Contrasted levels of genetic diversity in a benthic Mediterranean octocoral: Consequences of different demographic histories?

Abstract Understanding the factors explaining the observed patterns of genetic diversity is an important question in evolutionary biology. We provide the first data on the genetic structure of a Mediterranean octocoral, the yellow gorgonian Eunicella cavolini, along with insights into the demographic history of this species. We sampled populations in four areas of the Mediterranean Sea: continental France, Algeria, Turkey, and the Balearic and Corsica islands. Along French coasts, three sites were sampled at two depths (20 and 40 m). We demonstrated a high genetic structure in this species (overall FST = 0.13), and most pairwise differentiation tests were significant. We did not detect any difference between depths at the same site. Clustering analyses revealed four differentiated groups corresponding to the main geographical areas. The levels of allelic richness and heterozygosity were significantly different between regions, with highest diversity in Algeria and lowest levels in Turkey. The highest levels of private allelic richness were observed in Algeria followed by Turkey. Such contrasted patterns of genetic diversity were not observed in other Mediterranean octocorals and could be the result of different evolutionary histories. We also provide new empirical evidence of contrasting results between tests and model‐based studies of demographic history. Our results have important consequences for the management of this species.

have been studied by Maggs et al. (2008), who proposed a theoretical framework to study glacial refugia and recolonization in North Atlantic benthic species. Their predictions are based on lower levels of genetic diversity after recolonization (a pattern potentially erased by secondary contacts; Petit et al., 2003). The reconstruction of demographic history, on the basis of sequence polymorphism, also suggested demographic expansion for three benthic species following sea-level rise in the Sunda Shelf (Crandall, Sbrocco, Deboer, Barber, & Carpenter, 2011). Nevertheless, the impact of past climatic fluctuations on the current genetic diversity remains to be studied for numerous marine species and oceanic basins.
The Mediterranean Sea is an interesting geographical and environmental context for the study of the demographic history of marine species. It comprises different basins with different current and past environmental conditions (Hayes, Kucera, Kallel, Sbaffi, & Rohling, 2005). A dozen different biogeographical areas have been described in the Mediterranean Sea which is a biodiversity hot spot (Bianchi et al., 2012). For numerous species, the different basins correspond to different genetic units (Borsa et al., 1997) which could have evolved more or less independently. The past sea-level variation added additional constraints to marine species, with a level 120 m lower than present at the LGM around French coasts (Hayes et al., 2005;Lambeck & Bard, 2000).
Differences in levels of genetic diversity between basins have been demonstrated in several cases. Reduced levels of genetic diversity have been observed in Adriatic and Black Seas for the sprat Sprattus sprattus (Limborg et al., 2012), in the Eastern Mediterranean for the red gorgonian, Paramuricea clavata (Mokhtar-Jamaï et al., 2011), or for deep populations of the red coral, Corallium rubrum (Costantini et al., 2011; but see Cannas et al., 2016). In the seagrass Posidonia oceanica, higher genetic diversity has been observed in central populations, potentially as the consequence of a secondary contact between Western and Eastern populations . Different approaches allow the study of demographic history which might explain the observed differences in genetic diversity (eg, Beaumont, 1999;Cornuet & Luikart, 1996;Drummond, Rambaut, Shapiro, & Pybus, 2005;Girod, Vitalis, Leblois, & Fréville, 2011;Rogers & Harpending, 1992). In all cases, genetic structure can bias the results and should be taken into account for such approaches (Städler, Haubold, Merino, Stephan, & Pfaffelhuber, 2009).
Octocorals are good models to study patterns of genetic diversity and demographic history in the Mediterranean Sea. Previous studies have identified well-differentiated populations for these sessile species (eg, Costantini, Fauvelot, & Abbiati, 2007b;Mokhtar-Jamaï et al., 2011). These species present low dispersal abilities (Costantini, Fauvelot, & Abbiati, 2007a;Garrabou et al., 2009;but see Martínez-Quintana, Bramanti, Viladrich, Rossi, & Guizien, 2015), and they could be more impacted by sea-level and climatic fluctuations than species with higher dispersal. No clear past demographic fluctuations have been demonstrated for the red coral Corallium rubrum in the Mediterranean Sea  on the basis of tests of mutation-drift equilibrium, but other methods could be more informative (Girod et al., 2011).
We studied here the genetic diversity and the genetic structure of the yellow gorgonian, Eunicella cavolini (Koch 1887) (Figure 1), one of the most abundant gorgonians in the Mediterranean (Weinberg, 1978). E. cavolini was impacted by mortality events linked with thermal anomalies over the past two decades with variable levels of necrosis depending on location, depth, and individuals (Garrabou et al., 2009).
Its wide range, from Western Mediterranean to Marmara Sea, allows comparing the history of different basins. Up to now, there was no extended genetic study on this species because of a lack of adequate molecular markers (Calderon, Garrabou, & Aurelle, 2006).
Our aim was to study the genetic diversity of E. cavolini in different parts of the Mediterranean Sea. First, we will describe the genetic structure of this species at different spatial scales. We include a comparison between depths to test the differentiation along an environmental gradient. We will then test whether populations from different geographical areas present the same levels of diversity and similar demographic histories. We will study past demographic events with tests of mutation-drift equilibrium and with estimates of current and past effective sizes. These results will be useful for the management of this ecologically important species (Ballesteros, 2006).

| Sampling
Five hundred and eighty-four individuals of the yellow gorgonian  (Table 1). Small fragments (3-5 cm) were collected randomly (approximately 30 colonies sampled per site) and then preserved in 95% ethanol at −20°C for further use.

| Molecular markers
Total genomic DNA was extracted using two methods: either the QIAamp ® DNA Mini Kit (Qiagen) following the manufacturer's instructions or a salting-out procedure (Mokhtar-Jamaï et al., 2011). All individuals were genotyped at seven microsatellite loci: C21, C30, C40, S14 (Molecular Ecology Resources Primer Development Consortium et al., 2010), Ever007, Ever009 (Holland, Dawson, Horsburgh, Krupa, & Stevens, 2013), Mic56 (This study). All loci were amplified according to the PCR protocols described in Appendix S1. PCR products were analyzed on an ABI 3130 Genetic Analyser using an internal size standard  Table S1). For the following analyses, only one representative of each MLG was retained corresponding to a final set of 575 samples.
GENETIX was used to compute single and multilocus F IS on the basis of the estimator of Weir and Cockerham (1984), and its significance was tested with 1000 permutations. The HP-Rare software (Kalinowski, 2005) was used to estimate allelic richness [Ar(g)] and private allelic richness [Ap(g)] with a rarefaction analysis and 18 as minimum sample size. Differences in genetic diversity and allelic richness were tested between groups of populations, using the one-sided probability test implemented in FSTAT 2.9.3.2 software (Goudet, 2001). The groups of populations were defined on the basis of geographical location, and of clustering analyses, and were Turkey, Algeria, France, and islands (Corsica and Menorca).

| Demographic history
In order to test whether the analyzed population underwent recent population changes, we used two different approaches. First, the null hypothesis of mutation-drift equilibrium was tested using the software BOTTLENECK 1.2.02 (Piry, Luikart, & Cornuet, 1999). The tests were based on 1000 replicates under a two-phase mutation model (TPM) with 95% of the stepwise mutation model (SMM) and variance among multiple steps equal to 12 (Cornuet & Luikart, 1996).
Second, we used the MSVAR 1.3 software to evaluate the most probable demographic history on the basis of Markov Chain Monte Carlo (MCMC) simulations (Beaumont, 1999). First, we tested the sensitivity of the software to different starting points concerning ancestral and current effective sizes (respectively N anc and N curr ) on one population: We used either the same distributions for N anc and N curr , or distributions indicating either reduction or expansion of populations. As this led to similar results indicating population decline (data not shown), we focused for the main analyses on an approach without a priori, with the same distributions for N anc and N curr . Considering computation time, differentiation comparison (see results). For continental France, the two retained samples corresponded to two depths and sites, and gave different results with BOTTLENECK. We also analyzed each region separately by grouping the corresponding population samples.
In order to evaluate the impact of mutation model on the obtained results, we analyzed the results for the seven loci separately in the Algerian region. As genetic structure can impact the results of MSVAR analysis, an analysis at the deme level inside the French region was launched using two population samples in that region. We also pooled two individuals from the 12 population samples (total: 24 individuals) from the Marseille area, as an approximation to the method proposed by Chikhi, Sousa, Luisi, Goossens, and Beaumont (2010). This was not done in other areas because of a reduced number of independent samples. The parameters used for the MSVAR analyses are provided in Appendix S1 (Tables S2 and S3). We ran four independent chains with identical priors and starting values for each region. Each chain led to 20 000 lines of output. We tested the convergence of the MCMC chains with the Brooks, Gelman, and Rubin statistic (Brooks & Gelman, 1998;Gelman & Rubin, 1992). Values of the multivariate Gelman and Rubin's convergence diagnostic between 1.0 and 1.1 indicate reasonable convergence, whereas values >1.1 indicate poor convergence. In this regard, the last 10 000 output lines of each chain were retained to make a combined consensus chain of 40 000 data points for each region, which was assumed to summarize the posterior distribution of N anc and N curr (Storz & Beaumont, 2002). The output of MSVAR was analyzed by focusing on the detection and on the direction of demographic changes (expansion or contraction). We also compared the magnitude of changes between regions using both natural (N curr , N anc ) and scaled parameters (θ curr = 4N curr μ, θ anc = 4N anc μ) over the four replicated data sets. All outputs were analyzed with the R CODA package (Plummer, Best, Cowles, & Vines, 2006).

| Genetic structure
Pairwise F ST were calculated with GENETIX according to Weir and Cockerham (1984). Their significance was tested with 1000 permutations. The excluding null allele (ENA) method in FreeNA (Chapuis & Estoup, 2007) was used to calculate pairwise F ST to avoid potential bias induced by null alleles. As a complementary estimate of genetic differentiation, we computed the Jost's D statistic (Jost, 2008) with the SMOGD software (Crawford, 2010). between populations (Rousset, 1997). The correlation was tested with a Mantel test (n = 10 000 permutations) in IBDWS 3.16 (Jensen, Bohonak, & Kelley, 2005).
An analysis of molecular variance (AMOVA) was performed with F ST and R ST estimators with ARLEQUIN v.3.5 (Excoffier & Lischer, 2010) and by using the main geographical areas as groups, that is, Turkey, Algeria, continental France, and Menorca and Corsica islands. For these last two islands, we conducted the AMOVA both by separating and by grouping them, as the STRUCTURE analysis grouped them (see results). One thousand permutations were used to test the significance of the different estimates of fixation indices of the AMOVA.
A clustering analysis was performed with the Bayesian method implemented in STRUCTURE v.2.2 (Falush, Stephens, & Pritchard, 2003Pritchard, Stephens, & Donnelly, 2000) launched with admixture model, 500 000 iterations after a burn-in period of 50 000, and 12 replicates for each configuration. A first round of analyses was launched with the whole data set to assess structure at the Mediterranean scale with K varying from 1 to 16. A second round of analyses was performed on each genetic group depicted by the initial round with the same parameter set of the first round, and K varying from 1 to 12 for France, and 1 to 5 in other cases. The outputs were analyzed through the STRUCTURE HARVERSTER website (Earl, 2012) to choose the value that captured the major structure in the data. The number of clusters was estimated based on the Delta (K) criterion (Evanno, Regnaut, & Goudet, 2005).
To analyze genetic structure without relying on the model implemented in STRUCTURE, we performed a discriminant analysis of principal components (DAPC; Jombart, Devillard, & Balloux, 2010) implemented in the adegenet R package (Jombart, 2008). Data were analyzed in two rounds, one with all samples and a second round with French samples only. The number of clusters was determined based on the Bayesian information criterion (BIC).
In all cases, for multiple tests, significance levels were corrected using a 5% false discovery rate (FDR) (Benjamini & Hochberg, 1995).

| Genetic diversity
The total number of alleles per locus ranged from eight for Ever007 to 40 for Mic56 and a mean value of 18 alleles (

| Demographic fluctuations
The analysis of departure from mutation-drift equilibrium using BOTTLENECK indicated no significant heterozygosity excess, expected following a bottleneck, on the basis of one-tailed Wilcoxon test (Table 3). On the other hand, significant heterozygosite deficiency, expected after population expansion, was detected in 10 samples over 19, and nine tests remained significant after FDR correction. These signals of expansion were observed in five over twelve northern populations (Marseille) and in Algeria and Balearic Islands.
Contrastingly, MSVAR results indicated a strong historical decline for all the analyzed samples whether separately (Appendix S3, Figure S1, Table S1) or pooled per region ( Figure 3; Appendix S3, Table S2). At the region level, the inferred N anc /N curr reached 10 4 in Turkey and around 2 × 10 3 in Algeria (Appendix S3, Table S2). When considering the scaled parameters, the current value of θ = 4Neμ was again highest in Algeria, intermediate in France and islands, and lowest in Turkey. The θ estimate was ten times higher in Algeria than in Turkey (0.21 vs 0.02, respectively; Table 4; Appendix S3, Figure S2). Current effective sizes were also lower in Turkish samples than in other samples at deme level (mean values: 0.78 for AYV, and 0.67 for SIV; Appendix S3, Table S1). In the French region with a pool of 24 individuals, two for each site, results also indicated a population decline, but at a lower intensity than with the regional analysis with two demes (N anc /N curr around 300 and 9 × 10 3 for the pool and the regional analysis, respectively; Appendix S3, Table S3, Figure S3). The inference of demographic decline was coherent along all our loci in the Algerian area, all indicating signatures of population declines (Appendix S3, Figure S4, Table S4).   Table S4). The smallest geographical distance for which significant genetic differentiation was observed was 763 m, with F ST = 0.02 for KIA vs SPI.

| Genetic structure
The AMOVA indicated significant differences between geographical groups of samples, both by separating the two islands (Table 5) and by grouping them (Appendix S2, Table S5). For the analysis separating the two islands, differences among groups were significant (F CT = 0.19 and F CT = 0.04 with F ST -like and R ST -like analyses, respectively; Table 5). The differences between populations within groups appeared significant with F ST but not significant with R ST (F SC = 0.03 and −0.01, respectively). A significant positive correlation was evident between genetic distances and the logarithm of the geographical distances, indicating a pattern of IBD at the Mediterranean scale (R² = .567, p < .0001; Appendix S2, Figure S1) and within the French region (R² = .169, p = .009; Appendix S2, Figure S2). At the global scale, the IBD pattern seemed to be separated in two parts with a lower slope at short distance compared to a much higher slope at higher distances. French samples than to other areas but well separated from them.

| Clustering analysis
The first round of STRUCTURE with K = 2 separated samples from Algeria and Turkey in cluster 1 and samples from France, Menorca, and Corsica in cluster 2 ( Figure 5). For K = 3, samples from France, Menorca, and Corsica were assigned to cluster 1, while samples from Algeria and Turkey were separated in two different clusters (2 and 3, respectively). At K = 3, four replicates over twelve grouped Algerian and Turkish samples and were not retained here. For K = 4, samples from islands Menorca and Corsica were assigned to a new group, while other samples were clustered as above but with France partly admixed with the islands cluster ( Figure 5). The Delta(K) criterion indicated K = 5 as the best clustering solution (Appendix S2, Figure S3). With K = 5, Algerian samples were in cluster 1, Menorca and Corsica in cluster 2, Turkey in cluster 3 and all French samples subdivided between clusters 4 and 5 but with high admixture between these two putative clusters.
For the second STRUCTURE analysis on French samples, K = 3 was the best solution followed by K = 6 (Appendix S2, Figure S4), but there was no clear genetic structuring (Appendix S2, Figure S5).
A STRUCTURE analysis on Turkish samples alone indicated a clear separation of both populations at K = 2 (Appendix S2, Figure S6). The STRUCTURE analysis on Menorca and Corsica samples indicated a distinction between these two islands but with quite high levels of admixture (Appendix S2, Figures S7 and S8).
For the DAPC analysis, the Bayesian information criterion (BIC) was minimal between K = 15 and 21 but without a single clear informative value (Appendix S2, Figure S9). We present here the results obtained with K = 15 clusters, and other analyses around this value gave similar results. Higher K values did not bring more information on the general structure at the Mediterranean scale. The analysis confirmed the main groupings evidenced with STRUCTURE but with the additional separation between the two Turkish samples from Marmara Sea and Aegean Sea in clusters 11 and 14, respectively ( Figure 6).
Samples from Algeria were assigned to clusters 2 and 7. The two samples of Menorca and Corsica were mainly grouped in clusters 5 and 15, respectively, while French samples were mainly assigned to the remaining clusters (Table 6). High percentages (>0.70) of reassignment to the original clusters were observed apart from clusters 1, 4, and 10 corresponding to samples from France (Appendix S2, Table S6). A F I G U R E 4 Plot of the first two axes from the principal coordinate analysis based on Nei's unbiased genetic distance. Percentage of variation explained by axis 1: 51.1. and by axis 2: 22 T A B L E 5 Results of AMOVA. The groups of populations were defined on the basis of geographical location and of clustering analyses and were Turkey, Algeria, France, and islands (Corsica and Menorca)

| DISCUSSION
We have demonstrated (1) strong genetic structure between samples from different regions in the Mediterranean, (2) we did not observe any significant differentiation between depths for a given site in France, and (3)

| Genetic structure of E. cavolini and comparison with other Mediterranean octocorals
We identified four main clusters corresponding to geographical subdivisions: northwestern Mediterranean, Balearic and Corsica islands, and Algeria and Turkish samples. These differences between regions were statistically significant. These results can be discussed in the more general context of the biogeography of the Mediterranean Sea.  (Borsa et al., 1997).
Concerning E. cavolini, the strong differentiation between eastern and western Mediterranean samples could be explained by several potential oceanographic barriers, including the Siculo-Tunisian strait, but their exact location remains to be studied (Berline, Rammou, Doglioli, Molcard, & Petrenko, 2014). Additionally, the gaps in the distribution range of E. cavolini between Turkey and Algeria could contribute to this differentiation (Sini, Kipson, Linares, Garrabou, & Koutsoubas, 2014). Isolation by distance could lead to the identification of wellseparated clusters as well, if distant populations are analyzed without geographical intermediates (Aurelle & Ledoux, 2013). Nevertheless, in most cases, the precise location of the genetic break could not be determined, especially its position relative to the Marmara  (Beşiktepe et al., 1994). This, along with the strait systems delimiting the Marmara Sea, provides a strong isolating factor for octocorals, which are restrained to deeper locations.
In all cases, considering the important differentiation observed between some of these clusters, especially the eastern-western differentiation, genetic incompatibilities may contribute to the observed differentiation as well (Bierne, Welch, Loire, Bonhomme, & David, 2011). Genome scan approaches would be useful here to go further on this topic.

| Genetic structure at regional and local scales
In the Marseille area, the maximum pairwise F ST reached 0.07 for populations separated by 15 km. Such local genetic structure has been demonstrated for other octocorals in this area, with maximum F ST reaching 0.2 for C. rubrum  and 0.1 for P. clavata (Mokhtar-Jamaï et al., 2011). This has been linked to reduced dispersal abilities of the larval stage in Mediterranean octocorals (Martínez-Quintana et al., 2015). In E. verrucosa, the lecithotrophic larvae are supposed to have a short, but unknown life span (Sartoretto & Francour, 2011). In E. singularis, experimental results indicated that in the presence of favorable substrates, settlement could take place in less than 30 hr (Weinberg & Weinberg, 1979). If similar larval traits are present in E. cavolini, this could explain, along with important genetic drift, our observation of a strong local genetic structure.
No significant differentiation was evidenced between samples from different depths within the same sites near Marseille as observed in a preliminary study (Pivotto et al., 2015). This suggests the occurrence of regular gene flow or low genetic drift that leads to a genetic homogeneity between depths. This was also observed for E. singularis (Cataneo, 2011), but it contrasts with previous findings of genetic structure between depths for C. rubrum (Costantini et al., 2011; and P. clavata (Mokhtar-Jamaï et al., 2011). Such differences between species could be linked to the buoyancy or the vertical movements of the larvae of these species. The precise timing of larval release, relative to the onset of thermocline, could explain these results and would require a precise study of phenology according to water stratification. The observation of a lack of genetic differentiation between depths despite clear thermotolerance differences questions the possibility of local adaptation in E. cavolini (Pivotto et al., 2015). In the Carribean octocoral Eunicea flexuosa, adaptation to different depths coincided with distinct genetic lineages (Prada & Hellberg, 2013). For E. cavolini, an intron locus seems to indicate significant differences according to depth and could be linked to a selected polymorphic locus (Aurelle et al. submitted). It will thus be necessary to study more loci to test for possible genetics-environment associations.

| Contrasting results between BOTTLENECK and MSVAR approaches
Concerning demographic history, BOTTLENECK tests and estimates of past versus current effective sizes gave contrasting results.
Whereas BOTTLENECK indicated either no demographic fluctuation or population expansion, the MSVAR approach suggested a generalized population decline with different strengths. Such discrepancies between these two methods have been demonstrated by Girod et al. (2011): In a simulation of population decline, these authors observed that MSVAR could indeed detect the correct demographic change, but in some simulation cases with the oldest changes, BOTTLENECK suggested a population expansion. We provide here additional empirical evidence of contrasting results between tests and model-based studies of demographic history. The MSVAR analysis can lead to false inferences of population decline in cases of strong departures from a stepwise mutation model (SMM; Girod et al., 2011;Faurby & Pertoldi, 2012) or in case of underlying genetic structure (Chikhi et al., 2010).
Concerning departures from SMM, the inference of demographic decline was coherent along all our loci which present different levels of variability and distributions of allele sizes. Regarding genetic structure, the analysis at the deme level inside regions gave similar results to pooled samples, but the analysis of a single deme can lead to spurious inference of decline as well (Chikhi et al., 2010). A decline, though less strong, was also inferred for a pool of individuals scattered along different demes as suggested by Chikhi et al. (2010). These observations suggest that E. cavolini populations were indeed impacted by a demographic decline, but the estimates of the magnitude of this decline may be biased by population structure. As suggested by Faurby and Pertoldi (2012), we focus the following interpretation of MSVAR results on the inferred relative levels of current effective size.
At the deme level, the inferred current effective sizes were much lower in Turkey samples than in other Mediterranean areas.

| Genetic diversity and inferences on evolutionary history
E. cavolini is a high diversity species among metazoans (Romiguier et al., 2014). We evidenced here that this diversity is highly heterogeneous among regions as a potential result of differences in effective sizes, a result which remained robust when considering estimates of scaled effective size. Turkish samples displayed the lowest levels of genetic diversity, whereas Algerian samples displayed the highest diversity (a twofold difference in allelic richness and a 42% reduction in expected heterozygosity in Turkey compared to Algeria).The lower reduction in heterozygosity compared to allelic richness is expected as a loss of rare alleles has a higher impact on the latter. Differences in the levels of genetic diversity between populations can be the consequence of different evolutionary histories (eg, bottleneck) and on differences in effective size for a same history. The lowest levels of genetic diversity in Turkey can be discussed according to the peculiar history of the Marmara Sea. Several scenarios of connection between Black Sea and Mediterranean Sea after the last glacial maximum have been proposed. Ryan et al. (1997) suggested that during the last glacial maximum, the Black Sea became a giant freshwater lake and that the Mediterranean Sea refilled it. A second hypothesis suggested that it was instead the Black Sea that first breached the Bosphorus and overflowed into the Marmara Sea (Aksu et al., 2002). In both cases, they imply recent colonization events for the Marmara Sea and possibly for the neighboring part of the Aegean Sea. A reduced diversity of Turkish populations compared to Western Mediterranean populations has also been observed in the red gorgonian P. clavata, but with only one Turkish population considered (Mokhtar-Jamaï et al., 2011). The population density of E. cavolini in the neighboring northern Aegean is also lower than in western Mediterranean, and recruitment frequency seems reduced in eastern populations as well (Sini, Kipson, Linares, Koutsoubas, & Garrabou, 2015). This agrees well with our inferences of lower current effective size and lower diversity in this area. The high level of private allelic richness (second after Algeria) in this area also points to an historical isolation of these populations (Waples, 2010) which could have strengthened the regional loss of diversity.
The highest diversity was observed in Algeria, but no demographic studies have been carried out for southern populations for comparison with northern ones. The highest private genetic richness observed here also underlines the isolation of this southern cluster. This could indicate that this area corresponded to a glacial refugia (Maggs et al., 2008) or, at least, to an area where the environmental conditions would have allowed a better demographic stability of these populations than in other regions. Quaternary climatic fluctuations led to variations in sea temperature and sea level. Winter surface temperature was estimated to be around 7°C in winter in the Gulf of Lion (compared to 13°C for present day) and the sea level was 120 m lower than present around French coasts (Hayes et al., 2005;Lambeck & Bard, 2000). These variations could have had different demographic impacts according to location, as the cooling was lower along Algerian coasts (Hayes et al., 2005). If southern areas were refugia for E. cavolini, the recolonization of northern Mediterranean could have led to a "southern richness to northern poverty" (Hewitt, 2000). For E. cavolini, the marked differentiation between northern and southern populations could point to an alternative scenario: There could be a northern refugia more affected by climatic fluctuations than southern one. A higher diversity in southern compared to northern populations has not been observed for other octocoral populations, apart for E. singularis Cataneo, 2011;Mokhtar-Jamaï et al., 2011). This could suggest different responses of octocorals to climate fluctuations with a higher sensitivity of E. cavolini to past climatic variations.

| CONCLUSION
The yellow gorgonian E. cavolini presents distinct genetic units depending on geographical locations with contrasted levels of genetic diversity. Therefore, protection of genetically rich populations (eg, in Algeria) or with an important private diversity (eg, in Turkey) should be a priority. This is particularly important when considering the current pressures on this species. In Turkey, some relatively dense E. cavolini populations, restricted to few areas in the Marmara Sea, are under various threats such as fisheries (Topçu & Öztürk, 2015). Their genetic particularity could make these populations more vulnerable against such threats that tend to lower their abundance. The impact of local environmental conditions on such species should be considered for protection. Genomic studies of this species could open the way to a better understanding of its evolution and adaptation in a heterogeneous and fluctuating environment. It would also be useful for a better understanding of the evolutionary history of this species.

ACKNOWLEDGMENTS
We acknowledge the help of the staff of the molecular biology service