From population connectivity to the art of striping Russian dolls: the lessons from Pocillopora corals

Abstract Here, we examined the genetic variability in the coral genus Pocillopora, in particular within the Primary Species Hypothesis PSH09, identified by Gélin, Postaire, Fauvelot and Magalon (2017) using species delimitation methods [also named Pocillopora eydouxi/meandrina complex sensu, Schmidt‐Roach, Miller, Lundgren, & Andreakis (2014)] and which was found to split into three secondary species hypotheses (SSH09a, SSH09b, and SSH09c) according to assignment tests using multi‐locus genotypes (13 microsatellites). From a large sampling (2,507 colonies) achieved in three marine provinces [Western Indian Ocean (WIO), Tropical Southwestern Pacific (TSP), and Southeast Polynesia (SEP)], genetic structuring analysis conducted with two clustering analyses (structure and DAPC) using 13 microsatellites revealed that SSH09a was restricted to the WIO while SSH09b and SSH09c were almost exclusively in the TSP and SEP. More surprisingly, each SSH split into two to three genetically differentiated clusters, found in sympatry at the reef scale, leading to a pattern of nested hierarchical levels (PSH > SSH > cluster), each level hiding highly differentiated genetic groups. Thus, rather than structured populations within a single species, these three SSHs, and even the eight clusters, likely represent distinct genetic lineages engaged in a speciation process or real species. The issue is now to understand which hierarchical level (SSH, cluster, or even below) corresponds to the species one. Several hypotheses are discussed on the processes leading to this pattern of mixed clusters in sympatry, evoking formation of reproductive barriers, either by allopatric speciation or habitat selection.


| INTRODUCTION
Studying population genetic connectivity is first a matter of knowing what we work on, that is, accurately delimiting the evolutionary units. Indeed, the populations among which we want to assess the exchanges of alleles must be composed of individuals that belong to a unique and same species, in order to estimate genetic distances among comparable entities. That being said, it seems trivial, but the increasing discovery of highly divergent populations or divergent clusters among populations reveals the presence of possible cryptic species, somehow by serendipity, that the sole use of traditional taxonomic characters may have not highlighted. So population genetics data collected for estimating connectivity, and more broadly phylogeographic analyses or barcoding, may in turn be used to refine taxonomic knowledge at the species rank, making them hardly ineluctable in an approach of integrative taxonomy. As an illustration, trying to understand the biogeographic shift in the kelp Lessonia nigrescens, Tellier, Meynard, Correa, Faugeron, and Valero (2009) identified two cryptic species using a combination of four genes among 1,000 individuals covering more than 2,500 km of coastline. The two divergent genetic lineages show a parapatric latitudinal distribution: one extends from southern Peru (17°S) to central Chile (30°S), and the other from central Chile (29°S) to Chiloe Island (42°S), both lineages spatially overlapping in a narrow area (29-30°S) in discrete patches where individuals belong to either the northern or southern species. Likewise, studying the ecological interactions between a coral host and its crustacean exosymbionts, Rouzé et al. (2017) used barcoding methods to identify the exosymbionts and found two cryptic species in the shrimp Alpheus lottini, revealing the key role of cryptic diversity in structuring communities of mutualists and the importance of taking into account this diversity in ecological studies to better perceive the complexity of ecological processes. Similarly, Souter, Henriksson, Olsson, and Grahn (2009) studied the connectivity pattern in the coral Pocillopora damicornis in East Africa and, after identifying two cryptic lineages using mitochondrial markers, chose to analyze them separately. Also, exploring the species diversity of the hydrozoans from the Aglaopheniidae family, Postaire, Magalon, Bourmaud, and Bruggemann (2016) revealed extensive lineage diversity and cryptic species in two common species, Lytocarpia brevirostris and Macrorhynchia phoenicea. Then, studying the connectivity of one of the cryptic species within the L. brevirostris complex using microsatellites, Postaire, Gélin, Bruggemann, and Magalon (2017) found a high genetic differentiation among populations, each island housing an independent evolutionary lineage, probably representing different species. In fine, two populations that have diverged enough can be considered as distinct units (e.g., species) on which, several studies, for example, genetic structuring and connectivity, environmental responses in face of perturbation, and conservation plan, will be set up. Nevertheless, the speciation process is slow and gradual. Thus, it is sometimes tricky to put a frontier between different units, notably when the speciation process is not achieved (De Queiroz, 2007), that is, in the gray zone starting from one species and ending to two new ones.
In marine systems, a large number of studies have shown that species with no or low dispersal (often linked with larval phase) tend to present significant genetic structure over small spatial scales while high dispersal abilities are not correlated with population subdivisions (Kelly & Palumbi, 2010). In general, high dispersal species present large population sizes, huge ranges, and rapid gene flow: characteristics that should slow species formation, confirmed by fossil data (Jablonski, 1986). Nevertheless, high dispersal potential does not always lead to high gene flow because of selection (Hilbish & Koehn, 1985), local genetic drift (Reeb & Avise, 1990), and complex homing behavior (Baker et al., 1990). Indeed, habitat preferences could lead to segregate individuals and promote divergence till sympatric speciation even with remaining gene flow (see for reviews, Pinho & Hey, 2010or Bowen, Rocha, Toonen, & Karl, 2013. Among marine organisms, scleractinians represent a good example of species with potentially high dispersal [e.g., larval lifetime of 30 days for Heliopora coerulea (Harii, Kayanne, Takigawa, Hayashibara, & Yamamoto, 2002), 100 days in Pocillopora damicornis (Richmond, 1987), and >200 days in some species from the genera Acropora, Favia, Goniastrea, Monstastrea (Graham, Baird, & Connolly, 2008)], but their dispersal and settlement are constrained by several biotic and abiotic factors (hydrodynamics, light, temperature, gravity, surface texture, or presence of conspecifics; Rodriguez, Ojeda, & Inestrosa, 1993).
Like almost all the morphospecies from the genus Pocillopora, P. eydouxi has been described widely distributed in the Pacific Ocean and Indian Ocean, and the Red Sea, but absent from the Atlantic Ocean.
Some recent studies have revisited Pocillopora taxonomy in light of molecular data. Using species delimitation methods based on mitochondrial markers, Gélin, Postaire, Fauvelot and Magalon (2017) found that P. eydouxi forms a primary species hypothesis [PSH, sensu Pante et al. (2015)], named PSH09 therein [see Gélin, Postaire, et al. (2017) Pante et al. (2015)]: SSH09a is restricted to the Western Indian Ocean, and SSH09b and SSH09c are found in sympatry but restricted to the Pacific Ocean. Generally, colonies belonging to PSH09 present common morphological characteristics: large colonies presenting robust erected or horizontal branches, rounded or flattened, with more or less pronounced verrucae that are uniform in shape and spacing. Nevertheless, the corallum macromorphology is not a diagnostic character in Pocillopora genus [e.g., Paz-Garcia et al. (2015) or Gélin, Postaire, et al. (2017)]. It is present on all reef slopes and less frequently in lagoons, from surface to 40 m (HM, pers. obs.). Its three-dimensional structure is an element of reef structuring, and its broad interbranch width provides habitats for a huge variety of species, making it a key species of coral reef ecosystems from the Indo-Pacific and Red Sea. On a biological point of view, P. eydouxi morphospecies has been described as a broadcast spawner (Hirose, Kinzie, & Hidaka, 2001). Moreover, they evidenced that zooxanthellae were maternally inherited in oocytes, suggesting that larvae are autotrophic and intuitively meaning that their dispersal abilities are not limited by intrinsic resources as lecithotrophic larvae could be. Apart from these studies, focusing mainly on biology, ecology, and taxonomy, no one has investigated yet population genetic diversity in this species complex.
In front of such incomplete knowledge in P. eydouxi, it seems urgent to collect and reinforce data for this too long ignored scleractinian species complex despite its undeniable role in reef architecture and maintenance over the Indian Ocean, the Pacific Ocean, and the Red Sea. Thus, this study aimed to explore the cryptic genetic diversity within PSH09, and more precisely, within each SSH (SSH09a, b, c) using a combination of population genetics data (13 microsatellites) along with different kinds of clustering analyses, and so at different spatial scales (reef < island < ecoregion < province). The hierarchical sampling focused on three understudied provinces (from a genetic connectivity point of view): the Western Indian Ocean, the Tropical Southwestern Pacific, and Southeast Polynesia. This should allow refining species delimitation in this species complex, better estimating reef biodiversity and identifying the units on which connectivity should be assessed.

| Sampling
In the aim of exploring the Pocillopora genus diversity and in fine studying population connectivity, colonies of Pocillopora genus were sampled [tip of branches + photographs except for Tromelin Island (Scattered Islands) and Polynesia] independently of their corallum macromorphology, as it is not a diagnostic character in Pocillopora genus. So species identification was realized molecularly a posteriori of sampling and a priori of data analyses (see below), leading to a subset of 2,507 colonies corresponding to PSH09 sensu Gélin, Postaire, et al. (2017). The sampling was achieved from April 2011 to October 2016, in three marine provinces extended over six ecoregions (Spalding et al., 2007): the Western Indian Ocean (WIO), the Tropical Southwestern Pacific (TSP), and the Southeast Polynesia (SEP). The sampling followed a hierarchical scheme with several islands within a province and several sites within an island (province > ecoregion > island > site; Figure 1, Table 1). It represented a total of 12 islands (included large islands: Madagascar and New Caledonia) and 65 sites. For a given site, colonies were usually sampled at the same depth (8-14 m), during one single dive, so that the range of sampling for each site did not exceed some hundreds of m² and the distance between two colonies within a site varied from few centimeters to few meters, depending on the density of Pocillopora colonies.

| DNA extraction, sequencing, and microsatellite genotyping
From the sampled colonies, DNA was extracted using DNeasy Blood & Tissue kit (Qiagen ™ ). Genotyping was performed following the same  T A B L E 1 (Continued) (Continues) protocol than in Gélin, Postaire, et al. (2017) and so using the same 13 microsatellite loci. PCR products were genotyped using an ABI 3730 genetic analyser (Applied Biosystems), and allelic sizes were determined with GeneMapper v.4.0 (Applied Biosystems) using an internal size standard (Genescan LIZ-500, Applied Biosystems). Because colonies were sampled based on their corallum macromorphology, P. eydouxi lineage identity was verified a priori using assignment tests performed with Structure (Pritchard, Stephens, & Donnelly, 2000).
For this purpose, we combined all sampled colonies for this study and the 943 colonies from Gélin, Postaire, et al. (2017) that were assigned to the different PSHs as in Gélin, Postaire, et al. (2017). From that, a total of 2,507 colonies grouped together in PSH09 and constituted the final dataset.

| MLG identification and assignment to the SSHs
To check for clonal propagation among the sampled colonies, identical multilocus genotypes (MLG) were identified using the R (R Development Core Team 2016) package RClone (Bailleul, Stoeckel, & Arnaud-Haond, 2016).
The colonies showing an assignment probability for each of the five runs >0.75 to a given SSH were assigned to this SSH (practically identified by a unique color on Figures). Colonies that did not show an assignment >0.75 for a given SSH were considered admixed, that is, assigned to more than one SSH (due to hybridization or shared ancestry or bad assignment due to missing data). For each admixed colony, it was assigned to the SSHs presenting an assignment probability >0.1 (a probability < 0.1 was considered as noise). Moreover, we used NewHybriDS v.1.1 (Anderson & Thompson, 2002), running 5 × 10 5 iterations after a burn-in period of 5 × 10 4 , to detect whether the admixed colonies assigned to two SSHs found in Structure could be considered as hybrids. NewHybriDS calculates the posterior probability that sampled colonies fall into each of a set of hybrid categories (Parent 1, Parent 2, F1, F2, backcross to Parent 1, backcross to Parent 2). Once identified, the admixed colonies were not considered in the subsequent analyses.

| Structuring analyses within each SSH
Further analyses were performed on each SSH separately. First, to determine the most likely number of genetically homogenous groups (K) within each SSH, a Bayesian analysis was performed using Structure v.2.3.4 (Pritchard et al., 2000) with the same conditions as described above. This analysis assumes that, within a set of samples, there are K genetic groups and colonies are assigned to each putative genetic groups under Hardy-Weinberg equilibrium (HWE) and minimized linkage disequilibrium (LD). To allow comparing the different outputs from the different types of analyses (see below), the colonies were assigned to clusters following the same rule than above (replacing SSH by cluster, and one color per cluster) and the same rule was applied to constitute the dataset (i.e., removing admixed colonies) for further analyses. In a hierarchical approach, these analyses were repeated on each cluster found in each SSH separately. Commonly, using Structure and DAPC, when the finest level of structuring is reached, adding a supplementary cluster leads to inconclusive assignments with colonies assigned to several clusters in the same proportions. Here, some colonies kept ongoing assigned to clusters when K increased above the most likely K, either using Structure or DAPC. Moreover, because the methods traditionally used to detect the most likely number of genetic groups within a dataset [Pr(X|K) (Pritchard et al., 2000), ΔK method (Evanno, Regnaut, & Goudet, 2005), the deviance information criterion (DIC) (Gao, Bryc, & Bustamante, 2011), the Bayesian information criterion (BIC) (Jombart et al., 2010), and the thermodynamics integration Considering the clusters finally kept within each SSH, the pairwise differentiation among clusters within each SSH was estimated using the F ST (Weir & Cockerham, 1984) with arlequiN v.3.5 (Schneider et al., 2000) and D est using the R package DEMEtics (Gerlach et al., 2010). Then, for each cluster, the allelic frequencies spectrum for each locus were plotted, and considering each SSH, the global F ST (Weir & Cockerham, 1984) for each locus was calculated using FStat v.2.9.3 (Goudet, 2001). Additionally, minimum spanning trees based on the shared alleles distance (DAS) were performed with eDeNetwork (Kivelä, Arnaud-Haond, & Saramäki, 2015) and drawn with colonies colored according to the clusters found in the assignment analyses. Moreover, we used NewHybriDS v.1.1 (Anderson & Thompson, 2002) between pairs of clusters within each SSH to detect whether the admixed colonies found in Structure could be considered as hybrids (same conditions as above). Finally, a hierarchical analysis of molecular variance (AMOVA) was performed using arlequiN v.3.5 (Schneider et al., 2000) considering the whole dataset with SSH as group and clusters within SSH as populations.

| MLG identification and assignment to the SSHs
Among the 2,507 Pocillopora colonies assigned to PSH09 ), each represented a unique MLG. First, assigning the colonies to the three SSHs, all the colonies from the TSP and the SEP (n = 1,075) were assigned to both SSH09b and SSH09c (none to SSH09a), while almost all the colonies from the WIO (n = 1,430) were assigned to SSH09a (n = 1,403), except 12 and 15 that were assigned to SSH09b and SSH09c, respectively ( Overall, this indicates a high genetic differentiation among the three SSHs, indicating restricted gene flow between these SSHs. Thus, from now, we will consider these three SSHs as independent genetic lineages and will perform the subsequent analyses on each SSH separately (N SSH09a = 1,403; N SSH09b = 323; N SSH09c = 781; Table 1).

| SSH09a
Considering all the colonies assigned to SSH09a (exclusively located in the WIO; n = 1,403), DAPC showed nearly similar individual assignments than Structure for both K = 2 and K = 3: for K = 2, we observed two clusters, SSH09a-1 and SSH09a-2, while for K = 3, SSH09a-2 was further divided into two clusters (SSH09a-2 and SSH09a-3, Figure 2). To explore in depth the partitioning, we reanalyzed alone the first cluster (SSH09a-1) found for K = 2: it did not separate anymore using Structure nor DAPC. Conversely, the second cluster (SSH09a-2) when re-analyzed alone did split in three distinct subclusters with Structure while with DAPC, two of these subclusters were overlapping. Then, reconsidering all the colonies and now K = 4, Structure did find four clusters, but they did not correspond to the ones found using DAPC, nor to the ones found when treating the two-first clusters independently (in this case, SSH09a-1 and SSH09a-2 both split into two).
Nevertheless, only 12 of these 93 colonies presented no missing data, suggesting that the other admixed colonies (21% of missing data in average) might be the result of bad assignment. Comparing both methods, only 4.7% of the colonies were assigned differently. The minimum spanning tree retrieved the three clusters ( Figure 2d). Surprisingly, these latter were found nearly evenly distributed in all the sampling sites with no apparent geographical pattern, cluster SSH09a-1 being the most represented (46%; Figure 2). Pairwise F ST estimates among the three clusters were of the same order, between 0.125*** and 0.150*** (Table 3). Comparatively, D est values appeared lower, comprised between 0.085*** and 0.092*** (  Jost, 2008). In parentheses, is indicated the number of colonies. *** P < 0.001.

| SSH09c
Concerning SSH09c (n = 781), for K = 2, the structuring pattern nearly corresponded to a geographical pattern. Indeed, on one hand, colonies from Chesterfield Islands together with those of LOY4, LOY5 (Loyalty Islands) and the 15 colonies from the WIO were grouped in a first cluster (SSH09c-1; Figure 4b), and on the other hand, all the other colonies from New Caledonia (Grande Terre and Loyalty Islands) composed the second cluster (SSH09c-2; Figure 4b). For K = 3,

SSH09c-1 further split into two clusters, the colonies from Loyalty
Islands all appearing differentiated in a third cluster (SSH09c-3) while colonies from Chesterfield Islands segregated in both SSH09c-1 and SSH09c-3. Using DAPC, the same partitioning was observed either for K = 2 or K = 3 ( Figure 4b).
Thus, following our choice criterion, three clusters were identified. At K = 3 in Structure, each colony was assigned identically among the five runs. Then, 756 colonies (97%) were assigned to a unique cluster (N SSH09c-1 = 273, N SSH09c-2 = 302, and N SSH09c-3 = 181), and 25 colonies (3%) were admixed in several clusters: two colonies between SSH09c-1 and SSH09c-2, 11 colonies between SSH09c-1 and SSH09c-3, nine colonies between SSH09c-2 and SSH09c-3, and three colonies among the three clusters. Nevertheless, only four colonies presented no missing data, suggesting that the other admixed colonies (19% of missing data in average) might be the result of bad assignment.
Only 1.9% of the colonies were assigned differently among the two methods. The three clusters were retrieved in the minimum spanning tree (Figure 4d), and genetic differentiation among them was high, either considering F ST (from 0.16*** to 0.34***) or D est (from 0.09*** to 0.31***) estimates (Table 4). Global F ST per locus ranged from 0.001 ns to 0.737*** for Pd4 (mean F ST ± SE = 0.097 ± 0.063; Appendix S5). T A B L E 3 Genetic differentiation between the three identified clusters within SSH09a estimated with Weir and Cockerham's F ST (lower diagonal; Weir & Cockerham, 1984) and with Jost's D est (upper diagonal;Jost, 2008). In parentheses, is indicated the number of colonies . *** P<0.001.

| Russian dolls
The present study focuses on the genetic variability among colonies from the Primary Species Hypothesis PSH09 found in Gélin, Postaire, et al. (2017). PSH09 corresponds to two ORF and four Dloop haplotypes and was attributed P. eydouxi name. In this previous analysis, based on microsatellite loci, Gélin, Postaire, et al. (2017) identified three genetically distinct clusters within PSH09, two of which were found in sympatry, leading to the recognition of three Secondary Species Hypotheses within PSH09 (SSH09a, b, and c). Here, using microsatellite data and assignment tests on a much larger sampling, we further revealed that each of them was found in sympatry with one or the other, with the three being found on a single reef in the WIO (Reunion Island, REU1). Furthermore, according to our choice criterion to determine the number of genetic groups K, we even found that each of these SSHs was partitioned into two to three additional clusters, which were also all found in sympatry at least at the island scale within the distribution range of their respective SSH, with very rare hybrids (only F2 and none F1). Noteworthy, these clusters were as much genetically differentiated from each other as SSHs were, or as SSHs belonging to different PSHs were [see Appendix S13 in Gélin, Postaire, et al. (2017)]. To summarize, when dissecting the genetic variability of PSH09, we revealed several nested hierarchical levels (PSH > SSH > cluster), each level hiding highly differentiated genetic groups as in Russian dolls, not in agreement with geography for most of them.
This pattern of interspersed genetic clusters among populations, revealed using Structure assignments and microsatellite data, has already been reported in the literature. In the coral Seriatopora hystrix in Japan, Nakajima et al. (2017) found three interspersed genetic clusters in sympatry among the sampled sites corresponding to three distinct genetic lineages identified by mitochondrial DNA. Aside from a mix of mitochondrial lineages, the existence of ecological gradients has been proposed to explain a mix of clusters at the site level. As an illustration, Van Oppen, Bongaerts, Underwood, Peplow, and Cooper (2011) found that colonies of S. hystrix in North Australia were assigned in genetic clusters that corresponded to depth. Moreover, they evidenced vertical migration between shallow and deep habitats in each site explaining the mix of clusters found for some particular depths.

| Is this the real life? Is this just fantasy?
Here, we used microsatellite genetic data coupled with different methods to refine lineage boundaries within the P. eydouxi complex (PSH09; Gélin, Postaire, et al. (2017)). Thus, despite the fact that microsatellites are not commonly used in species delimitation but rather in population genetics, their high polymorphism can help in refining boundaries between closely related species (e.g., Dawson et al., 2010).
Although Structure and DAPC differ in their a priori hypotheses, we observed very few differences (3.6% over the whole dataset) between the outputs of these two methods, highlighting their congruency and comforting us that the observed clusters were independent of the method. The main problem when using clustering methods is to determine the true number of genetically differentiated clusters, K, present in the dataset, subject of debate over the last decade. Nevertheless, all the estimators of K traditionally used can all give different values for a given dataset, and not a single estimator can provide a true value of K, as suggested by Jombart et al. (2010). Recently, Verity and Nichols (2016) suggested that K should be viewed as a flexible parameter that describes just one point on a continuously varying scale of population structure. Therefore, comparing the outputs of the different analyses and keeping the highest value of K that gave congruent results among analyses seemed to be a good compromise to estimate the best number of genetic groups.
Overall in the whole sampling (2,507 colonies), we did not find at least two colonies sharing the same MLG, suggesting that the analyzed colonies do not exhibit clonal propagation, in accordance with their massive morphology and their stout branches, limiting fragmentation.
So the genotypic linkage observed in the whole dataset was not due to the presence of repeated MLGs (clones), but to some particular alleles that were preferentially associated within SSH, as well as in SSH09b considering the non-random association of alleles within each cluster.
Thus, (1) the number of private alleles found in each SSH and in each cluster within SSH along with (2) the variation of allele frequencies among SSHs and also among clusters within SSH (high values of global F ST per locus) both played a role in the genetic differentiation observed among SSHs and clusters.
Finally, rather than structured populations within a single species, these three SSHs, and even the eight clusters, likely represent distinct genetic lineages, though incompletely sorted for the ORF gene, engaged in a speciation process (half-way in the "gray zone" between two populations highly differentiated and two sister species) or real species following the unified concept of De Queiroz (2007). In this way, Johnston et al. (2017) highlighted a genetic distinction between P. eydouxi and P. meandrina (which are not diagnosable using the ORF mitochondrial marker), suggesting that both might be recognizable using nuclear DNA. Thus, the SSHs (or even clusters) we observed might be the reflection of the presence of these two genetic lineages among our samples distinguished by the set of microsatellites used.
The issue is now to understand in our case which hierarchical level (SSH, cluster or even below) corresponds to the species one. An extended integrative approach (microstructure, microenvironment ecology, symbionts, and genomics) is needed to fully conclude where to put the boundaries between species, whether these three SSHs actually correspond to three distinct species or whether each SSH would represent a complex of species, each cluster being a single species (or a complex of, more or less cryptic) with its own distribution area.  (Briggs, 1974), in generating the genetic divergences among SSHs. With the onset of Pleistocene glaciation cycles about 3 million years ago (Mya), global sea levels have fluctuated with maximum amplitudes of up to 140 m (Lambeck, Esat, & Potter, 2002). The sea level reached 120 m below present-day level twice over the last two glacial periods, with the last one, occurring ca.

| Origin of divergence
17-18,000 years ago, largely exposing Sunda and Sahul Shelves and restricting water exchanges between Indian Ocean and Pacific Ocean (Bard et al., 1996;Voris, 2000). As a consequence, the opportunity for genetic exchange for marine organisms between the two oceans strongly decreased and populations evolved independently on both sides of this semi-permeable barrier, providing an occasion for population differentiation and incipient speciation to occur. Meanwhile, these sea-level regressions profoundly affected the distribution of shallow-water reef habitats (Lambeck & Chappell, 2001), generating population size reductions in reef-associated organisms, which may have caused local extinction of some species (Fauvelot, Bernardi, & Planes, 2003). Although the magnitude and timing of relative sealevel stands vary across the Indo-Pacific region (Woodroffe & Horton, 2005), during low sea-level periods, viable populations survived in coral reef refuges that were isolated from each other (Pellissier et al., 2014). Once sea levels rose again (to reach present-day level), surviving populations, by then forming differentiated lineages, re-expanded to reach their current geographic distributions (Fauvelot et al., 2003).
Then, several nonexclusive hypotheses may explain the nested partitioning within each SSH (i.e., the presence of differentiated clusters in sympatry at the island scale): 1. In all our provinces, genetically divergent clusters within SSH the colonies studied (Baums et al., 2014;Pinzón et al., 2012) were described as P. damicornis-like colonies, a morph that presents thin branches susceptible to break, favoring fragmentation.

3.
Populations experiencing a reduction in gene flow (such as in allopatric speciation processes) could exhibit heterogeneity in genetic differentiation along the genome (Tine et al., 2014). This heterogeneity could lead to consider two populations as two distinct species when looking at the highly differentiated zones of the  (Combosch & Vollmer, 2015). Frequent events of hybridization could comfort Veron's suggestion about metaspecies (or syngameon) existence in corals (Veron, 1995).
Whichever the causes, PSH09 (and even the whole Pocillopora genus) may represent a metaspecies with some hybrids between entities (such as the colonies that were assigned to two different SSHs or clusters herein). Nevertheless, a recent study resolving the phylogenetic relationships among seven species of Pocillopora using RAD-seq did not provide any proof for hybridization among P. eydouxi and other Pocillopora species which were all found to be reciprocally monophyletic, although possible introgressive hybridization may have occurred between the most recently derived sister species P. damicornis and P. acuta (Johnston et al., 2017). while the colonies from two sites were found at different depths (MAD7, a reef flat between 1 and 2 m and MAD9, a pinnacle between 13 and 26 m), no genetic differentiation was found related to depth. Likewise, previous studies failed to detect a link between depth and genetic groups in the Seriatopora genus (Flot et al., 2008;Nakajima et al., 2017). All in all, while we cannot fully reject this hypothesis, depth does not appear an explanatory factor of genetic differentiation in our case, but microhabitat might be, considering fine-scale variations of both abiotic and biotic factors that could be indiscernible to the human eye. Particularly, coral larvae are attracted to the substratum by sensory receptors (e.g., Tran & Hadfield, 2012) and are notably sensitive to biofilms produced by crustose coralline algae (e.g., Morse, Hooker, Morse, & Jensen, 1988). Therefore, the affinity for different kinds of coralline algae may play a role in micro-habitat selection, as revealed in the pea aphid for which chemosensory gene families are determinant in host plant specialization (Smadja et al., 2012). To date, we cannot favor one or another hypothesis and more investigations are needed to fully conclude regarding the origin of the genetically divergent clusters observed within PSH09. Several hypotheses imply an ancestral reduction in gene flow that created reproductive barriers or genome incompatibilities among the different clusters for each SSH, which contemporary gene flows have not homogenized yet.

| CONCLUSION AND PERSPECTIVES
Examining the population structure of the Pocillopora eydouxi species hypothesis (PSH09; Gélin, Postaire, et al. 2017) revealed a nested partitioning of the different SSHs, obliging to think about the unit on which connectivity should be assessed. Whatever the causes, facing to this over-partitioning of our dataset, the matter is not how to estimate connectivity but on what. As each SSH is a mix of several genetically differentiated clusters found in sympatry, we prefer considering the eight different clusters as our reference unit to assess genetic differentiation among populations.
Several hypotheses have been proposed to explain the observed pattern, but more investigations are needed to understand the structuring pattern of this species complex. Knowing more about Pocillopora genome seems now to be a fundamental key to improve the understanding of its history of divergence and would offer clues to favor one or another exposed hypothesis.

CONFLICT OF INTEREST
None declared.