Maintenance of genetic differentiation across a transition zone in the sea: discordance between nuclear and cytoplasmic markers

Authors

  • C. Lemaire,

    1. Laboratoire Génome, Populations, Interactions, Adaptation, UMR 5171 CNRS-IFREMER-Université Montpellier II, Station Méditerranéenne de l'Environnement Littoral, Sète, France
    Search for more papers by this author
  • J.-J. Versini,

    1. Laboratoire Génome, Populations, Interactions, Adaptation, UMR 5171 CNRS-IFREMER-Université Montpellier II, Station Méditerranéenne de l'Environnement Littoral, Sète, France
    Search for more papers by this author
  • F. Bonhomme

    1. Laboratoire Génome, Populations, Interactions, Adaptation, UMR 5171 CNRS-IFREMER-Université Montpellier II, Station Méditerranéenne de l'Environnement Littoral, Sète, France
    Search for more papers by this author

François Bonhomme, Laboratoire Génome, Populations, Interactions, Adaptation, UMR 5171 CNRS-IFREMER-Université Montpellier II, Station Méditerranéenne de l'Environnement Littoral, 1 Quai de la daurade, 34200 Séte, France.
Tel.: (+33) 4 67 46 33 72; fax: (+33) 4 67 46 33 99;
e-mail: bonhomme@univ-montp2.fr

Abstract

To investigate the origin and maintenance of the genetic discontinuity between Atlantic and Mediterranean populations of the common sea bass (Dicentrarchus labrax) we analysed the genetic variation at a fragment of mitochondrial cytochrome b sequence for 18 population samples. The result were also compared with new or previously published microsatellite data. Seven mitochondrial haplotypes and an average nucleotidic divergence of 0.02 between Atlantic and Mediterranean populations that matches a Pleistocene allopatric isolation were found. The frequency variation at the cytochrome b locus was many times greater between Atlantic and Mediterranean populations (inline image = 0.67) than at microsatellite loci (inline image = 0.02). The examination of the different evolutionary forces at play suggests that a sex-biased hybrid breakdown is a likely explanation for part of the observed discrepancy between mitochondrial and nuclear loci. In addition, an analysis is made of the correlation between microsatellite loci points towards the possible existence of a hybrid zone in samples from the Alboran Sea.

Introduction

Understanding the origin and maintenance of genetic discontinuities remains a central point in evolutionary biology, as exemplified by recent publications that examine the classical dilemma of allopatric vs. sympatric speciation (Orr & Smith, 1998; Dieckmann & Doebeli, 1999; Kirkpatrick, 2000; Schluter, 2001). The question is especially relevant in the sea, where effective physical barriers are difficult to find (Avise, 1994; Palumbi, 1994). Basic information enabling us to identify genetic discontinuities is far less abundant for marine than for terrestrial species (e.g. Bermingham et al., 1997; Bowen & Grant, 1997; Lessios et al., 1998; Arnaud et al., 2000; Avise, 2000; Perrin & Borsa, 2001). Regardless of the possible existence of sympatric or nearly sympatric (microallopatric) cases of genetic divergence (Carlon & Budd, 2002), current or ancient physical barriers can also play an important role in promoting reproductive isolation in the marine environment, leading in some instances to allopatric speciation (e.g. Knowlton et al., 1993).

Pleistocene glacial episodes involve cyclical variation in sea level and surface temperatures which have significantly influenced the physical connection between water masses and populations of their inhabiting species (Avise, 2000). This is also true around the Strait of Gibraltar where cold episodes are thought to have pushed populations of temperate species southwards along the African coast in the Atlantic Ocean and south-eastwards in the Mediterranean Sea off the coasts of Libya and Egypt. Thus, many species that inhabit both sides of the Strait of Gibraltar are expected to have experienced cyclical fragmentation and isolation events that could have allowed genetic differentiation (e.g. Borsa et al., 1997). It is unclear whether evidence of these isolation events remains or whether their effects vanish at each interglacial warm episode because of subsequent gene flow.

In the common (or European) sea bass [Dicentrarchus labrax L. (1758), Teleost; Perciformes, Moronidae] evidence of past fragmentation is still detectable. This fish is found in coastal waters of the Atlantic Ocean from Scandinavia to Western Sahara and throughout the Mediterranean Sea. Previous genetic studies based on allozymes and mitochondrial markers have led to the identification of three genetically distinct zones: the north-eastern Atlantic Ocean, the western Mediterranean and the eastern Mediterranean (Benharrat et al., 1983; Patarnello et al., 1993; Allegrucci et al., 1997; Caccone et al., 1997; Cesaroni et al., 1997). On the basis of six microsatellite loci the transition zones have been localized between the Atlantic (plus the Alboran Sea) and the western Mediterranean at the Almeria-Oran oceanic front (Naciri et al., 1999) and between eastern and western Mediterranean somewhere around the Siculo-Tunisian strait (Bahri-Sfar et al., 2000).

Here we focus on the study of the first transition zone. Understanding the maintenance of such transition zones is particularly relevant to discussions about speciation mechanisms in marine taxa. For example, D. labrax has great dispersal potential because of a planktonic larval phase of 2–3 months (Pickett & Pawson, 1994) and migratory behaviour of adults (Barnabé, 1980; Pickett & Pawson, 1994), both of which could lead to movements over several hundreds of kilometres (Pickett & Pawson, 1994). Nevertheless, a clear transition zone appears to exist in the eastern part of the Alboran Sea (Naciri et al., 1999). The equilibrium gene flow (Nem) between the two basins has been estimated to be the result of only approximately 11 effective migrants per generation (Naciri et al., 1999; but see also Whitlock & McCauley, 1999 for limitations of translation of FST into Nem), whereas the same study also revealed fairly homogeneous populations on either side of the transition zone (inline image = 0.010, P < 0.001 in the Atlantic and inline image = 0.002, P > 0.05 in the western Mediterranean). These results support the hypothesis that the physical barrier has not been sufficient to prevent gene flow between Atlantic and Mediterranean individuals. Thus, while a physical barrier may well be the source of this genetic discontinuity, a genetically determined barrier is invoked to explain its persistence.

A genetic barrier may result from a number of mechanisms, including post-zygotic isolation linked to hybrid breakdown (endogenous factors) or disruptive environmental selection (exogenous factors), prezygotic isolation mechanisms such as philopatry (Fitzsimmons et al., 1997), assortative mating/sexual selection (Nagel & Schluter, 1998) or habitat choice (Behegaray & Levy, 2000; Bierne et al., 2002). These mechanisms may all contribute to speciation (Kirkpatrick & Ravigné, 2002). In the presence of local adaptations, parapatric speciation can occur between adjacent populations if such behavioural processes and genetic drift are sufficiently strong to counteract the homogenizing effects of gene flow (e.g. Gavrilets et al., 2000), and thus, the effective dispersal behaviour of individuals is a key parameter.

While the whole genome should roughly be equally affected by a physical barrier, genomic heterogeneity may arise because of semi-permeable genetic barriers to gene flow caused by differing selective pressures (Barton & Hewitt, 1989; Harrison, 1990; Rieseberg et al., 1999). Because the mitochondrial genome is not physically linked to the nuclear one it may exhibit different dynamics, crossing the genetic barriers more rapidly and being influenced by sex-biased isolation mechanisms (e.g. sex-biased hybrid breakdown or even philopatry), or it may itself be involved in some kind of reproductive isolation (Clark & Lyckegaard, 1988; Burton et al., 1999). It has thus been important to compare mitochondrial genome differentiation with that of the nuclear genome (Harrison, 1990).

We compare nuclear microsatellite and mitochondrial DNA data in D. labrax to investigate potential presence of a genetic barrier between Mediterranean and Atlantic populations, as well as to shed light on its maintenance.

Materials and methods

Sampling

Eighteen sampling sites were chosen around the Almeria-Oran front: seven samples were located in Atlantic Ocean, three in Alboran Sea and eight in western Mediterranean. Average within population sample size was 33 ± 6.35 individuals with a maximum of 63 and a minimum of nine individuals (Fig. 1). Most DNA samples were the same as those surveyed in previous microsatellite studies (Bahri-Sfar et al., 2000: TISK, TBIZ, TGOU; García de León et al., 1997: FSET, FMRS; Lemaire et al., 2000: FPRE; Naciri et al., 1999: BANV, FBRE, PAVR, MTN1, MTH1, MRBT, MKS1, MMAR, MNAD, AGLA). The two remaining population samples (FYEU and EMAM) were specific to this study.

Figure 1.

Geographic location of the 18 samples of D. labrax, numbers between brackets indicate sample size.

Molecular analyses

The 18 samples were analysed using a 304 bp fragment of the mitochondrial cytochrome b gene. Specific primers (CBF1: 5′-CGGTTCGCTCTTAGGCCTATGC-3′ and CBR1: 5′-GGGCAACACATAGCCTACGAAGGC-3′) were designed for D. labrax from the sequence of cytochrome b available in Genbank (accession number: AF143189). DNA amplifications were performed by polymerase chain reaction (PCR) in microtitre plates, with each reaction containing 3 μL DNA solution (10–20 ng), 10 pmol of each primer, 0.2 mm each dNTP, 2.5 mm MgCl2, 1X polymerase buffer and 0.5 U Taq polymerase (25 μL final volume). Each well was overlaid with 20 μL mineral oil, and PCR was carried out in a Crocodile III thermocycler. Discrimination of haplotypes was performed by Single Strand Conformation Polymorphism (SSCP; Orita et al., 1989) on a vertical nondenaturing 12% acrylamide gel (ratio acryl : bisacryl 37 : 1; BDH, Poole, UK) with 1X TBE buffer. DNA bands were visualized by ethidium bromide staining under a UV transilluminator.

Genotypes at six microsatellite loci (García de León et al., 1995) were available for most of the samples (García de León et al., 1997; Naciri et al., 1999; Bahri-Sfar et al., 2000) and two new samples, FYEU (Yeu Island, Gulf of Biscay) and EMAM (lagoon of Mar Menor, Western Mediterranean) were typed for these loci following the protocol described in Naciri et al. (1999).

DNA sequencing

For each mtDNA haplotype identified, the 304 bp PCR products were sequenced directly for two individuals per haplotype by the method of Sanger et al. (1977) using primer CBF1 dye-labelled with fluorescing CY5. Sequences were performed with the ‘Thermo Sequenase Fluorescent Primer Cycle Sequencing’ kit with 7-deaza-dGTP (USB Corporation, Cleveland, OH, USA) and analysed with the automatic A.L.F. DNA Sequencer (Pharmacia Biotech, Uppsala, Sweden).

Sequence alignment

All the sequences were automatically aligned and manually corrected with the alignment interface of the automatic sequencer: Alfwin Sequence Analyser Module v2. 10. 06, 1998 (Amersham Pharmacia Biotech). Alignments were manually performed with the BioEdit software (Hall, 1999). Primer sequences were removed from all of the haplotypes, analyses were conducted on the remaining 288 bp.

Statistical analyses

Allele and haplotype frequencies as well as population genetic parameters were estimated with the Genetix v4.02 software (Belkhir et al., 1996–2002). Gene and haplotype diversity (Henb, Nei, 1987) were estimated respectively on microsatellite and cytochrome b data for each sample with correction for haplotype data [(2N − 1)/(2N − 2); N = sample size]. Confidence intervals for gene diversities were estimated among samples of each basin (Mediterranean and Atlantic/Alboran). Wright's fixation indices FIS and FST were estimated respectively by inline image and inline image according to Weir & Cockerham (1984). Estimates of f were only computed on microsatellite data. All tests (f or θ = 0) were performed by permutation of alleles or individuals respectively. inline image and inline image will be used to denote cytoplasmic (mitochondrial) and nuclear (microsatellites) θ estimates respectively. The sequential Bonferroni method (Rice, 1989) was used for multiple tests. All distance trees were inferred from Reynolds et al. (1983) coancestry genetic distance [D = 1 −  ln (1 − θ)] matrix by the Neighbor procedure of the phylogenetic package Phylip 3.2 (Felsenstein, 1993).

For each sample, nucleotidic diversities π (Tajima, 1983; Nei, 1987) were estimated for mitochondrial haplotypic data. Phylogenetic relations between sea bass mitochondrial haplotypes were also investigated. The average transition/transversion ratio among pairwise comparisons of sequences and nucleotidic distances were estimated using the Kimura 2-parameters model (Kimura, 1980) implemented in the MEGA 2.1 software (Kumar et al., 2001). The different D. labrax haplotypes were compared with a sequence of the congeneric species D. punctatus, the spotted sea bass (Genbank accession number: AF240740). A network of phylogenetic relationships between haplotypes was reconstructed by the split decomposition method implemented in SplitsTree 3.2 (Huson, 1998; see also Posada & Crandall, 2001).

Joint estimations of gene flow and divergence time on cytochrome b between the Atlantic-Alboran and the Mediterranean groups were carried out with the Bayesian inference method of Nielsen & Wakeley (2001) implemented in the program mdiv. Using a Markov Chain Monte Carlo simulation of the coalescence process, this program obtains posterior estimates of the classical parameters of diversity (θ = 2Nefμ), migration (M = 2Nefm) and divergence time between the two populations (T = t/2Ne) with Nef being the female haploid effective size, μ the mutation rate, m the female migration rate and t the number of generations. Initially, the method was tested for different values of M (0; 2 and 5) and T (0; 5 and 10). After this first step, both parameters were adjusted (M = 5 and T = 10) and the process was repeated five times with a different seed number for checking the stability of estimates. For each replicate the settings of the Markov Chain were fixed at 5 × 106 steps and a burning time of 106 steps.

We also performed a neutrality test using molecular variation on the cytochrome b gene. The Tajima's neutrality test (Tajima, 1989) is based on haplotype sequence information and compares two estimators of θ (2Nefμ; Nef being the female haploid effective size and μ the mutation rate) under an infinite-site model without recombination, which is appropriate for mitochondrial genome. It detects a possible departure from mutation-drift equilibrium, which could be caused either by selection or changes in effective population size like bottlenecks. This test was carried out on the haplotypes of cytochrome b using the Arlequin software (Schneider et al., 2000).

Results

Gene and haplotype diversities

The analyses of the 304 bp fragment of cytochrome b by SSCP on 594 individuals revealed seven haplotypes denoted A to G. Sequencing of two individuals sharing each haplotype revealed no sequence heterogeneity within haplotypes. We found no heteroplasmy in any individual, a result concordant with the data of Patarnello et al. (1993) on the cytochrome b and contrary to Cesaroni et al. (1997) on the D-loop (Table 1). Cytochrome b data gave haplotype diversities ranging from 0.00 (FSET, one haplotype fixed) to 0.69 (MNAD). At the nuclear loci, estimates of multilocus Henb ranged from 0.84 (TGOU and TISK) to 0.91 (PAVR). As previously observed by Naciri et al. (1999), gene diversities of the samples from Atlantic-Alboran region (Henb = 0.90 ± 0.01) were significantly greater than those of Mediterranean (Henb = 0.86 ± 0.01; Mann–Whitney U-test, n = 18, P < 0.0001). This difference was confirmed at the mitochondrial level where haplotypic diversities were four times greater in Atlantic and Alboran (Henb = 0.42 ± 0.105) than in Mediterranean samples (Henb = 0.10 ± 0.07; Mann–Whitney U-test, n = 18, P < 0.0001). Furthermore, estimates of average nucleotidic diversity π were almost 10-fold greater in Atlantic-Alboran samples (0.00550 ± 1.0 × 10−3) than in Mediterranean ones (0.00067 ± 6.0 × 10−4; Mann–Whitney U-test, n = 18, P < 0.0001).

Table 1.  Mitochondrial haplotype frequencies and diversities (Henb: haplotypic diversity; π: nucleotidic diversity). Mediterranean samples are in italic.
SamplesNHaplotypesHenbπ
ABCDEFG
BANV410.93 0.020.02 0.02 0.142.0 × 10−3
FBRE380.790.16 0.05   0.364.0 × 10−3
FYEU220.770.090.050.09   0.405.0 × 10−3
PAVR480.790.08 0.020.02 0.080.375.0 × 10−3
MTAN420.690.260.020.02   0.466.0 × 10−3
MTAH310.680.26 0.06   0.496.0 × 10−3
MRBT350.710.26 0.03   0.445.0 × 10−3
MKSR430.790.140.020.020.02  0.365.0 × 10−3
MMAR140.570.43     0.537.0 × 10−3
MNAD90.220.56  0.11 0.110.691.0 × 10−2
EMAM32 0.97  0.03  0.062.0 × 10−4
FPRE20 0.95  0.05  0.104.0 × 10−4
FSET14 1.00     
FMRS60 0.95  0.05  0.104.0 × 10−4
AGLA510.020.98     0.045.0 × 10−4
TISK38 0.92  0.08  0.156.0 × 10−4
TBIZ290.140.83  0.03  0.313.0 × 10−3
TGOU27 0.96  0.04  0.073.0 × 10−4
Atlantic & Alboran3230.690.220.010.030.02 0.020.425.5 × 10−3
Western Mediterranean2710.020.95  0.04  0.106.7 × 10−4

Because of the presence of null alleles at the locus Labrax-6 (see García de León et al., 1997), estimates of multilocus f are given with and without this locus. The two new samples EMAM and FYEU do not demonstrate significant departures from Hardy-Weinberg equilibrium. Excluding the Labrax-6 locus, Western Mediterranean samples seemed to be close to panmixia (inline image = 0.005, P > 0.05) while this was not the case for the Atlantic-Alboran samples (inline image = 0.039, P < 0.001), which has already been noted by Naciri et al. (1999).

Population differentiation

Figure 2 describes the variation of cytochrome b haplotype frequencies for different samples, characterized by a replacement of the main haplotype (A) in Atlantic samples by the B haplotype in Mediterranean samples. These latter samples comprised only two haplotypes (B and E), except in AGLA and TBIZ samples where haplotype A was also observed. The subsequent global inline image is 0.53 (P < 0.001). The new sample of Mar Menor (EMAM), close to the Almeria-Oran front was significantly differentiated from all other samples at nuclear loci (inline image = 0.023, P < 0.001 against Mediterranean samples and inline image = 0.021, P < 0.001 against samples from Atlantic-Alboran), whereas it was genetically similar to Mediterranean samples at the cytochrome b locus, but still different from Atlantic-Alboran's samples for this marker (inline image = 0.613, P < 0.001). The tree in Fig. 3 inferred from microsatellite data illustrates that the newly analysed Mar Menor (EMAM) sample falls between the Mediterranean and Atlantic clades, despite a bootstrap value of only 77% with the former. The new sample (FYEU) was clearly assigned as Atlantic at both nuclear and cytoplasmic loci.

Figure 2.

Mitochondrial haplotype frequencies in the 18 samples of Dicentrarchus labrax. Rare haplotypes present in Atlantic samples were combined under the ‘Atlantic pool’.

Figure 3.

Neighbour-joining tree inferred from Reynolds’ genetic distance at six microsatellite loci in 18 Dicentrarchus labrax samples. Only bootstrap scores >50% are indicated.

Genetic differentiation at the mitochondrial locus appeared to be several-fold greater than that observed at microsatellite loci. The overall θ estimate was more than 20-fold greater at cytochrome b (inline image = 0.53, P < 0.001) than at microsatellite loci (inline image = 0.02, P < 0.001). Intra-basin estimates were inline image = 0.023 (P > 0.05) vs. inline image = 0.004 (0.01 < P < 0.05) for western Mediterranean samples (excluding the Mar Menor sample), and inline image = 0.061 (P < 0.001) vs. inline image = 0.008, P < 0.001 for Atlantic-Alboran samples. The most important difference between cytoplasmic and nuclear differentiation occurs at the inter-basin level (Mediterranean Sea without EMAM vs. Atlantic-Alboran) where inline image = 0.676 (P < 0.001) was more than 30-fold larger than inline image = 0.023 (P < 0.001).

Phylogeny of haplotypes

Kimura-2-parameters distances (Kimura, 1980) were estimated with a mean transition/tranversion rate of 2.6 (Table 2). These estimates ranged from 0.004 (B and E) and 0.039 (C and F), corresponding to one and nine mutations steps respectively. The mean distance between D. labrax and D. punctatus was 0.133 ± 0.006. The split decomposition network of the Atlantic-Alboran and Mediterranean haplotypes demonstrated that, with the exception of haplotype C being close to E, the haplotypes clustered according to their main area of occurrence (Fig. 4).

Table 2.  Estimates of nucleotidic divergence on a 238 bp of the mitochondrial cytochrome b of D. labrax. Above diagonal: distances of Kimura 2 parameters; below diagonal: number of pairwise changes.
 ABCDEFGD. punctatus
A0.0130.0210.0080.0170.0170.0130.121
B30.0080.0210.0040.030.0260.137
C520.030.0040.0390.0350.148
D2570.0260.0260.0210.131
E41160.0350.030.143
F479680.0170.131
G3685740.131
D. punctatus26293128302828
Figure 4.

Split decomposition network of Dicentrarchus labrax rooted with haplotype of D. punctatus. Haplotypes typical of Mediterranean Sea are indicated in bold italic.

Estimates of mitochondrial gene flow and time of divergence between individuals from the Atlantic-Alboran and the Mediterranean basins are also given by the mdiv program (Nielsen & Wakeley, 2001). According to this method, the global genetic diversity estimate was inline image = 0.785 and the migration rate estimate was inline image = 0.384, suggesting that on average one female is exchanged between the two basins every three generations. Estimates of T were more difficult to assess because the distribution of T reached a plateau at T = 3.5 and did not decrease for higher values (not shown). Yet, we could consider that the value of T = 3.5 might represented the minimum divergence time.

No departure from mutation/drift equilibrium was observed except at the BANV sample (Tajima's D = −2.39, P < 0.01; data not shown). Nevertheless, after a Bonferroni correction for multiple tests (Rice, 1989) the departure did not remain significant.

Discussion

Natural populations of the European sea bass have been the subject of many studies using several types of markers, although few studies have focused on transition zones. Genetic variation at the mitochondrial cytochrome b was analysed and compared to previous published data on microsatellite loci (Naciri et al., 1999). In the following, we discuss the possible phylogeographic origin of D. labrax present-day populations, exploring the cause(s) of the strongly contrasted levels of mitochondrial and microsatellite θs observed at the inter-basin level. Furthermore, we discuss of the implication of these findings for the maintenance of the genetic transition zone between Atlantic-Alboran and Mediterranean populations.

Phylogeography of D. labrax

The mtDNA variation in sea bass is consistent with an ancient split between the Atlantic and Mediterranean populations. When looking at the network in Fig. 4, it is apparent that the most frequent Mediterranean haplotype (B) and the most frequent Atlantic one (A) are distinct. The overall mean of Nei & Li's (1979) nucleotide divergence, which takes into account ancestral polymorphisms, is 0.018 ± 0.001 between the Atlantic-Alboran and Mediterranean basins. This translates to a divergence time of 0.9–3.6 My assuming a divergence rate of 0.5–2% per My (see Brown et al., 1979 for a standard reference in fish). The estimate of T given by mdiv (Nielsen & Wakeley, 2001) implies a divergence beginning at least at t = 3.5 × 2Nef. Assuming divergence rates indicated above, the mutation rate per lineage is estimated as 0.25–1.10−8 mutations per nucleotide per year. Assuming a generation time of three years and a 304 bp long fragment, we expect a haplotype mutation rate between μ = 0.25 × 10−8 × 3 × 304 ≈ 2.5 × 10−6 and 10−5 mutations per generation for our cytochrome b fragment. Thus, considering the estimate of θ = 2Nefμ = 0.785 and the time of divergence in number of generation t = T × 2Nef, the divergence time can be expressed as t = θT/μ. Thus, the last divergence event may have occurred between 0.27 and 1.1 My. These divergence times are consistent with the hypothesis of the last isolation episode occurring during the Pleistocene glaciation cycles (1.8–0.01 My ago). These results should be interpreted cautiously because the underlying model in mdiv assumes a constant effective size since divergence and equal effective sizes in the two populations. Nevertheless, Tajima's test does not detect any recent significant change in population effective sizes (for example a bottleneck followed by a demographic expansion) although our power to detect this is limited given the low amount of nucleotide polymorphism in each population.

Rooting of the network by the outgroup D. punctatus suggests that the common Mediterranean haplotypes derive from a larger Atlantic coalescent, which seems more diversified. This fits well with the observed higher haplotype and nucleotide diversities for mtDNA and the higher expected heterozygosity for microsatellites we found for the Atlantic samples. This, together with the lack of reciprocal monophyly, points towards a probable Atlantic origin of Mediterranean sea bass. However, the contrary scenario cannot be completely excluded, especially if the effective sizes of populations have been very different. Nevertheless, the presence of haplotypes B and C in the Atlantic, and the haplotype A in the Mediterranean suggests that rare exchanges may occur from time to time. Assuming a simple model of divergence, with the possible occurrence of migration between two populations (Nielsen & Wakeley, 2001), we obtained an estimate of M = 0.385 (about one successful female every three generations). It is nevertheless very difficult to unambiguously prove that mtDNA flow still occurs across the barrier. However, the frequency gradient observed within the Atlantic (Fig. 2) seems to be compatible with an introgression tail, which would have been homogenized if gene flow had stopped a long time ago.

Reduction of inter-basins gene flow

Given the high dispersal potential of D. labrax (Pickett & Pawson, 1994) our estimate of genetic differentiation appears to be quite high. Several hypotheses could be proposed to explain, the observed genetic differentiation. The equilibrium between migration and drift may not have been reached since the last contact event. In this case, despite high levels of gene flow, current genetic differentiation may reflect historical differentiation and incomplete re-homogenization. Yet, supposing that contact between the two differentiated populations occurred at the end of the last glaciation (ca. 15 000 years ago) and a generation time is 2–3 years, then 5000–7500 generations would have elapsed since renewed contact. In a simple model with migration and drift, Slatkin (1985) has shown that equilibrium is reached in about t = 1/m generations. Assuming a migration rate of m ≈ 10−3, which could be regarded as a minimum estimate, the time required for reaching equilibrium would be around t = 103 generations. This is about 5000 generations less than the time elapsed since the last glaciation. Therefore, even with a low migration rate compared with the potential for dispersal of the sea bass, the equilibrium has probably been reached. A nonequilibrium state could have persisted since the presumed time of secondary contact only if the migration rate were much lower. We can thus consider that with equilibrium reached or not, the effective gene flow between the Atlantic and Mediterranean basins is limited (a few successful individuals per generations).

A possible explanation for this low gene flow could be that the Almeria-Oran front, where incoming Atlantic waters meet resident Mediterranean layers, might constitute a physical barrier for migration. This may be plausible for the larval stages, but it is certainly not the case for juveniles and adults, which are excellent swimmers and able to withstand large salinity and temperature changes (Barnabé, 1980; Pickett & Pawson, 1994), even if certain genotypes seems to be more adapted than others to salinity changes in Mediterranean (Allegrucci et al., 1994, 1997; Lemaire et al., 2000). Moreover, although genetic discontinuity between Atlantic and Mediterranean populations has been reported for some species (e.g. Daguin & Borsa, 1999), genetic homogeneity has nonetheless been observed on each side of the water front in species with similar or even less pronounced dispersal abilities (Borsa et al., 1997). Hydrological factors alone cannot therefore explain the persistence of the observed genetic disjunction. One should therefore consider the potential presence of a genetically determined barrier resulting from population vicariance in conjunction with a physical barrier, both of which are expected to coincide (Barton, 1979). There are some good examples in the literature where the problem of the possible co-occurrence of a physical and a genetic barrier is addressed, including Point Conception in California (reviewed in Burton, 1998) and Cap Canaveral in Florida (e.g. Cunningham & Collins, 1994). In both cases, although currents are thought to reduce dispersal, they seem to be insufficient by themselves to explain the differentiation observed for certain species as genetic homogeneity is found in the same area for other species with similar dispersal abilities.

Comparisons of differentiation levels between nuclear and cytoplasmic loci

The inter basin θC estimate at mitochondrial cytochrome b is more than 30 times greater than the θN estimated at the six nuclear microsatellite loci. Let Ne and Nef be the total and female effective sizes respectively, and m and mf the total and the female migration rate, respectively. In populations with balanced sex-ratio at mutation-drift equilibrium, and because of haploidy (vs. diploidy) and maternal (vs. biparental) inheritance, the estimate of θ at a cytoplasmic locus is expected to be about four-fold larger than at a nuclear one.

With the expressions of θ being inline image for a nuclear locus and inline image for cytoplasmic locus with maternal inheritance (Fitzsimmons et al., 1997), the upper limit of the ratio inline image (1 + 2Nefmf) cannot exceed 4 when Nem ≫ 1 in the absence of both biased sex ratio (Nef = Ne/2) and sex-biased migration rate (mf = m).

Any departure from these conditions could lead to another ratio between nuclear and mitochondrial θs and gene flow estimates. For example, a sex ratio bias in favour of males would lead to a reduced female effective size and therefore an increased sensitivity to drift effect that implies a higher inline image. Sex related migration rate might also introduce a strong bias. A female effective migration smaller than the male one would directly increase inline image.

In the following we discuss the possible contribution of drift, mutation, migration and finally reproductive isolation in relation to the strong discrepancy observed between genetic variation at nuclear and mitochondrial loci.

Drift

A reduced female effective population size, which would increase drift effects and decrease female gene flow, could lead to a much larger estimate of θ at a mitochondrial locus. If differences between cytoplasmic and nuclear θ were because of biases in sex ratio, we should observe the same ratio between the two at intra- and inter-basin levels. Our data show that the expectation of a ratio of 4 between nuclear and mitochondrial θs seems to be quite well fulfilled at the intra-basin level (2.71 for Mediterranean and 4.11 for Atlantic/Alboran), but not for inter-basin comparisons (37.65). Thus male biased sex ratio, if it does exist, cannot in itself explain the observed differences of genetic variation between loci. Indeed, observations in natural samples are discordant as there is evidence of both male (Blasquez et al., 1995) and female biased sex ratios (Arias, 1980).

Mutation

Differences in mutation rate between classes of markers could be invoked to explain the differences in the level of genetic variation. On the one hand, mitochondrial DNA of vertebrates is reputed to evolve rapidly (review by Rand, 2001) and could lead to elevated diversity and high divergence between isolated populations. Microsatellite loci are also known to have high mutation rates (e.g. Jarne & Lagoda, 1996), a feature which is often associated with homoplasy (Garza & Freimer, 1996; Viard et al., 1998) and which could mimic gene flow and lead to an underestimation of genetic differentiation (but see Estoup et al., 2002). In addition, the small amount of inter-population variation at microsatellites could be explained by the high intra-population gene diversities compared with the ones observed at cytochrome b (Hedrick, 1999).Yet, the genetic differentiation of microsatellite loci in sea bass is far from the limit as shown by the two following examples. First, the genetic differentiation among populations (FST) in the congeneric species D. punctatus is five times larger than in D. labrax at the same microsatellite loci, despite a greater number of alleles (Bonhomme et al., 2002). Secondly, genetic variation at an intronic locus was assessed on the same set of Atlantic and western Mediterranean samples (excepted MMAR and MNAD) (C. Lemaire & F. Bonhomme, unpublished). Only three alleles were found and inline image reached 0.047, which is of the same order of magnitude as what we found for microsatellites, and indeed far from what was expected, if it was a quarter of the mitochondrial value (i.e. 0.167). We can thus conclude that differences in diversity level are not likely to explain the discrepancy between nuclear and cytoplasmic markers.

Migration

Sex-biased migration rates have been deduced from comparison of cytoplasmic and nuclear markers in many organisms and especially marine ones (e.g. Fitzsimmons et al., 1997). As previously stated, if differences in θ estimates between cytoplasmic and nuclear markers were because of a systematic sex-biased migration, we should observe the same differences at intra- and inter-basin levels.

Reproductive isolation mechanisms

Finally, to explain a far more pronounced reduction in mtDNA compared with nuclear gene flow between the two basins rather than within each basin, sex-biased reproductive isolation or direct selection acting on mtDNA may be invoked. Although selection is known to occur on mtDNA (see Rand, 2001 for a review), a direct involvement of mtDNA in the barrier, in the form of disruptive exogenous natural selection, seems unlikely. On the other hand, accumulating evidence suggests that the mitochondrial genome is involved in epistatic relation with nuclear genes and that it could adapt differentially to differentiated nuclear backgrounds (Clark & Lyckegaard, 1988; Rand, 2001). A second possibility lies with a female-biased hybrid breakdown that may account for an impeded maternal transmission. One can hypothesize that female hybrids could be counter-selected, or that hybrid individuals differentiate much less often into females than males. Indeed, a large part of sex-determination seems to be eco-phenotypic in sea bass, for quite a long undetermined period of 6 months (Saillant et al., 2003), the female corresponding on the average to the fastest growing individuals, and it is possible that this mechanism is perturbed by hybridization. Female gene flow would thus be restricted as observed in our data.

A hybrid zone in Alboran Sea?

Two observations suggest that the differentiated genomes may meet in a hybrid zone in Alboran Sea. First, the sample from Mar Menor (EMAM), which borders the Alboran sea at its eastern side (Fig. 1), is significantly differentiated from all the other samples at nuclear loci and is typically Mediterranean at mitochondrial locus. Three microsatellite loci (Labrax-3, Labrax-8 and Labrax-29) cluster this sample to the Mediterranean group while the three remaining loci clusters it nearer to Atlantic samples (details not shown). This may reflect differential introgression close to the hypothetical contact zone.

Secondly, intermediate mtDNA frequencies are observed in two samples from Alboran Sea (MMAR, MNAD). Naciri et al. (1999) suggested the Almeria-Oran oceanic front as a barrier against gene flow. The occurrence of ‘Mediterranean’ haplotypes at higher frequencies in the Alboran Sea means that Mediterranean individuals have crossed the Almeria-Oran front. Moreover, as expected in the centre of hybrid zones where incoming gene flow from source populations overrides the dispersion of hybrid genotypes out of the central part (Barton & Hewitt, 1989), these two samples show a higher correlation coefficient between loci as shown on Fig. 5. Linkage disequilibria sensu stricto were not estimated because of small sample size of the two Alboran samples. Such an observation further corroborates a post-zygotic component of the barrier to gene flow.

Figure 5.

Average of the coefficient of correlation between loci (Rij) (Weir, 1979) at the 18 samples of Dicentrarchus labrax.

Genetic divergence is often correlated with hydrographical features in the sea. Nevertheless, accumulating evidence suggests that such physical barriers are often insufficient to explain observed genetic differentiation. The comparison of genetic variation at nuclear and cytoplasmic loci for the common sea bass has enabled us to infer that the Atlantic/Mediterranean differentiation is not due in itself to the Almeria-Oran front but more likely proceeds from biological reproductive isolation processes finding their roots in glacial vicariance, as is the case also for many terrestrial taxa (Hewitt, 2002). The barrier to gene flow detected seems to rely, at least partly, on sex-biased reproductive isolation that may be associated with a certain amount of hybrid breakdown as is the case in hybrid zones. Precisely, it is expected that such zones should come to be localized in regions of limited dispersal (Barton, 1979), the Almeria-Oran oceanic front being indeed likely to play this role.

Acknowledgments

The authors are particularly indebted to L. Bahri-Sfar, M. Naciri and F. García de León for their contribution to this study. They also wish to extend their thanks to P. Boudry, N. Bierne, J.-D. Durand and P. Borsa for helpful comments. Many thanks also to Tim Sharbel and Scott M. Blankenship, as well as an anonymous reviewer for all the corrections on the manuscript. This work was financed by contract IFREMER URM no. 16 to F.B.

Ancillary