± 0.1 × 10⁶ reads). Quality controls and adapter trims removed 6.9% of the bases. Between 780 and 1,593 locus reference sequences, with an overall mean length (± s.e.) of 997.6 ± 7.9 bp, were de novo reconstructed per species, on which 11.6%–61.1% of the individual trimmed reads were successfully mapped (mean ± s.e.: D. heliopora: 55.9 ± 1.1%, C. monile: 40.7 ± 0.4%, and P. profundacella: 34.4 ± 1.1%), with a mean coverage depth (± s.e.) of 84.7× (± 0.1). Final specific SNP datasets were as follow: D. heliopora: 49 colonies (plus two replicates) × 1,139 SNPs, C. monile: 114 colonies (plus one replicate) × 757 SNPs, and P. profundacella: 143 colonies (plus two replicates) × 646 SNPs, with a mean SNP coverage (± s.e.) of 44.4× (± 0.1), and from 0.8% to 46.3% missing data per colony (mean ± s.e. = 5.4 ± 0.4%). Most samples with higher missing data originated from BAL and had been collected ca. 20 years ago, yielding more fragmented and less concentrated DNA extracts and indicating potential preservation issues.

On average (± s.e.), sequencing replicates differed by 0.7 ± 0.1% of the alleles called, while distinct colonies differed from 9.2% to 30.9% (Figure S4). No colonies thus belonged to the same clonal lineage, and the overall clonal richness (R)⁴² was 1. Once replicates were removed, population sizes (N) ranged from 1 to 16, with 4 (over 4), 12 (over 15), and 13 (over 16) populations with N ≥ 5 for D. heliopora, C. monile, and P. profundacella, respectively. The proportion of missing data per population (%NA) ranged from 1.9% to 12.4%, except for the BAL population of P. profundacella (%NA = 36.0%; Table S4). Due to the uneven numbers of sampled populations per species and SNP filtering strategies, the number of variable SNPs per population (N_SNP) varied among species, ranging from 1,046 to 1,062 for D. heliopora (over 1,139 SNPs), from 185 to 656 for C. monile (over 757 SNPs), and from 155 to 603 for P. profundacella (over 646 SNPs). In contrast, allelic richness⁴³ (AR) rarefied to 10 alleles was relatively constant, varying from 1.59 to 1.89 alleles (over 2), except for one C. monile population (MAL, AR = 1.35 alleles). In the latter, the observed heterozygosity (Ho = 0.14) was lower than in other populations, which varied from 0.19 to 0.27 for C. monile and P. profundacella, and from 0.31 to 0.32 for D. heliopora. Expected heterozygosities (He) followed similar trends, resulting in inbreeding coefficients (F_IS) ranging from −0.04 to 0.06 (Table S4). Most F_IS did not significantly differ from zero after false discovery rate (FDR) correction⁴⁴ (p_adj ≥ 0.05), consistent with the absence of clonality and suggesting little to no substructure within populations.

Population genetic structure

Assignment tests (sNMF⁴⁰ and DAPC⁴⁵) produced congruent results up to three, five, and four genetic clusters (k) for D. heliopora, C. monile, and P. profundacella, respectively. Higher k values did not reveal additional genetic structure, and individuals from previously defined genetic clusters were instead admixed in similar proportions between two clusters (Figures S5–S7). For each species, the genetic partitioning at the retained k value aligned well with geography and environmental clusters. In D. heliopora, the three genetic clusters roughly grouped colonies from 1 – Red Sea, 2 – Gulf of Tadjoura, and 3 – Maldives (Figure 2A). A similar partitioning was obtained for C. monile, in which additional clusters were found due to its wider distribution. The five clusters corresponded to colonies from 1 – Red Sea, 2 – Gulfs of Tadjoura and Aden, 3 – Arabian Sea and Gulf of Oman, 4 – Arabian/Persian Gulf, and 5 – Maldives (Figure 2B). In contrast, in the widespread P. profundacella, colonies from the Red Sea and the Gulf of Tadjoura and Aden were grouped, as well as colonies from the Gulf of Oman and the Arabian/Persian Gulf. The Arabian Sea and Maldives colonies remained in two distinct clusters (Figure 2C).

Genetic differentiation indices (F_ST⁴⁶) estimated across all conspecific populations varied from 0.0037 to 0.0789 for D. heliopora, from 0.0000 to 0.4195 for C. monile, and from 0.0000 to 0.1850 for P. profundacella, with values on average 10–15 times higher and more significant between populations from distinct genetic clusters (Table S5). To account for unequal population sizes (N; D. heliopora: 10–13, C. monile: 1–15, and P. profundacella: 1–16), which can bias F_ST estimations, mean F_ST over 100 bootstraps of five (10 for D. heliopora) randomly sampled individuals per population were also computed and compared to raw estimates (populations with N < 5 were not considered). Average (± s.e.) absolute deviation between bootstrapped and raw F_ST was 0.0051 ± 0.0005 (range: 0.0000–0.0357), with significant differences (one-sample t-tests; p_adj < 0.05) for 24 over 150 population pairs. Most of these involved MAL, and when this locality was excluded, the average (± s.e.) absolute deviation decreased to 0.0016 ± 0.0002 (range: 0.0000–0.0081), with only 12 population pairs showing significant differences (Table S5), suggesting reliable F_ST estimations.

Relative migrations among populations were further estimated using divMigrate,⁴¹ following the same bootstrap procedure as for F_ST (thus excluding populations with N < 5). Although the algorithm is experimental,⁴¹ it can provide broad estimates of gene flow direction and strength. Average relative migrations were more important between populations within genetic clusters, with maximum migrations reported in the Red Sea (D. heliopora: from LIT to KAU, C. monile: from AQA to WAJ, and P. profundacella: from WAJ to DUB). In particular, networks became fragmented when only relative migrations above 0.5 were kept, with few connections between populations from different clusters remaining (Figure 2).

Seasonal larval dispersal modeling

To assess the potential impact of seasonality and currents on connectivity, larval dispersal around the AP was simulated from surface currents of a general, high-resolution (∼ 2 km), circulation model (MITgcm⁴⁷) for two seasons (summer and winter, with daily spawning events in May-June and November-December, respectively) over the period 2000–2018. Given the limited data available on the reproduction of the focal species, dispersal characteristics were set in line with previous studies modeling the larval dispersal of broadcast spawning corals^48,49 (see details in Methods). Our simulations highlighted different larval trajectories between seasons (Figure S8), resulting in different connectivity matrices (Figure S9). For instance, in summer, some larvae released in the southern Red Sea were able to cross the Strait of Bab-el-Mandeb. In winter, however, no larvae from the Red Sea were exported outside, but larvae released as far as Mirbat (Arabian Sea) were able to enter. In the Gulf of Aden, Arabian Sea, and Gulf of Oman, trajectories were predominantly eastward in summer, while westward in winter. Finally, for both seasons, very few larvae from Muscat (Gulf of Oman), Kuwait (Arabian/Persian Gulf), and the Maldives were able to reach other sampling localities (Figure S8).

Genetic structure predictors

Potential impacts of geography (isolation by distance, IBD), oceanography (isolation by resistance, IBR), and environment (isolation by environment, IBE) on the population genetic structure of C. monile and P. profundacella (D. heliopora was not considered due to the small number of populations sampled, which would have led to imprecise correlations) were first evaluated using Mantel correlation tests,⁵⁰ considering or not the Maldives population. Previously estimated bootstrapped F_ST were used as response variables (populations with less than five samples were thus not considered). Over-water geographic distances and inverse log-scaled maximum probabilities of connectivity obtained from spawning simulations (in summer for C. monile, in winter for P. profundacella; Figure 3A) were used to test for IBD and IBR, respectively. For IBE, Euclidean distances were computed for the ePC1-4 (together and separately) previously used for the hierarchical clustering of sampling localities (Figures 1B and S1), and for the 65 environmental variables separately (Table S3). Correlations were stronger (R² on average 1.7 times higher) and more significant when the Maldives population was not considered (Figure S10). We therefore discuss only the results excluding this locality. Of the 144 tests (72 variables × 2 species), 28 were significant after FDR correction (p_adj < 0.05; C. monile: 15, P. profundacella: 13), with R² varying from 0.20 to 0.85 (Figure 3B). Strong correlations were found between F_ST and over-water geographic distances (C. monile: y = 4.6 × 10⁻⁵x - 5.2 × 10⁻⁴, N = 55, R² = 0.84, p = 2.2 × 10⁻¹⁶; P. profundacella: y = 2.2 × 10⁻⁵x - 3.6 × 10⁻³, N = 66, R² = 0.60, p = 1.6 × 10⁻¹⁴; Figure 3C), and, to a lesser extent, between F_ST and inverse probabilities of connectivity (C. monile: y = 1.9 × 10⁻¹x - 6.1 × 10⁻², N = 55, R² = 0.32, p = 4.8 × 10⁻⁶; P. profundacella: y = 1.0 × 10⁻¹x - 3.0 × 10⁻², N = 66, R² = 0.30, p = 9.5 × 10⁻⁷; Figure 3D). Some environmental parameters, such as the diffuse attenuation coefficient at 490 nm (a proxy of water clarity and irradiance level) or the dissolved iron and phosphate concentrations, also showed relatively strong and significant correlations with F_ST for both species (Figure 3B). Of course, most of these variables also correlated with geographic distances (Figures S11 and S12), so redundancy analyses (RDAs) were performed to disentangle the effects of IBD, IBR, and IBE on both species’ genetic structures.

Figure 3.

Population genetic differentiation and correlation tests

(A) Genetic differentiation (lower triangle) and connectivity (upper triangle) among populations for Coscinaraea monile and Psammocora profundacella, represented as heatmaps of max-normalized pairwise F_ST⁴⁶ (i.e., F_ST divided by the maximum F_ST value for each species, to allow inter-species comparison; reds indicate stronger differentiations) and of max-normalized inverse log-scaled maximum probabilities of dispersal obtained from the summer (C. monile) or winter (P. profundacella) spawning simulations (each value represented corresponds to the maximum of the two unidirectional flows inferred between two localities; blues indicate stronger connectivity; gray denotes no larval exchanges), respectively. Locality codes as in Figure 1C, colored according to environmental clusters (see Figure 1B). Crossed cells indicate no F_ST (unsampled populations for the given species). Asterisks denote F_ST significance after false discovery rate (FDR)⁴⁴ correction ∗: 0.01 ≤ p_adj < 0.05, ∗∗: 0.001 ≤ p_adj < 0.01, and ∗∗∗: p_adj < 0.001.

(B) Correlations between bootstrapped F_ST and over-water geographic (overwater.geo), oceanographic (oceano), and environmental distances among populations (without the Maldives) for both species, represented as a heatmap of the determination coefficients (R²; reds indicate stronger correlations). Environmental distances were computed as the Euclidean distances among localities for all and each of the environmental principal components used for the hierarchical clustering of the sampling localities (ePC1-4; see Figures 1B and S1) and for each of the 65 environmental variables previously considered (codes as in Table S3). The number of pairwise comparisons used to estimate the correlations (N) is indicated in parentheses above. Asterisks denote significance of Mantel tests⁵⁰ after FDR correction, as in (A). Detailed correlations between bootstrapped F_ST and over-water geographic (C) or oceanographic (D) distances are shown, with (all points; gray linear regressions and R²) or without (only black points; black linear regressions and R²) the Maldives.

RDAs were first performed separately for IBD (using distance-based Moran’s eigenvector maps (dbMEMs)⁵¹ to decompose over-water geographic distances; Figures S13A and S13B), IBR (using asymmetric eigenvector maps (AEMs)⁵² derived from directed acyclic graphs with edges weighted using spawning simulations; Figures S13C–S13F), and IBE (using either ePCs to reduce collinearity among the 65 environmental variables previously considered or one representative variable per collinear group (|r | ≥ 0.75; Figures S11 and S12) to ease interpretations). Each model underwent forward selection to identify significant predictors to include in subsequent models combining all explanatory variable types (i.e., dbMEMs, AEMs, and ePCs or representative environmental variables). As response variables, the first seven and eight PCs from PCA on Hellinger-transformed allele frequencies⁵³ were retained for C. monile and P. profundacella, respectively, cumulatively explaining 87.4% and 86.4% of the total genetic variation, respectively. Forward selection of dbMEMs identified four (dbMEM1-2-3-4) and two (dbMEM1-2) significant predictors, respectively, all representing large-scale spatial structures. RDAs with only those selected variables were significant (p < 0.001) for both species, with adjusted R² of 0.532 (C. monile) and 0.246 (P. profundacella; Table 1). For AEMs, only one significant predictor (AEM6) was retained for P. profundacella, and the resulting RDA was significant (R_adj² = 0.078; p = 0.014). Finally, for IBE, when considering ePCs derived from all 65 environmental variables, three (ePC1-2-3, accounting for 42.5%, 24.8%, and 17.3% of the variability, respectively) and two (ePC1-6; 44.0% and 2.2% of the variability) significant predictors were retained for C. monile and P. profundacella, respectively, and the resulting RDAs were both significant (C. monile: R_adj² = 0.372, p < 0.001; P. profundacella: R_adj² = 0.140, p = 0.012). Instead, grouping highly correlated (|r | ≥ 0.75) environmental variables led to 24 and 23 collinear groups for C. monile and P. profundacella, respectively (Figures S11 and S12). Keeping one representative variable per group, forward selection identified the minimum diffuse attenuation coefficient at 490 nm (“da.min”) and mean phosphate concentration (“phosphate.mean”), both singletons in their collinear group (Figure S11), as significant predictors for C. monile. The same significant predictors were identified for P. profundacella, plus the minimum silicate concentration (“silicate.min”), representative of 19 other variables (= group 20 in Figure S12). RDAs with these selected variables were also significant (p < 0.001) and explained a similar (C. monile, R_adj² = 0.340) or higher (P. profundacella, R_adj² = 0.310) proportion of variability than ePCs. In the final models combining selected predictors, whether considering ePCs or representative environmental variables did not affect the results. Indeed, for both species, environmental predictors were excluded as not explaining additional genetic variation (possibly due to collinearity with dbMEMs, but the latter also had unique contributions). Thus, only results including ePCs are presented. Global RDAs (C. monile: dbMEM1-2-3-4 + ePC1-2-3, R_adj² = 0.595; P. profundacella: dbMEM1-2 + AEM6 + ePC1-6, R_adj² = 0.346) were significant (p < 0.001) for both species, but none of the variables significantly contributed to the model (p ≥ 0.05). When partitioning out the variation explained by the different sets of predictors, ePCs alone were explaining 6% and 0% of the variability for C. monile and P. profundacella, respectively. Thus, for C. monile, only dbMEM1-2-3-4 were kept in the final model (R_adj² = 0.532, p < 0.001, all variables significant). The first two large-scale spatial structures (dbMEM1-2) were associated with the differentiation of western (Red Sea and Gulf of Tadjoura) vs. eastern (Arabian Sea, Gulf of Oman, and Arabian/Persian Gulf) populations on the first RDA axis (explaining 37.5% of the variability), while the following two (dbMEM3-4) mostly separated Red Sea vs. Gulf of Tadjoura and the four eastern populations among them on the second axis (14.5%; Figures 4A and 4C). In contrast, for P. profundacella, dbMEM1-2 and AEM6 were complementary in explaining the genetic variability (dbMEMs: 27.5%, AEMs: 10.8%, both: 0.0%), and the resulting RDA, as well as all included predictors, were significant (R_adj² = 0.354, p < 0.001). As for C. monile, dbMEM1-2 separated western (Red Sea and Gulf of Tadjoura) vs. eastern (Arabian Sea, Gulf of Oman, and Arabian/Persian Gulf) populations on the first RDA axis (27.7%), but the population from the Gulf of Aden (BAL) was rather separated with AEM6 on the second axis (14.8%; Figure 4B). This AEM seemed to reflect a break in the physical connectivity between the Gulf of Aden and Arabian Sea (Figure 4D).

Table 1.

Redundancy analyses (RDAs)

Significant variables		RDA1	RDA2	RDA3	Radj2	p
Coscinaraea monile
►dbMEM	1-2-3-4	37.5%	14.5%	13.7%	0.532	<0.001	∗∗∗
AEM	N/A	–	–	–	–	–
ePC	1-2-3	32.8%	14.3%	9.0%	0.372	<0.001	∗∗∗
EnvGrp	11–21	34.7%	12.5%	–	0.340	<0.001	∗∗∗
dbMEM+ePC		37.7%	14.7%	13.8%	0.595	<0.001	∗∗∗
dbMEM+EnvGrp		37.7%	14.6%	13.8%	0.572	<0.001	∗∗∗
Psammocora profundacella
dbMEM	1–2	25.7%	12.6%	–	0.246	<0.001	∗∗∗
AEM	6	16.2%	–	–	0.078	0.014	∗
ePC	1–6	14.8%	14.8%	–	0.140	0.012	∗
EnvGrp	14-21-20	25.2%	12.7%	11.9%	0.310	<0.001	∗∗∗
dbMEM+AEM+ePC		27.8%	15.2%	14.2%	0.346	<0.001	∗∗∗
dbMEM+AEM+EnvGrp		28.1%	16.3%	13.3%	0.319	<0.001	∗∗∗
►dbMEM+AEM		27.7%	14.8%	10.5%	0.354	<0.001	∗∗∗

Comparison of RDA results with over-water distance-based Moran’s eigenvector maps (dbMEMs),⁵¹ asymmetric eigenvector maps (AEMs)⁵² derived from the spawning simulations, and environmental principal components (ePCs) or representative variables of the collinear environmental groups (EnvGrps; see Figures S11 and S12) for Coscinaraea monile and Psammocora profundacella (without the Maldives). Final retained models (Figure 4) are indicated by arrowheads. Only variables included in the best model by forward selection are shown. Proportions of variation explained by the first three RDA axes are indicated (significant axes in bold; p < 0.05). R_adj²: adjusted coefficient of determination, p: RDA significance (∗: 0.01 ≤ p < 0.05, ∗∗: 0.001 ≤ p < 0.01, and ∗∗∗: p < 0.001).

Figure 4.

Redundancy analyses (RDAs)

Biplot of the first two RDA axes for the final models of (A) Coscinaraea monile and (B) Psammocora profundacella, with the spatial visualization of the selected significant variables (i.e., over-water distance-based Moran’s eigenvector maps – dbMEMs⁵¹ – and asymmetric eigenvector maps – AEMs⁵² – derived from the spawning simulations; see Table 1) in (C) and (D), respectively. Dots indicate sampling localities (locality codes as in Figure 1C) and are colored according to environmental clusters (A-B; see Figure 1B) or to scaled variable scores (C and D).

Discussion

In this first genomic study of the population structure of three reef-building corals around the Arabian Peninsula (AP), contrasting patterns among species were found, with different barriers overlying a general isolation by distance (IBD). While ocean currents (isolation by resistance; IBR) appear to shape some barriers (particularly in P. profundacella) and, in fine, part of the genetic structure of corals in this region, environmental gradients (isolation by environment; IBE) covarying with spatial separation could also intervene. Differences in the location of genetic breaks between D. heliopora and C. monile (summer-spawners) vs. P. profundacella (winter-spawner) were identified, potentially coinciding with reproduction periods combined with differences in seasonal currents and/or environmental conditions (Figure 5). This complex combination of covarying variables and isolation processes demonstrates the importance of multi-species and multi-disciplinary approaches to assess population genetic structure, and provides insights into the differentiation of AP and global reef communities.

Figure 5.

Contrasting patterns of genetic structure

Graphical summary of the patterns of genetic structure observed for Coscinaraea monile (and, to a lesser extent, for Diploastrea heliopora) (A) and Psammocora profundacella (B), with the indication of barriers previously reported around the Arabian Peninsula. Seasonal currents, gyres, and coastal upwellings at the species’ spawning periods (A – summer; B – winter) are indicated schematically (adapted from^{19,54,55,56,57,58,59,60,61}). Dots represent sampling localities and are colored according to environmental clusters (see Figure 1B).

IBD has frequently been reported across organisms around the AP.^16,30,62 This region is also renowned for considerable environmental variability.^56,63,64 In the Red Sea, salinity, temperature, productivity, or dissolved oxygen concentration show marked north-south gradients^65,66 correlating with distance. Significant correlations were thus found between genetic and over-water geographic distances, but also between genetic distances and some environmental parameters, to the point that significant predictors of geographic distance (dbMEMs) were redundant with environmental ones (ePCs) and explained more of the genetic variability in C. monile. Given that both sets of variables were sometimes highly correlated, it is challenging to disentangle the relative contributions of geography and environment in influencing genetic structure. Environmental gradients (e.g., temperature, salinity, and thus, density) are also major drivers of winds and currents, adding complexity to separating their influence on isolation processes. Studying these processes individually and combined is therefore crucial to better understand their effect on connectivity.

Strong genetic differentiation was found between the AP and the Maldives, consistent with previous findings on reef fishes.⁶⁷ The AP is isolated from the rest of the Indian Ocean by sandy and coral reef-poor coastlines (e.g., Somalia and India) and large oceanic expanses, which constitute a major dispersal barrier for reef-dwelling organisms. Regional and local surface circulation, highly influenced by the seasonal monsoon system,^56,68 further isolates it. This genetic gap is not only visible at the population level, but also at the species and community levels, as evidenced by the high rate of endemism of the AP reef fauna.^37,69,70 AP reef communities are thus isolated and unique, making them more vulnerable to threats. Understanding regional patterns of genetic structure is therefore critical for addressing these threats.

Differences in population genetic structure among species were evident at three boundaries: between the Red Sea and Gulf of Aden (break in C. monile and D. heliopora), between the Arabian Sea and Gulf of Oman (break in P. profundacella), and between the Gulf of Oman and Arabian/Persian Gulf (break in C. monile). Although multiple factors may contribute to these differences, they coincide with species’ reproductive periods (summer for C. monile (P. Range, pers. comm.) and D. heliopora³⁵; winter for P. profundacella³⁶), which, combined with the high seasonality of currents and environmental conditions, may have contributed to the observed patterns in population structure.

The Red Sea is a semi-enclosed basin connected to the Gulf of Aden through the narrow (∼ 26 km) and relatively shallow (∼ 300 m depth) Strait of Bab-el-Mandeb. Circulation in the Strait is highly seasonal, with strong surface inflow from the Gulf of Aden to the Red Sea in winter (October to February), progressively altering to a weak outflow in summer (June to September).^63,71 Thus, exchanges of planktonic larvae between the Red Sea and the Gulf of Aden are expected to be stronger in winter. Our and previous^33,34 larval dispersal simulations support such seasonal differences in larval transport. The absence of genetic differentiation among Red Sea, Gulf of Tadjoura, and Gulf of Aden populations in the winter-spawning P. profundacella, compared with the isolation of Red Sea populations of the summer-spawning D. heliopora and C. monile, also matches this expectation (Figure 5). No previous study has assessed the genetic structure of corals across the Bab-el-Mandeb Strait. In reef fishes, similar variation is evident, with some species showing no genetic differences across the Strait,^16,30,31,67 while others showed a marked discontinuity.^16,67 However, reproductive cycles remain unknown for most species investigated, preventing testing whether differences in connectivity are related to seasonal reproduction. Data on spawning, larval behavior, duration, and ultimately dispersal are needed to better understand the broader role of seasonality on connectivity. The Strait of Bab-el-Mandeb is associated with a change in faunal composition,^69,70,72 also supporting the importance of ocean currents and environment in influencing species' ability to disperse and colonize in or out of the Red Sea.

Previous studies on corals,^26,27 as well as sponges²⁵ and fishes,^16,29,30,31 highlighted a genetic break around 16–20°N in the Red Sea in some species, coinciding with an environmental and faunal change.^73,74 Unfortunately, due to logistical constraints, the two species that show differentiation between the Red Sea and Gulf of Aden (D. heliopora and C. monile) were only sampled until Al-Lith, at the border between the central and southern Red Sea. It is thus unclear whether the genetic structure observed between populations inside and outside of the Red Sea in these species is related to the Strait of Bab-el-Mandeb (as discussed above), to the 16–20°N break reported in previous studies, or to both. Further sampling efforts in the southern Red Sea would help to understand how populations are structured in this area. Nevertheless, all investigated reef-dwelling species (in this study and in previous ones^{16,25,26,27,30}) show high genetic homogeneity between the central and northern Red Sea (including the Gulf of Aqaba), suggesting a single evolutionary and management unit for species distributed all along this part of the basin.

Like the Strait of Bab-el-Mandeb, the Strait of Hormuz (∼ 56 km wide), connecting the Gulf of Oman and Arabian/Persian Gulf waters, seems to act as a seasonal barrier. High genetic differences between both sides of the Strait were inferred for C. monile, contrasting with the homogeneity in P. profundacella. Previous studies found similar variability. For instance, the Arabian regional endemic coral Acropora downingi²⁸ (spawning in April-May^75,76) and the angelfish Pomacanthus maculosus⁷⁷ showed restricted gene flow across the Strait (as in C. monile), while no differentiation was found in other reef fishes^78,79 (as in P. profundacella). Surface circulation in these water bodies is highly complex, showing seasonal and interannual variability^60,61,80 (Figure 5). The southern part of the Strait generally alternates between a net inflow during winter/spring to a net outflow during summer/fall,⁶⁴ although the latter contrasts with larval dispersal models that show year-round inflow, with larvae released in the Gulf trapped within.³⁴

The boundary between the Gulf of Oman and the Arabian Sea at Ras al Hadd has been considered one of the sharpest biotic transitions known in marine biogeography,⁸¹ clearly evident in an environmental transition (Figure 1C) as well as in circulation modeling in summer, but not winter (Figure S8). The summer spawner C. monile shows a break here, while the winter spawner P. profundacella does not. In the Gulf of Oman, both empirical measurements¹⁹ and dispersal models³⁴ show that the outflow alternates between running along the Arabian coast in winter and toward the Indian coast in summer, which could lead to the different genetic structuring patterns observed. This alternation of westerly to easterly currents is driven by the intertropical convergence zone migration,²⁰ which triggers the seasonal upwelling off the Arabian Sea coast. The system reaches its maximum intensity in summer, bringing cold and nutrient-rich deep waters to the surface and inducing easterly currents, parallel to the shore.^19,20,82

Although stronger correlations were estimated between F_ST and geographic rather than oceanographic distances, currents generally explained the broad patterns well. Mismatches between model outputs and genetics or empirical current measurements were also evident (e.g., at the Strait of Hormuz). Accurate modeling of meso-to small-scale currents in large-scale simulations is challenging, as emphasized in the literature.^83,84 Lack of fundamental information on species’ reproductive behavior regionally is also limiting, and the use of reproductive parameters derived from other broadcast spawning species may partly explain mismatches with genetic patterns. For a fine-scale understanding of connectivity, high-resolution, low-extent circulation models, coupled with extensive sampling efforts to investigate local gene flow in a model (i.e., heavily studied, including for its reproduction) species, are necessary. This is particularly important for marine spatial planning and designing networks of marine protected areas, but is more resource- and time-consuming, which could be critical for coral reefs. Compromises among scale, resolution, sampling effort, time, and so forth are thus needed and significantly influence our understanding of connectivity.

In conclusion, our results highlight the isolation of the AP, and within it, several genetic breaks influenced by seasonal currents (IBR) and possibly environmental conditions (IBE), on top of a general isolation by distance (IBD). From a conservation perspective, the different and species-specific genetic clusters should represent distinct management units,⁸⁵ and will ultimately help defining effective management areas. Red Sea and Arabian/Persian Gulf populations also appears particularly isolated, which supports the uniqueness of their communities, but also their sensitivity to climate change and local threats. More generally, this study demonstrates the pivotal roles of both geography and oceanographic features in structuring coral diversity across reef ecosystems, providing critical insights to guide stakeholders in implementing long-term effective conservation efforts. It also represents a methodological framework, confirming the utility of UCEs and exons capture to resolve population genetic structure in corals, and provides guidelines for future research on this topic.

Limitations of the study

Here, genetic structure has been inferred as a proxy of gene flow and connectivity. While this association has been widely used in the literature, it is important to note that the inference of gene flow from allele frequency data relies on several simplifying assumptions, including the expectation that populations are at or near drift-migration equilibrium. While high levels of gene flow can rapidly homogenize allelic frequencies, the converse is not true, and genetic differences can take several generations to reach equilibrium. As a result, observed patterns of genetic structure may reflect historical averages of gene flow rather than contemporary dynamics. Additionally, unequal and small population sizes may affect allele frequency estimations; however, the use of genomic data and our bootstrap approach helped provide more reliable results. Population structure was inferred without explicitly separating neutral from adaptive variation, aiming at capturing the overall genetic differentiation. Thus, results do not distinguish between patterns driven by selection vs. those arising from demographic processes. The accuracy of larval dispersal models depends on several parameters, including biological traits, which were largely unavailable for the understudied focal species and were therefore inferred from related species with similar known characteristics. Future research would benefit from improved knowledge of the biology of these species. Similarly, while the 5-arcmin (∼ 9 km for the study area) resolution of environmental variables captures broad regional environmental gradients, it does not represent microhabitat-level conditions. Integrating higher-resolution, site-specific environmental data (e.g., in situ measurements or remote sensing at finer scales) may reveal additional patterns not captured in our regional analysis. Finally, the distribution of sampling localities around the Arabian Peninsula implied almost linear networks along the coastline, making it difficult to disentangle the effects of spatially correlated predictors, especially in the Red Sea, where latitudinal environmental gradients are observed. Additional sampling along the western Red Sea coast and in the Arabian/Persian Gulf, overcoming local logistic challenges, is needed to better assess the influence of the various contributing factors.

Resource availability

Lead contact

Requests for further information and resources should be directed to the lead contact, Nicolas Oury (nicolasoury@hotmail.fr).

Materials availability

Raw sequencing reads generated in this study were deposited on the NCBI (BioProject PRJNA1074537).

Data and code availability

• •Raw sequencing reads have been deposited on the NCBI (BioProject PRJNA1074537).
• •All original code has been deposited on Zenodo at https://doi.org/10.5281/zenodo.17540001.
• •All other data underlying this article (including colonies in situ pictures, datasets, and intermediate files) have been deposited on Zenodo at https://doi.org/10.5281/zenodo.17540001.
• •Any additional information required to reanalyze the data reported in this article is available from the lead contact upon request.

Acknowledgments

This study was supported by NEOM OSSARI grant RGC/3/6027-01-01 and KAUST baseline research funds BAS/1/1090-01-01 awarded to FBe. Research was undertaken in accordance with KAUST research policies and procedures, and all permissions from the applicable governmental agencies in Saudi Arabia have been obtained. Sampling in the Gulf of Aqaba and northern Red Sea was performed during the Red Sea Deep Blue expedition onboard the OceanXplorer. We thank NEOM for facilitating and coordinating the expedition and, specifically, A. Alghamdi, T. Habis, A. A. Eweida, P. Mackelworth, P. Marshall, J. Myner, and G. Palavicini. Thanks to OceanX and the crew of the OceanXplorer for their operational and logistical support. Special thanks to M. Rodrigue for the scientific coordination of the expedition. In the central and southern Red Sea, sampling was sponsored by KAUST baseline funds to MLB and FBe, and CRG-1-BER-002 was awarded to MLB. We wish to thank KAUST Coastal and Marine Resources Core Laboratory for fieldwork logistics, as well as the captains and crews of the MV Dream Master, MV Dream Island, and MV Tempest. We would also like to extend our thanks to the Marine Enhancement and Environmental Operations teams from the Red Sea Zone Environmental Protection and Regeneration department at Red Sea Global (RSG) for their support in the logistics and operations in Al-Wajh. For Djibouti, the last author is grateful to the 2020 Dolphin cruise leader D. Barshis (Old Dominion University, VA). We also thank the Ministry of the Environment and Sustainable Development (MEDD) and Ministère de l’Urbanisme, de l’Environnement et du Tourisme (MUET) for their critical assistance in supporting the research in Djibouti, particularly B. H. Ismail and Y. M. Omar, through coordination of the research, permits, and sample transport. We also thank C. Canellakis and O. Awaleh from the US Embassy for their logistical support. We are also grateful to E. Karsenti (EMBL) and E. Bourgois (Tara Expeditions), the OCEANS Consortium for allowing sampling during the Tara Oceans expedition in Djibouti. We thank the commitment of the following people and additional sponsors who made this singular expedition possible: CNRS, EMBL, Genoscope/CEA, VIB, Stazione Zoologica Anton Dohrn, UNIMIB, ANR (projects POSEIDON/ANR09-BLAN-0348, BIOMARKS/ANR-08-BDVA-003, PROMETHEUS/ANR-09-GENM-031, and TARA-GIRUS/ANR-09-PCSGENM-218), EU FP7 (MicroB3/No.287589), FWO, BIO5, Biosphere 2, the Veolia Environment Foundation, Région Bretagne, World Courier, Illumina, Cap L’Orient, the EDF Foundation EDF Diversiterre, FRB, the Prince Albert II de Monaco Foundation, E. Bourgois, the Tara schooner, and its captain and crew. Tara Oceans would not exist without continuous support from 23 institutes (https://oceans.taraexpeditions.org). This article is contribution number 165 of the Tara Oceans expedition 2009–2012. In Yemen, fieldwork organization, logistics, and sampling permits from the relevant authorities were possible thanks to the collaboration of M. Pichon (MTQ), E. Dutrieux (Creocean), C.H. Chaineau (Total SA), R. Hirst, and M. Abdul Aziz (YLNG). Omani fieldwork was supported by NSF DEB-1457817 to GP, and we thank M. & E. Claereboudt, N. Stauft, K. Samimi-Namin, S. Wilson, and other members of the Oman Bioblitz expeditions for facilitating fieldwork, and the Environment Authority of Oman for permits. We are thankful to M. Pichon, S. Sartoretto, C. Marschal, and S. Alhazeem (KSIR) for facilitating the sampling in Kuwait. Sample collection in the Maldives was performed under research permit (OTHR)30-D/INDIV/2022/42, obtained with the help of M. Haleem (Ministry of Fisheries, Marine Resources, and Agriculture). We also thank A. Shazly (MNU) and H. Ahmed (Save the Beach Maldives) for sponsorship and logistic support, D. Telli, and all the crew of the MY Ocean Sapphire. Special thanks to J. Bouwmeester and P. Range for their work on coral reproduction. We acknowledge the Plateforme iGenSeq of the Institut du Cerveau et de la Moelle épinière (ICM, Paris, France) for library preparations and sequencing. Bioinformatics analyses and larval dispersal models were run on the KAUST Ibex cluster and KAUST Supercomputing Laboratory (Shaheen III), respectively.

Author contributions

Conceptualization: N.O., F.Be., T.I.T., Y.W., I.H., and F.M.; methodology, formal analysis, and investigation: N.O., Y.W., J.M., F.Be., and F.M.; resources: all authors; software: N.O., Y.W., and J.M.; data curation: N.O., T.I.T., F.Ba., F.Be., Y.W., J.M., F.M., and G.P.; funding acquisition: F.Be. and I.H.; supervision: F.Be. and I.H.; writing – original draft preparation: N.O.; writing – review and editing: N.O., F.Be., G.P., R.P., Y.W., I.H., and T.I.T.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological Samples
Diploastrea heliopora	see Table S1	see Table S1
Coscinaraea monile	see Table S1	see Table S1
Psammocora profundacella	see Table S1	see Table S1
Chemicals, peptides, and recombinant proteins
Ethanol (absolute)	VWR International, Radnor, PA	Cat #20821.330
Sodium hypochlorite (domestic grade)	N/A	N/A
Critical commercial assays
DNeasy® Blood & Tissue Kit	QIAGEN GmbH, Hilden, Germany	Cat #69506
NEBNext® UltraTM II FS DNA Library Prep kit for Illumina	New England Biolabs, Ipswich, MA	Cat #E7805L
Illumina NovaSeq X Series 1.5B Reagent Kit (300 cycles)	Illumina, San Diego, CA	Cat #20104705
Deposited data
Environmental layers	BioOracle v2.2	https://www.bio-oracle.org/downloads-to-email-v2.php?version=2_2
UCE raw sequencing data	NCBI – This study	PRJNA1074537
Other data from this study	Zenodo – This study	https://doi.org/10.5281/zenodo.15340482
Oligonucleotides
hexa-v2-scleractinia bait set	Cowman et al.38	myBaits design #D10350PCCRL
Software and algorithms
R v4.0.4	R Core Team86	https://www.r-project.org
sdmpredictors (R package)	Bosch and Fernandez87	https://CRAN.R-project.org/package=sdmpredictors
raster (R package)	Hijmans88	https://CRAN.R-project.org/package=raster
NbClust (R package)	Charrad et al.89	https://CRAN.R-project.org/package=NbClust
poppr (R package)	Kamvar et al.90	https://CRAN.R-project.org/package=poppr
hierfstat (R package)	Goudet91	https://CRAN.R-project.org/package=hierfstat
PopGenReport (R package)	Adamack and Gruber92	https://CRAN.R-project.org/package=PopGenReport
poppr (R package)	Kamvar et al.90	https://CRAN.R-project.org/package=poppr
LEA (R package)	Frichot and François93	https://doi.org/10.18129/B9.bioc.LEA
adegenet (R package)	Jombart94	https://CRAN.R-project.org/package=adegenet
StAMPP (R package)	Pembleton et al.95	https://CRAN.R-project.org/package=StAMPP
diveRsity (R package)	Keenan et al.96	https://CRAN.R-project.org/package=diveRsity
marmap (R package)	Pante and Simon-Bouhet97	https://CRAN.R-project.org/package=marmap
ade4 (R package)	Dray and Dufour98	https://CRAN.R-project.org/package=ade4
adespatial (R package)	Dray et al.99	https://CRAN.R-project.org/package=adespatial
igraph (R package)	Csardi and Nepusz100	https://CRAN.R-project.org/package=igraph
vegan (R package)	Oksanen et al.101	https://CRAN.R-project.org/package=vegan
FastQC v0.11.7	Babraham Bioinformatics	http://www.bioinformatics.babraham.ac.uk/projects/fastqc
MultiQC v1.7	Ewels et al.102	https://multiqc.info
cutadapt v2.1	Martin103	https://github.com/marcelm/cutadapt
Trim Galore! v0.6.0	Babraham Bioinformatics	http://www.bioinformatics.babraham.ac.uk/projects/trim_galore
SPAdes v3.13.0	Bankevich et al.104	https://github.com/ablab/spades
Phyluce	Faircloth105	https://github.com/faircloth-lab/phyluce
CAP3	Huang and Madan106	http://doua.prabi.fr/software/cap3
BWA v0.7.17	Li and Durbin107	https://bio-bwa.sourceforge.net
Picard v2.20.7	Broad Institute	https://broadinstitute.github.io/picard
GATK v3.8.1	McKenna et al.108	https://gatk.broadinstitute.org
BCFtools v1.9	HTSlib	http://samtools.github.io/bcftools
Clumpak	Kopelman et al.109	https://clumpak.tau.ac.il
MITgcm	Marshall et al.47	https://mitgcm.org
Other
Code and scripts	Zenodo – This study	https://doi.org/10.5281/zenodo.15340482

Experimental model and study participant details

Colonies of Diploastrea heliopora (N = 49), Coscinaraea monile (N = 114), and Psammocora profundacella (N = 143) were collected across 17 localities around the Arabian Peninsula and the Maldives (Table S1). All applicable permissions from the relevant authorities were obtained for collection and research (see Acknowledgments section).

Method details

Sampling

Colonies of the three target coral species – Diploastrea heliopora (Lamarck, 1816), Coscinaraea monile (Forskål, 1775), and Psammocora profundacella Gardiner, 1898 – were sampled while scuba diving or snorkelling during various cruises and expeditions which took place between September 2004 and December 2022. A total of 306 colonies (D. heliopora: 49, C. monile: 114, and P. profundacella: 143) were collected in 17 localities, from the Gulf of Aqaba to Kuwait around the Arabian Peninsula (AP), and the Maldives (Figures 1 and 2; Tables S1 and S2). A minimum distance of 5 m between sampled colonies was maintained to minimize the risk of sampling clonemates. Each colony was photographed in situ, its depth noted, and, depending on its size, fragmented or entirely collected using chisel and hammer (see Table S1 and ‘Data and code availability’ section for individual metadata and in situ pictures, respectively). Subsamples of ∼ 1 cm² were immediately taken after diving and preserved in absolute ethanol at 4°C for further genetic analyses. The remaining part of the corallum was bleached in sodium hypochlorite for 48h, rinsed in freshwater, and air-dried for morphological identification following.^73,110,111 Specimens are currently deposited at King Abdullah University of Science and Technology (KAUST, Thuwal, Saudi Arabia), in F. Benzoni collection (Milan, Italy), or at Florida Museum of Natural History, University of Florida (UF, Gainesville, FL, USA).

Environmental classification of sampling localities and surface waters

Surface (< 0.5 m depth) environmental variability among sampling localities was characterized using data available from Bio-Oracle v2.2.²⁴ The database provides long-term (averaged between 2000 and 2014) global-scale environmental rasters at a resolution of 5 arcmin (∼ 9 km for the study area) for different variables related to climate (temperature, cloud cover, photosynthetically available radiation), chemistry (salinity, pH, dissolved oxygen), nutrients (silicate, nitrate, phosphate), and productivity (chlorophyll, diffuse attenuation). To avoid any biological and ecological assumptions, all available present-day surface variables (except those related to ice) were considered. These 65 environmental variables (Table S3) were downloaded and queried in 25-km buffers centered on sampling locality coordinates using the R v4.0.4⁸⁶ libraries ‘sdmpredictors’⁸⁷ and ‘raster’.⁸⁸ All variables measured at the 17 sampling localities were normalized to have a mean of zero and unit variance, and a principal component analysis (PCA) was performed to reduce collinearity. The first environmental principal components (ePCs) explaining a cumulative variance above 80% were retained and used to compute a matrix of Euclidean distances among sampling localities. From that, an environmental hierarchical clustering of the sampling localities was performed and ‘NbClust’⁸⁹ was used to determine the optimal number of environmental clusters using the ‘kmeans’ method.

To further visualize environmental breaks in surface waters around the AP, a PCA was performed on the same set of environmental variables but considering all 5 × 5 arcmin (∼ 9 × 9 km) surface water cells in the study area (defined from 10°N to 31°N and from 32°E to 61°E). The 65 variables were normalized on these 26,264 values/cells as previously. Individual scores were then plotted as heatmaps for the first ePCs.

Laboratory and bioinformatics steps

Total genomic DNA of the samples was extracted using the DNeasy Blood & Tissue kit (QIAGEN GmbH, Hilden, Germany), according to manufacturer’s protocol. Samples were then PE150 sequenced with an Illumina NovaSeq 6000 (Illumina, San Diego, CA) at the platform iGenSeq (ICM, Paris, France), following a capture protocol targeting 1,127 ultraconserved elements (UCEs) and 1,347 exons (=hexa-v2-scleractinia bait set³⁸), as in Oury et al.¹¹² One or two random samples per species were independently prepared and sequenced twice (sequencing replicates).

Following sequencing, reads were demultiplexed according to individual-specific indexes (no mismatch allowed), then quality checked with FastQC v0.11.7 (http://www.bioinformatics.babraham.ac.uk/projects/fastqc) and MultiQC v1.7¹⁰², before and after adapter contamination and low-quality bases removal with cutadapt v2.1¹⁰³, available in the wrapper script Trim Galore! v0.6.0 (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore). Specific reference sequences for the targeted loci were de novo constructed from the trimmed reads of 10 random individuals per species. These reads were assembled de novo using SPAdes v3.13.0¹⁰⁴ and the resulting scaffolds were matched to the 16,114 RNA capture probes, using the phyluce_assembly_match_contigs_to_probes script from the Phyluce program.¹⁰⁵ All scaffolds assigned to the same locus were then extracted using custom R scripts and a consensus was generated using CAP3.¹⁰⁶ When scaffolds matched multiple loci, these loci were considered as one, but loci producing multiple consensuses were discarded (as representing potential paralogs). Consensuses then went through a second round of matching to the probes for refinement and eventually merging when they were overlapping (indicating physical linkage disequilibrium among loci). The resulting sequences constitute the specific references for subsequent analyses.

All individual trimmed reads were then mapped to the corresponding specific references using BWA v0.7.17.¹⁰⁷ Read sorting and duplicate marking were performed with Picard v2.20.7 (https://broadinstitute.github.io/picard), followed by local realignment with the genome analysis toolkit (GATK) v3.8.1,¹⁰⁸ as in Van der Auwera et al.¹¹³ BCFtools v1.9 (http://samtools.github.io/bcftools) was used for single-nucleotide polymorphism (SNP) calling, treating all conspecific samples simultaneously and requiring a base quality (BQ) and a mapping quality (MQ) of at least 20 and 30, respectively. SNP densities in 1-kbp sliding windows were computed in R and regions with an SNP density greater than twice the average value (indicating potential mapping issues) were discarded. From that, only SNPs with a minimum quality score (QUAL) of 20 and genotypes with a minimum read depth (DP) of 12× and non-significant strand biases (SP ≤ 13) were retained. Finally, one bi-allelic SNP with less than 20% missing data and a minor allele frequency (MAF) of at least 0.05 was randomly retain per locus, leading to the final specific datasets for genetic diversity and population structure analyses. Thereafter, a population was defined as all conspecific colonies sampled in the same locality.

Clonal and genetic diversity

Although the three focal species are encrusting or massive corals and are expected to reproduce asexually only occasionally,¹¹⁴ the absence of clonemates in the sampling was verified following Oury et al.¹¹⁵ Briefly, the number of different alleles (estimated with the diss.dist function from the R library ‘poppr’⁹⁰) over the number of comparable SNPs (i.e., genotyped for both individuals) was computed for each pair of conspecific individuals. The distribution of these genetic distances among individuals was plotted and used to define a threshold separating intra-clonal (in small distances) from inter-clonal (in higher distances) comparisons for each species. Sequencing replicates were used to help position the threshold. Clonal lineages were then visualized in R with a hierarchical clustering of the individuals based on genetic distances, using the hclust function with the UPGMA algorithm (‘average’ method). The clonal richness (R)⁴² of each species and each population was then calculated as $\frac{N_{lineages}-1}{N-1}$, with N and N_lineages, the total numbers of colonies and of clonal lineages, respectively. When appropriate, only one representative (the one with the least missing data) per clonal lineage was kept for further analyses.

Finally, multiple indices were computed per population with the R libraries ‘hierfstat’,⁹¹ ‘PopGenReport’,⁹² and ‘poppr’⁹⁰ to characterize their genetic diversity: proportion of missing data (%NA), number of variable SNPs (N_SNP), allelic richness⁴³ (AR) rarefied to 10 alleles, observed and expected heterozygosities (Ho and He, respectively), and inbreeding coefficient (F_IS).¹¹⁶

Population genetic structure

For each species, assignment tests were performed with sNMF⁴⁰ and discriminant analyses of principal components (DAPC),⁴⁵ implemented in the R libraries ‘LEA’⁹³ and ‘adegenet’,⁹⁴ respectively. Five repetitions per k, with k varying from 2 to 10, were run for sNMF, with a maximum of 500 iterations before reaching stationarity. Results of both methods were visualized with Clumpak.¹⁰⁹ The optimal number of genetic clusters (k) was then defined as the highest k for which sNMF and DAPC remained congruent and for which k+1 did not further partition the dataset (typically the additional cluster implies admixing in similar proportions between two clusters all individuals from a previous cluster). F_ST⁴⁶ were computed with the R library ‘StAMPP’⁹⁵ for each pair of conspecific populations, and significativity was assessed with 1,000 permutations, applying a false discovery rate (FDR)⁴⁴ correction. F_ST values were max-normalized (i.e., divided by the maximum value) for each species and visualized as heatmaps for inter-species comparisons. Additionally, to account for unequal population sizes (ranging from 10 to 13 for D. heliopora, from 1 to 15 for C. monile, and from 1 to 16 for P. profundacella) which can bias F_ST estimations, mean F_ST over 100 random sampling of five (10 for D. heliopora) individuals per population were also computed and compared to raw estimates using one-sample t-tests. Populations with less than five individuals were not considered. Finally, directional gene flow among populations was assessed by constructing a relative migration network with divMigrate,⁴¹ implemented in the R library ‘diveRsity’.⁹⁶ Although the method has not been thoroughly tested and may be subject to uncertainties, especially when sampling sizes are unequal,⁴¹ it has been increasingly used and can provide broad estimates of gene flow relative direction and strength. Analyses were run and averaged over 100 random sampling of five (10 for D. heliopora) individuals per population (thus not considering populations with less than five individuals), using G_ST¹¹⁷ and 1,000 permutations.

Seasonal larval dispersal modeling

Oceanographic drivers of connectivity were assessed using a Lagrangian particle tracking model^33,118 coupled with a general circulation model (MITgcm⁴⁷) configured and extensively validated for water bodies around the AP¹¹⁹ to simulate the seasonal flow of coral larvae among the 17 sampling localities. General circulation was modeled for the study area (including the Maldives) with a horizontal resolution of 0.02° (∼ 2 km) for the period 2000–2018. Simulated daily surface circulation was then used to model coral larval dispersal. In line with previous studies,^34,48,49 and given the limited data available on the reproduction of the studied species, coral larvae were considered positively buoyant, otherwise passive particles that drift with local surface circulation for up to 120 days (based on the maximum competency duration observed for broadcast spawning corals¹²⁰), with a pre-competency period of 3 days and a half-life of 35 days to account for mortality. Particles were re-inserted back to the ocean to avoid getting stranded on the beach. To test for the effect of seasonality and accommodate the different species’ spawning periods previously reported (D. heliopora: May in the Red Sea³⁵; C. monile: April-July in the Arabian/Persian Gulf (P. Range, pers. comm.); P. profundacella: September-November in the Red Sea³⁶), seasonal two-months daily spawning events were simulated in May-June (summer) and in November-December (winter) of each year. Ca. 12,000 particles were released every day from each of 17 squared (0.24 ° × 0.24 °) polygons centered on the coordinates of the 17 sampling localities, yielding ∼ 12.5 million particles for one simulation event. Proportions of particles released in each locality and reaching the other localities were computed for each spawning event, then averaged per season over the 19 simulated years to obtain a matrix of mean dispersal probabilities for each spawning season.

Genetic structure predictors

Mantel tests⁵⁰ were performed to evaluate the correlation between genetic distances (bootstrapped F_ST) and over-water geographic, oceanographic, or environmental distances among populations for C. monile and P. profundacella (D. heliopora was not considered due to the small number of populations sampled, which would have led to imprecise correlations). Over-water geographic distances were computed along the least-cost path (i.e., the shortest path excluding land) using the R library ‘marmap’,⁹⁷ while oceanographic distances corresponded to inverse log-scaled probabilities of dispersal among localities obtained from the summer (C. monile) or winter (P. profundacella) spawning simulations. Environmental distances were calculated as the Euclidean distances among localities for each of the 65 environmental variables previously considered (Table S3) and all or each of the ePCs used for the hierarchical clustering of the sampling localities. Tests were performed with and without the Maldives using the R library ‘ade4’⁹⁸ and 10,000 permutations, applying a FDR⁴⁴ correction. Additionally, as some explanatory variables (especially the environmental ones) may be highly correlated, pairwise correlation tests were performed among all variables. The 65 environmental variables were further hierarchically clustered based on correlations using the hclust function with the UPGMA algorithm (‘average’ method), and groups of highly correlated variables (|r | ≥ 0.75) were identified.

The effect of geography, currents, and environment on the genetic structure of C. monile and P. profundacella (without the Maldives) was further evaluated using redundancy analyses (RDAs). PCAs were performed on Hellinger-transformed allele frequencies⁵³ estimated over 100 random sampling of five individuals per population, and PCs explaining at least 5% of the total genetic variation were retained as response variables. Distance-based Moran’s eigenvector maps (dbMEMs)⁵¹ and asymmetric eigenvector maps (AEMs)⁵² were computed using the R library ‘adespatial’⁹⁹ to decompose over-water geographic distances and oceanographic distances, respectively. AEM variables were generated from a site-by-edge matrix, whose edges were weighted based on the highest dispersal probability between each pair of localities obtained from the spawning simulations. Directionality and acyclicity of the AEM connection diagrams were verified using the is.dag function from the R library ‘igraph’.¹⁰⁰ Centered and scaled environmental values at the sampling localities were decomposed in environmental PCs (ePCs) using PCAs to reduce collinearity. However, as ePCs may be difficult to interpret, analyses were also performed considering one representative environmental variable per collinear group previously defined. RDAs were first performed separately for each type of explanatory variables (i.e., dbMEMs, AEMs, ePCs, or representative environmental variables) to identify significant predictors with a forward selection procedure with 10,000 permutations using the R library ‘vegan’.¹⁰¹ Global and marginal analyses of variance (ANOVA) with 1,000 permutations were used to assess model significance and variable contributions. Finally, partial RDAs were run to partition the variance explained between the three sets of significant predictors (either dbMEMs, AEMs, and ePCs or dbMEMs, AEMs, and representative environmental variables).

Quantification and statistical analysis

All quantifications and statistical analyses are described in the method details. All means are presented with their standard error (±s.e.) and all p-values were adjusted using a FDR⁴⁴ correction. Adjusted p-values below 0.05 were considered statistically significant.

Published: December 13, 2025

Supplemental information

Table S1. Individual metadata

Sampling data, bioinformatic statistics, and NCBI accession numbers for the 306 colonies (+5 sequencing replicates) considered in this study.

Table S3. Environmental variables

List of the 65 present surface environmental variables downloaded from Bio-Oracle v2.2.