Seagrass meadows at the extreme of environmental tolerance: the case of Posidonia oceanica in a semi-enclosed coastal lagoon

Problem

Seagrass meadows are recognized as among the most widespread and productive coastal ecosystems worldwide (Hemminga & Duarte 2000), providing high-value ecosystem service, comparable with terrestrial habitats such as rainforests (Costanza et al. 1997). Indeed, seagrasses can be considered ‘ecosystem engineers’ given their ability to affect significantly physical, chemical and biological features of their environment (sensuWright & Jones 2006).

The endemic Mediterranean seagrass Posidonia oceanica (L.) Delile (den Hartog 1970) forms dense meadows from the surface down to a depth of 45 m and covers 2–4% of the whole benthic substratum in the basin (approximately 38,000 km²; den Hartog 1977; Procaccini et al. 2003). Sexual reproduction is irregular and is sporadic in some localities (Diaz-Almela et al. 2006, 2007), whereas clonal growth by means of stolons seems to be the predominant way of meadow expansion (Migliaccio et al. 2005; Arnaud-Haond et al. 2007b; Rozenfeld et al. 2007). Large clones can extend over more than 100 m (Ruggiero et al. 2002; Migliaccio et al. 2005; Zupo et al. 2006). Nevertheless, populations generally show high genetic variability as a result of sexual reproduction or immigration of foreign genotypes (Arnaud-Haond et al. 2005, 2007b). Posidonia oceanica can occur in salinities between 33‰ and 39‰ (Gobert et al. 2006; Sànchez-Lizaso et al. 2008) and tolerate temperatures between 9 °C and 29 °C (Boudouresque & Meinesz 1982). Coastal lagoons are stressful environments for P. oceanica because environmental conditions can reach the survival limits of this species (La Loggia et al. 2004). The Stagnone di Marsala, a semi-enclosed lagoon situated at the western coast of Sicily, is a singular environment where P. oceanica forms peculiar reef-like structures (Calvo & Fradà Orestano 1984). In some areas of the lagoon, these reefs are atoll-shaped, which is a very rare feature of Posidonia meadows. This phenomenon is also observed in small areas along the Tunisian, Turkish and Corsican coasts (Blanpied et al. 1979; Boudouresque et al. 1990; Pasqualini et al. 1995). Pergent & Pergent-Martini (1995) and Boudouresque et al. (2006) hypothesized that atolls derive from nearly circular patches where plagiotropic (horizontal) shoots only grow outwards, whereas the shoots on the central portion of the atolls die. According to this hypothesis, one or a few genotypes should be the founders of such structures without subsequent recruitment within the patch itself. In the Stagnone di Marsala, clear regression of P. oceanica stands has been recorded over the last decades, both in the overall distribution of the meadows and in the number and diameter of atolls (personal observation by the authors).

Our initial observations suggest that P. oceanica atolls in the Stagnone di Marsala occur at the limits of the species’ tolerance, with temperature and salinity values reaching 30 °C and 48‰, respectively (Mazzola & Vizzini 2005). We focused our research on this peculiar habitat because global change is expected to affect most strongly the populations occurring near the species’ environmental tolerance limits (Reusch & Wood 2007). Moreover, combined information on ecology and genetic diversity can add to our understanding of the effects of genetic isolation in a fragmented landscape (Manel et al. 2003; Procaccini et al. 2007).

The aim of this work was to test whether each atoll of P. oceanica is composed of a single genotype. If correct, then those structures are expected to be vulnerable to even small changes of the extreme environmental conditions to which the single genotypes are exposed, and could thus explain the observed regression. Moreover, we aimed to define the level of genetic isolation among stands growing inside, at the entrance and outside the lagoon. We used all 13 microsatellite loci available for the species (Procaccini & Waycott 1998; Alberto et al. 2003). We interpreted the results in respect to growth performances and occurrence of flowering in the last few years, obtained through reconstructive techniques, in an attempt to better understand the regression in atoll size and distribution observed in the last few years and provide support for the management of P. oceanica meadows within this important coastal lagoon.

Material and methods

Study area and sampling design

Six Posidonia oceanica meadows were selected for this study along the western coast of Sicily. Samples were collected in two sites inside the Stagnone di Marsala lagoon (Atolls and Récif) and four sites outside the lagoon (Plateau, Nubia, Marsala and Favignana), where continuous and extensive meadows are present (Calvo et al. 1995) (Fig. 1). The Stagnone di Marsala lagoon is sub-divided into a northern and a southern basin by an emerging P. oceanica reef barrier (Récif-barrière) (Calvo & Fradà Orestano 1984). The southern basin has an average depth of 2 m and is connected to the open sea through an opening 3000 m wide, in which it is possible to observe a vast P. oceanica reef platform (Plateau recifale). The northern basin, where P. oceanica grows in atoll-shaped structures, is characterized by an average depth of 1 m and connected with the open sea northwards through an opening 400 m wide and 20–30 cm deep. This basin shows more distinct lagoon features than the southern one, resulting from the limited water exchange and slow turnover (about 65 days; Calvo et al. 2005). Geomorphological conformation, shallow water, presence of isles and emerging vegetation also hinder water circulation (Mazzola & Sarà 1995; La Loggia et al. 2004). Indeed, higher annual variation in temperature and salinity values were recorded within the lagoon (10–30 °C and 33–48‰, respectively) with respect to the nearby open coastline (temperature: 14.1–26.4 °C; salinity close to 37‰) (Vizzini et al. 2002; Mazzola & Vizzini 2005).

In February 2006, about 40 individual orthotropic P. oceanica shoots (ramets growing vertically) were randomly sampled for each site. Shoots were collected at depths ranging between 1 and 6 m and at a minimum reciprocal distance of 8–10 m, to reduce the risk of sampling within the same clonal patch (Procaccini & Maltagliati 2004). The position of each shoot collected was recorded by GPS with a marginal error of about 1 m. In the northern basin of the lagoon, six to eight individual shoots were sampled from each of six atolls, selected by aerial geo-referenced orthophotos according to their central position in the lagoon and to their shape. The average width of sampled atolls ranged from 17 to 43 m (Fig. 1). From each shoot, rhizomes and bundles were taken for reconstructive and genetic analysis, respectively.

DNA extraction and microsatellite analysis

For each bundle, a leaf segment about 2 cm long was cleaned of epiphytes with a razor blade and stored in silica gel for further DNA extraction. Genomic DNA was extracted adapting the standard protocol of the nucleospin plant kit (Macherey–Nagel) to a new automated procedure. The new methodology involved automated genomic DNA extraction, PCR setup reactions and fragments analysis plates assembly and was developed for the Biomek FX automated platform (Beckman Coulter, Fullerton, CA) equipped with ORCA robotic arm, PCR machine and plates carousel. The Biomek Software allows for user-defined variables to be included in the method. Using our settings, a 96-well plate can be processed in approximately 50 min and up to 12 plates can be processed, unattended, in less than 530 min (see Borra et al. 2007 for details). In general, silica-gel-dried tissue samples were homogenized using a mixer mills and transferred to a 96-well plate, where the first extraction buffer, containing chaotropic salts, denaturing agents and detergents, was immediately added, as indicated in the kit manual. Using a vacuum manifold, working on the whole plate, lysis mixtures were cleared by filtration (300 s at 20 InHg) to remove polysaccharides, contaminations and residual cellular debris. Using a gripper tool and a 96-tip multi-pipetting head, the clear supernatant was mixed with binding buffer and ethanol to create conditions for optimal binding to the silica membrane and loaded on the 96-well binding plate. After washing with two different buffers, DNA was eluted and stored in low salt buffer for downstream applications. PCR reactions were set up, in automation, with 0.5–1 μl DNA; 0.2–0.5 (pmol·μl⁻¹) each oligo; Taq (0.075 u·μl⁻¹) in a total volume of 20 μl. A subset of reactions was randomly selected and gel verified (gel electrophoresis on a 1% agarose gel; TBE 0.5X stained with ethidium bromide). PCR reactions were combined in multiplex and fragment analysis sample plates were assembled in automation, with: 0.4–1.0 μl PCR samples; 0.4 μl size standard and formamide up to a total volume of 20 μl (Migliaccio et al. 2005). Fragments analysis plates were finally analysed on a Beckman coulter CEQ 2000XL DNA analyser.

Individual multilocus genotypes were assessed by means of 13 microsatellite loci, of which 12 were nuclear and one chloroplastic (Poc-trn) (Procaccini & Waycott 1998; Alberto et al. 2003). The analysis of the electrophoretic profiles was carried out with the software BECKMAN CEQ 2000 ver. 3.0. Allelic assignment has been refined in relation to existing datasets (Arnaud-Haond et al. 2007b; Serra et al. 2007). Samples which repeatedly failed to amplify or to give electrophoretic peaks of consistent size were excluded from the analysis. Final statistical analyses were performed on the total of 218 samples.

Statistical analysis

To establish whether ramets with the same genotypes were identical by descent (by clonal growth) or identical by chance, the p_sex value was calculated with the program GENCLONE 1.0 (Arnaud-Haond & Belkhir 2007). Furthermore, we refined our allele scoring to test for the existence of somatic mutations or scoring errors, re-calculating p_sex values for all genotype pairs differing by only one allele after deleting the variable locus (Arnaud-Haond et al. 2007a,b). A threshold value of 0.05, below which identical genotypes were considered truly clonemates, was selected.

The N_g value (Parks & Werth 1993; Ruggiero et al. 2005) was utilized to establish the theoretical number of genotypes according to the given number of loci and alleles. N_g was calculated as:

(1)

where L is the number of loci and a_i is the number of alleles at locus i (Reusch et al. 1998).

Genotypic richness for each meadow has been calculated through the R value (G–1/N–1), where G is the number of the multilocus lineages (MLL) obtained after re-scoring the database for somatic mutations or scoring errors, and N is the number of samples. R is equal to one if all the individuals identified are genetically distinct. Both R and the allelic richness (A) were calculated for the complete dataset and after standardization of sample sizes to N = 16 (equivalent to the smallest number of ramets from the Plateau population) and 1000 random resampling of sample units using GENCLONE 1.0 (Arnaud-Haond & Belkhir 2007). Only complete multilocus genotypes with no missing data for the 13 loci have been utilized for the calculations above.

Genetic variability has also been estimated by means of the allelic frequency for each locus, calculated with software GENALEX ver. 6.0 (Peakall & Smouse 2006), and through the observed (H_obs) and expected heterozygosity (H_exp) calculated using the software GENETIX ver. 4.1 (Belkhir et al. 1996–2002). The Weir & Cockerham (1984) inbreeding coefficient (F_is) was calculated with the software FSTAT ver. 2.9.3.2 (Goudet 1995). Linkage disequilibrium tests for all pairs of loci within each population were performed with the software ARLEQUIN (Excoffier et al. 2005).

The Loiselle spatial autocorrelation index (Loiselle et al. 1995) has been utilized to test for the existence of spatial autocorrelation within sampling sites. The coefficient of determination (r²) and slope (b) were reported. Analysis was performed with the software SPAGeDI (Hardy & Vekemans 2002).

Gene flow among populations was estimated using Wright’s F_st index (1951) and assignment tests. Assignment tests allow the possible assignment of a genotype to a population different from that of origin. This test was performed using a Bayesian method (Rannala & Mountain 1997) and tested over 1000 replicates, utilizing the software GENCLASS ver. 2.0 (Piry et al. 2004).

A CFA (correspondence factorial analysis) based on allele frequency was achieved with the software GENETIX ver. 4.1 (Belkhir et al. 1996–2002) to visualize genetic relationships among populations and individuals.

Results

Genetic variability and structure within sites

A total of 80 alleles were identified for the 13 microsatellite loci utilized. According to the total number of alleles, the theoretical number of genotypes was N_g = 1.12 × 10⁺¹⁴, showing the high polymorphism of the markers utilized. Seventeen genotypes from the total of 218 samples analysed were not complete for one to two loci and were excluded from the assessment of the basic statistics. Of the 201 complete multilocus genotypes, 192 were distinct (Table 1). Within-population pairwise analysis of allele sharing showed that, in the Atolls population, in 20 cases multilocus genotypes (MLG) were different only in one allele. Only one case was recorded among all the other populations analysed (i.e. Récif site). Nevertheless, among the 20 pairs present in the Atolls site, only in one case did the p_sex value suggest a merging of two MLGs in one MLL (Table 1).

Table 1. Parameters of genetic variability. N = number of samples with a complete multilocus genotype, no missing data; MLG = Multi Locus Genotypes; MLL = Multi Locus Lineages; clonal diversity (R) and number of alleles (A) for the entire sample and for the subset of 16 ramets obtained after 1000 permutations (R₁₆ and A₁₆); %pl = percent of polymorphic loci; pA = private alleles; F_is = inbreeding coefficient; H_exp= expected heterozygosity; H_obs = observed heterozygosity.

	N	MLG	MLL	R	R16	% pl	A	pA	A16	Fis	Hexp	Hobs
Atolls	39	32	31	0.79	0.88 ± 0.03	77	36	1	32.28 ± 0.04	−0.44	0.36	0.52
Récif	33	31	31	0.94	0.97 ± 0.02	100	48	4	45.05 ± 0.05	−0.08	0.50	0.55
Plateau	16	16	16	1.00	–	100	48	3	–	−0.02	0.50	0.53
Nubia	38	38	38	1.00	1.00 ± 0.00	100	52	6	42.29 ± 0.06	−0.14	0.45	0.52
Marsala	35	35	35	1.00	1.00 ± 0.00	92	47	5	41.76 ± 0.04	−0.21	0.51	0.62
Favignana	40	40	40	1.00	1.00 ± 0.00	100	46	2	42.02 ± 0.05	−0.15	0.56	0.65

Clonal richness, expressed as R, ranged from 0.79 in the Atolls to 1.0 in all the other populations except the Récif (Table 1). Average clonal diversity for the two Stagnone lagoon sites (R = 0.86; Atolls, Récif) was lower than for open sea sites (R = 1.00).

Expected heterozygosis (H_exp) and observed heterozygosity (H_obs) values ranged, respectively, from 0.36 (Atolls) to 0.56 (Favignana) and from 0.52 (Atolls and Nubia) to 0.65 (Favignana). An excess of heterozygotes was detected in all six sites (H_obs > H_exp). The highest difference was recorded in the Atolls (Table 1). The inbreeding coefficient (F_is) was also lower in the Atolls site (Table 1). Replicated genotypes (clones) did not have higher number of heterozygous loci with respect to genotypes represented by only one individual.

Potential linkage at some loci was present in all populations but the number of such loci was consistently higher in the two Stagnone populations (Atolls and Récif) than in the others.

The lowest number of alleles (A = 36) was also detected in the Atolls (Table 1). Moreover, at the same site, the lowest value of polymorphic loci was recorded (77%) as opposed to an average value of 98.4% in the remaining five sites. Ranking of allelic richness and genotypic richness values did not change after 1000 subsamplings of 16 ramets per population.

There was a greater similarity in the distribution of allelic frequencies between Atolls and Récif than between these two and the sites outside the lagoon (the table of allele frequency is available from the authors upon request). Indeed, the alleles Po5-40.224, Po15.153 and Po15.155 are almost exclusively present in the Atolls and Récif sites. The comparison of the two sites of the Stagnone shows some differences between the genotypes present in the southern sub-basin (Récif) and those in the innermost site (Atolls). In the Atolls, some alleles are fixed (i.e. Poc-trn.316, Poc-35.200 and Poc-26.316). For other alleles, either a negative or a positive trend in frequency can be noticed moving from more to less confined areas inside the lagoon (e.g.Po5.182 and Po5-49.240).

The number of private alleles was low, with the highest value in Nubia (6; Table 1). Surprisingly, the two sites inside the lagoon had fewer private alleles relative to external ones (5 versus 16).

Spatial autocorrelation was significant up to slightly below 300 m in the Atolls and Récif sites (Fig. 2). Autocorrelation in all the sites outside the lagoon was positive up to about 40 m. The coefficient of determination (r²) was higher in the Atolls (r² = 0.569) and in the Récif (r² = 0.198) than in the other populations (Plateau, r² = 0.07; Nubia, r² = 0.018; Marsala, r²=0.037; Favignana, r²=0.009). The slope was the highest in the Atolls (b = −5 × 10⁻⁴, P < 0.001) and the lowest in Favignana (b = −5.6 × 10⁻⁵, P = 0.033). The value was comparable (b = −1 × 10⁻⁴) for the other four populations (Récif, P < 0.001; Plateau, P = 0.019; Nubia, P = 0.009 and Marsala, P < 0.001).

Gene flow among sites

The highest pairwise values of F_st were recorded between Atolls and all the other sites (> 0.1) except for the Récif, in which the value was 0.075 (Table 2). F_st values among all the other sites were always lower than 0.05, showing consistent levels of gene flow.

Table 2. Pairwise F_st values for the six sites considered in the analyses.

	Atolls	Récif	Plateau	Nubia	Marsala	Favignana
Atolls	0.000
Récif	0.075	0.000
Plateau	0.164	0.050	0.000
Nubia	0.105	0.036	0.041	0.000
Marsala	0.181	0.065	0.041	0.060	0.000
Favignana	0.136	0.057	0.048	0.041	0.057	0.000

The correspondence factorial analysis (CFA) showed the distribution of genotypes in a three-dimensional space, according to similarities/differences in allele frequency. The three CFA axes explain 14.18% of total variance. Individual sample-points from the Atolls are tightly clustered and are clearly separated from the other sites (Fig. 3). In general, genotypes external to the lagoon are quite well grouped according to the site they belong to, although the position of sites in the graph is not completely consistent with their geographic distribution. Among all the sites considered in the analysis, Récif has the lowest internal compactness of sample-points. Récif genotypes are arranged in an intermediate position between the innermost area of the Stagnone and the outermost populations.

The isolation of the innermost area of the Stagnone is confirmed by results of assignment analysis, where only the 7.5% of genotypes can be assigned to a different population, and those are all exclusively assigned to the Récif. As for the Récif, 2.7% of genotypes can be assigned to the Atolls and 16.2% to the Plateau, whilst the Plateau preferentially exchange with the other three sites outside the lagoon.

Growth and flowering performance analyses

A total of 243 orthotropic shoots, for a total of 2039 annual rhizome segments, were analysed for the lepidochronological approach. This reconstructive method indicated that shoot age had a grand mean of 8.1 years and revealed a clear heterogeneity across the sites varying from 5.2 years at Marsala to 13.8 years at Atolls (Table 3). Shoot age was taken into account as a covariate before testing the effect of sites on rhizome elongation.

Table 3. Mean age distribution of shoots sampled in different sites.

site	age (years)
	minimum	maximum	mean
Atolls	2	28	13.8
Récif	1	20	8.2
Plateau	1	20	8.0
Nubia	4	13	6.6
Marsala	1	23	5.2
Favignana	1	21	6.5
total	1	28	8.1

Results of the statistical model indicated that age had a negative effect on rhizome elongation. Indeed, the estimate associated with this covariate is −0.3 ± 0.1 mm·year⁻¹ (P < 0.001). The baseline (Favignana) estimated for vertical growth was 8.6 ± 0.7 mm·year⁻¹ (P < 0.001) (Fig. 4A). Only Atolls showed significant statistical differences from Favignana, with a difference of 2.7 ± 1.0 mm·year⁻¹. As for the annual leaf formation, the linear model detected no significant differences among sites except for Atolls, where a difference of 1.2 ± 0.3 mm·year⁻¹ was observed (Fig. 4B).

Remains of floral stalks have been found on 68 of 9777 annual rhizome segments analysed from 1950 to 2005. No flowering was detected in the Atolls site. In all the other sites examined, flowering was most prevalent in 1987 (50%), 1991 (67%), 1996 (56%) and 2003 (50%). The global mean meadow flowering frequency (FF) was 0.196 ± 0.036, with the highest value at Nubia (0.282 ± 0.082) (Table 4). The global mean flowering probability (Pf) and intensity (FI) were respectively 0.0072 ± 0.0015 inflorescences · shoot⁻¹· year⁻¹ and 0.026 ± 0.007 inflorescences · shoot⁻¹. The Favignana site exhibited the highest value of both indexes, reaching 0.0108 ± 0.0016 inflorescence · shoot⁻¹· year⁻¹ and 0.032 ± 0.015 inflorescence · shoot⁻¹.

Table 4. Average frequency of meadow (FF, flowering years per year), shoot flowering probability (Pf, inflorescence · shoot⁻¹· year⁻¹), and flowering intensity (inflorescences · shoot⁻¹, calculated from meadows FI means) for the six sampling sites.

site	number of shoots		number of total shoots	time horizon (FF and Pf)	FF ± SE (nstation)	Pf (nstation)	Time horizon (FI)	FI ± SE (nstation)
	present study	previous study
Atolls	40	99	139	84–04	0.00 (ns = 1)	0.0000 (ns = 1)	92–00	0.00 (ns = 1)
Récif	40	53	93	89–04	0.188 (ns = 1)	0.0051 (ns = 1)	94–99	0.017 ± 0.002 (ns = 1)
Plateau	38	11	49	92–04	0.077 (ns = 1)	0.0022 (ns = 1)	–	–
Nubia	42	202	244	92–04	0.282 ± 0.082 (ns = 2)	0.0087 ± 0.0008 (ns = 2)	95–99	0.028 ± 0.015 (ns = 2)
Marsala	41	167	208	88–04	0.249 ± 0.018 (ns = 2)	0.0092 ± 0.0041 (ns = 2)	94–99	0.013 ± 0.005 (ns = 2)
Favignana	42	234	276	80–04	0.217 ± 0.063 (ns = 2)	0.0108 ± 0.0016 (ns = 2)	85–99	0.032 ± 0.015 (ns = 2)
total	243	766	1009	80–04	0.196 ± 0.036 (ns = 9)	0.0072 ± 0.0015 (ns = 9)	85–00	0.026 ± 0.007 (ns = 8)

Discussion

Results clearly show the peculiarity of the Posidonia oceanica atoll-shaped meadow in the innermost area of the Stagnone di Marsala coastal lagoon. Each atoll is composed of multiple genotypes, which falsifies the hypothesis of their monoclonal structure. Nonetheless, the atolls show characteristics of genetic isolation, namely the existence of alleles and genotypes which are distinct from those found outside the lagoon. This situation could generate inbreeding among the confined P. oceanica shoots with genetic drift and possible selection of adapted genotypes in the northern basin of the lagoon. The lower growth rate in the lagoon as inferred from lepidochronological data and the lack of flowering are indicative of persistent stressful conditions.

The Atolls population exhibits a lower number of alleles, lower percent of polymorphic loci, lower clonal diversity and higher heterozygosity excess relative to the other sites included in the present analysis. Different reasons can account for the high heterozygosity. First, in the literature high heterozygosity has been positively correlated with fitness, particularly when organisms live in unfavorable environmental conditions. This has been demonstrated experimentally in mice for functional traits represented by allozymes (Teska et al. 1990). In theory it should not apply if the markers utilized are truly neutral, which is generally assumed for microsatellites (Hansson & Westerberg 2002). Nevertheless, it has been demonstrated that microsatellites have a range of different functions within the genome (Li et al. 2002) and we cannot exclude the possibility that some loci utilized here show patterns not consistent with neutrality (see Oetjen & Reusch 2007 for Zostera marina). Secondly, excess of heterozygosity in neutral markers can also derive from the combined effect of linkage disequilibrium and partial inbreeding (Hansson & Westerberg 2002), a scenario which could easily apply to the Stagnone populations. Nevertheless, our results do not completely support this latter option. All populations analysed here, in fact, have a negative F_is value, but the value is consistently lower in the Atolls population than in the others, showing lower inbreeding in the more confined locality. The highest number of loci showing potential linkage, instead, is present at the Atolls site.

In general, species inhabiting temporally variable or spatially heterogeneous environments exhibited higher levels of genetic variability than those from less variable or more monotonous environments (Valentine 1976). In our study, this only true for heterozygosity, which is highest in the Atolls population, whereas the other genetic diversity estimators show a different trend. Genotypic diversity (R = 0.79 for Atolls), although lower than in the other populations analysed here, is still consistently higher than the average values recorded in the whole distributional range of the species. Without considering the Adriatic Sea, where monoclonal stands are present (Procaccini et al. 2002; Ruggiero et al. 2002; Arnaud-Haond et al. 2007b), the average R value calculated for 32 meadows within the whole Mediterranean basin is, in fact, below 0.64 (Arnaud-Haond et al. 2007b).

The results of the F_st analysis showed active gene flow among all the areas, excluding the innermost section of the lagoon, stressing the higher isolation of the Atolls from all the other sites except the Récif. This latter area seems to represent the only possible source of new genotypes for the innermost section of the lagoon, as resulting from the position of the Récif genotypes in the CFA. Indeed, all the Atolls genotypes assigned to different populations are exclusively assigned to the Récif. Results from the spatial autocorrelation analysis showed that the autocorrelation curve in the two sites within the lagoon is significant up to almost 300 m, which represents the putative maximum distance for local gene flow. The local gene flow decreases at this limit, suggesting that more distant atolls show lower genetic affinity. This value is higher than what is normally found in open-sea meadows (Migliaccio et al. 2005; Procaccini et al. 2007). The high correlation coefficient recorded in these two sites confirms that the enclosed system of the Stagnone lagoon is genetically isolated, with efficient water circulation and dispersal only at a local scale. In the open sea the more frequent immigration of genotypes from other meadows can reduce the correlation among spatially close genotypes.

Genetic isolation is likely to occur in populations living in confined environments (Cognetti & Maltagliati 2000). In seagrasses, studies carried out on Zostera marina (Muñiz-Salazar et al. 2006) have shown high genetic isolation of populations living in lagoons relative to the open sea. Long-term isolation of populations can enhance selection of locally adapted genotypes, whose existence can be detected with ad hoc genetic markers (see Vasemägi et al. 2005). The degree to which this will occur depends on the level of isolation and on the extremes of environmental conditions experienced (Hoffmann & Parsons 1997).

Biometric data indicate low variability in P. oceanica growth performance among plants growing in different areas. Only one significant deviation from the general intercept was detected in the linear model. In particular, Atolls plants exhibited a lower vertical growth and leaf formation compared to Favignana (baseline of Linear Model), whereas no significant variation was observed at the Récif site. Retro-dating analyses to estimate P. oceanica growth performance inside the lagoon were carried out in a previous study (La Loggia et al. 2004), where a reduction in vertical growth was detected in the presence of lower water exchange. This reduction is consistent with our study suggesting that growth performances may be sensitive to site specificities, depending in particular on local hydrodynamic regimes. In the northern and most enclosed area of the Stagnone, water exchange is twofold lower than in the Récif, where the maximum within-lagoon water exchange rate has been estimated (La Loggia et al. 2004). During the summer season, low water exchange determines maximum salinity and temperature values of 48‰ and 30 °C, respectively (Mazzola & Vizzini 2005), which are above the tolerance range reported for the species (Boudouresque & Meinesz 1982; Gobert et al. 2006). As observed for species growing in locations with temperatures above the optimum for growth or near the upper limit of thermal tolerance, a further increase in temperature could determine a decline in their productivity and distribution (Short & Neckles 1999).

The results of this study provide evidence that sexual reproduction is negatively affected by stressful conditions inside the lagoon. In the Atolls area no sexual reproductive events were recorded during 24 years of study, although in the Western Sicily area, flowering events occurred almost annually with a flowering index magnitude generally greater than the average in the Western Mediterranean basin (Diaz-Almela et al. 2006, 2007). Absence of flowering has also been observed in seagrass populations living at the geographical range limits of the species, where environmental conditions become extreme (Ashton & Mitchell 1989; Eriksson 1996; Reusch et al. 1999; Alberto et al. 2001; Billingham et al. 2003). Our observation is apparently in contrast to the general trend observed by Diaz-Almela et al. (2007), which indicates a positive correlation between P. oceanica flowering and temperature anomalies. We can hypothesize that the water temperature experienced by the plant in the Stagnone lagoon is above the threshold limit for flowering induction in the species. The water temperature of 30 °C recorded in summer in our study site, in fact, is consistently higher than values observed during temperature anomalies in the whole Mediterranean. As an example, the strongest temperature anomaly of the last 30 years, which corresponded to a clear increase in flowering, was recorded in 2003, when sea-surface temperature (SST) increased by about 2 °C in the Western Mediterranean, reaching a value of about 28 °C (Diaz-Almela et al. 2007). The very high salinity, characteristic of the Stagnone inner waters, surely plays an additional role in determining stressful environmental conditions.

How can the high genetic and genotypic diversity found within the lagoon be explained considering the genetic isolation and the absence of recent reproductive events in the innermost area of the lagoon? First, the genetic diversity of P. oceanica meadows is generally high in the whole geographic area where the Stagnone is located (i.e. Southern Tyrrhenian Sea and Sicily Strait). Values recorded in our study are comparable to previous results from meadows along the southern Sicilian and Tunisian coasts (Arnaud-Haond et al. 2007b), where flowering and seeds washed ashore have been observed more frequently than in the rest of the basin (Calvo et al. 2006). The presence of multiple genotypes in each of the atolls and the relatively high genetic diversity in the Atolls area could result from a high ancestral diversity in the lagoon. Influx of genotypes into the lagoon, which may have been frequent in the past, has diminished to almost nothing in recent times, due to an almost complete closure of the northern entrance (Agnesi et al. 1993). This hypothesis is also supported by recent findings of banquette formed by fruits of P. oceanica, which were probably able to enter the lagoon in the past from outside its northern opening (Calvo et al. 1995). Reduction in gene flow between the lagoon and the open sea may have led to genetic drift. The harsher environmental conditions may have also favored the selection of a reduced number of genotypes which were able to cope with the extreme temperature and salinity conditions of this area. The excess of heterozygosity present in this area could support the existence of selection, through the advantage of heterozygous genotypes, although replicated genotypes did not have a larger number of heterozygous loci than the other genotypes.

In light of these results, we can conclude that the P. oceanica atoll-like structures within the Stagnone di Marsala lagoon are composed of multiple genotypes. Although the worst scenario would have been if atolls were composed of a single genotype, the genetic isolation, lack of flowering and low growth performances are a cause for concern, threatening the survival of the species in this area. The P. oceanica population living in the northern basin of the lagoon under extreme environmental conditions could serve as an experimental model for what might happen under predicted environmental changes in the whole of the Mediterranean as a result of global warming.