Large discrepancies in differentiation of allozymes, nuclear and mitochondrial DNA loci in recently founded Pacific populations of the pearl oyster Pinctada margaritifera

Authors


 Sophie Arnaud-Haond, Station Méditerranéenne de l'Environnement Littoral, 1, quai de la daurade, 34200 Sète, France. Tel.: + 33 6746 3388; fax: +33 4 6746 3399; e-mail: s-arnaud@crit.univ-montp2.fr

Abstract

Abstract This study presents a comparative analysis of population structure applied to the pearl oyster (Pinctada margaritifera) from the Central Pacific islands using three classes of molecular markers: two mitochondrial genes (mtDNA), five anonymous nuclear loci (anDNA), and eight polymorphic allozymes. Very low levels of haplotype diversity and nucleotidic divergence detected for mtDNA validate the hypothesis of a recent (re)colonization of Polynesian lagoons after their exondation during the last glaciations. Some nuclear loci, however, showed highly significant FST values, indicating a reduced amount of larval exchange between archipelagos at present. A large interlocus variance of FST was nevertheless observed. We discuss whether this pattern is inherent to the stochasticity of the drift process since recolonization, or if it could result from balancing selection acting on certain loci. This study illustrates once more the need to combine the analysis of several kinds of loci when unrelated phenomena are likely to leave their footprints on genetic structure.

Introduction

Since their emergence, biochemical and molecular techniques have allowed the study of populations' genetic structure using allozymes, mitochondrial DNA (mtDNA) and a wealth of nuclear DNA markers [e.g. mini- and micro-satellites, single-copy nuclear DNA and anonymous nuclear DNA (anDNA)]. The underlying null hypotheses for interpreting allelic distribution as reflecting genetic exchanges between populations is that neutral gene flow and genetic drift are the predominant forces driving the evolution of population structure, and that observations are made at migration and sometimes mutation/drift equilibrium. Thus, patterns observed with different neutral loci should be comparable, provided that their respective effective sizes and mutation rates are properly taken into account.

Indeed, levels of differentiation estimated from organelle (or sex chromosome when they exist) and autosomal nuclear markers are expected to differ at equilibrium because of effective population size differences. When these differences are much higher than expected, it may be an indication of differential dispersion of gametes, as sometimes observed in plants (Latta & Mitton, 1997; Levy & Neal, 1999), or differential migration behaviour of males and females, as reported, for instance, for green turtles or fin whales (Bérubéet al., 1998; Lyrholm et al., 1999). Moreover, a marker locus with a lower effective size (like markers from organelle genomes) often has its diversity more strongly affected by historical events such as founder effects or bottlenecks than do autosomal nuclear genes. These demographic events can also generate discrepancies among loci irrespective of their effective sizes, as they may be responsible for an increased transient variance in the response of different loci to migration/drift. More generally, the question of this interlocus variance is still a matter of methodological debate, since the early work of Lewontin & Krakauer (1973) and all the ensuing discussion about what the neutral variance in a differentiation estimator as FST should be (see Baer, 1999 and literature cited therein). On the other hand, since the initial report on the enzyme Lap-1 in Long Island Sound mussels (Koehn et al., 1976), studies on marine species have often shown discrepancies among loci interpreted as evidences for balancing, or directional selection acting on allozymes with the argument that they have metabolic functions (for instance, Atlantic cod: Pogson et al., 1995; American oyster: Hare et al., 1996; sea bass: Lemaire et al., 2000).

Populations of the black-lipped pearl oyster Pinctada margaritifera cumingi from the Central Pacific are strongly dependent on coral formations and are distributed in lagoons, although occasional observations have been made on external slopes. Palaeogeographical studies have shown sea level fluctuations during Pleistocene glacial events, with episodes of emersion of Polynesian atolls, which made the survival of some of the species inhabiting lagoons impossible until about 8000 years ago in these areas (Ladd, 1960; Paulay, 1990; Woodroffe et al., 1990; Bard et al., 1996; Cabioch et al., 1999). According to Paulay (1990, 1991), approximately one-third of the bivalve species inhabiting Central Pacific islands became locally extinct during those glacial episodes. However, the same authors show that the soft-bottom bivalves were much more affected than the hard substrate ones, like Pinctada margaritifera, and that species inhabiting outer-reef slopes were much less affected than those exclusively inhabiting lagoons. Thus, present-day populations of P. margaritifera from the Central Pacific should show signs of a recent recolonization process. Populations from French Polynesia have previously been surveyed for allozymic polymorphism (Blanc, 1983; Blanc et al., 1985, 1996; Durand & Blanc, 1986). Despite these suspected Pleistocene demographic events, levels of variability were comparable with those commonly reported in the literature for allozymes in other bivalve species that supposedly had not undergone bottlenecks or founder events. No structuring was observed among samples located in the same lagoon. The differentiation between archipelagos was null or extremely low, except for the Marquesas sample that was found to be quite different from the others. The low level of genetic structure detected with allozymes could thus either result from high levels of gene flow (if populations were at migration/drift equilibrium), or from little present-day gene flow as a result of a colonization event sufficiently recent that no differentiation yet occurred. A third possibility could be that balancing selection on some allozymes maintains similar allele frequencies on the different islands and impedes differentiation.

To further test these three hypotheses and give some information on the reorganization of populations since the presumed recolonization of lagoons, we propose a combined analysis of heterogeneity in allelic frequencies of different markers in this study. For this purpose, we reanalysed data from the previous allozyme studies and compared them with the geographical distribution of allelic frequencies from two other kinds of markers: mitochondrial and anonymous nuclear loci. These were obtained from the same individuals as previous studies for all but one origin. Restriction fragment length polymorphism (RFLP) analysis was performed on two segments of mtDNA. In addition, five presumed neutral anDNA loci were analysed for length polymorphism.

Materials and methods

Sample collection

Sampling locations are mapped in Fig. 1. Samples from the Society [Manuae (Scilly), Maupihaa (Mopélia), Raiatea], Tuamotu (Takaroa) and Gambier (Mangareva) islands were collected by diving between summer 1981 and autumn 1983. Samples from Marquesas (Hiva Oa) were collected by diving during summer 1985, and samples from the Cook Islands (Suwarrow, Penrhyn), kindly provided by J. Benzie and B. Ballment, were collected during summer 1992.

Figure 1.

Sampling locations of Pinctada margaritifera among Polynesian archipelagos. Number of individuals successfully analysed is indicated for each class of markers, respectively, mitochondrial DNA/allozymes/nuclear DNA. Principal currents in the area are also reported.

All samples except those from the Cook Islands were preserved in liquid nitrogen (a piece of ground muscle stored in buffer Tris–HCL 0.1 m, pH 8.0). Samples from the Cook Islands consisted of pieces of adductor muscle fixed in 80% ethanol; no protein electrophoresis was performed on these samples.

DNA extraction for molecular analysis

Genomic DNA was obtained from approximately 0.5 g of chopped and subsequently air-dried tissue (for samples preserved in ethanol), or 200 μL of tissue extract, incubated in 300 μL of extraction buffer (0.02 m Tris–HCl, 0.02 m EDTA, pH 8), 10% SDS, in the presence of 0.25 μg of proteinase-K at 55 °C for 5–10 h. The procedures of DNA extraction, precipitation and storage were similar to those described in Sambrook et al. (1989). The nucleic acid pellet obtained after precipitation in 100% ethanol was washed with 70% ethanol, air-dried, resuspended in 100 μL of deionized water and preserved at −20 °C. Concentration of DNA extract was measured by fluorometry, and found to average 300 ng μL−1.

Genetic markers employed

Allozymes

The electrophoretic procedures were described in publications reporting the original results (Blanc, 1983; Blanc et al., 1985). Eighteen allozymic loci were analysed in these studies, among which the following eight were shown to be polymorphic at the 5% level: Alcohol dehydrogenase (Adh), Aldehyde oxydase (Ao), Amylase (Amy-1), Carbonic anhydrase (Car), Glucose phosphate isomerase (Pgi), Leucine aminopeptidase (Lap-2), Malate dehydrogenase (Mdh-1), and Octanol dehydrogenase (Odh).

Anonymous nuclear loci

The five anonymous nuclear markers used: pinucl1, pinucl2, pinucl3, pinU4 and pinaldo were described in a preliminary note (Arnaud et al., 2002). Two techniques were applied for the selection of those polymorphic nuclear loci, DALP (Direct Amplification of Length Polymorphism: Desmarais et al., 1998) for pinucl1, 2 and 3, and EPIC (Exon Primers Intron Crossing: Palumbi, 1995) for pinuclU4 and aldo. The two last markers were isolated with primers, respectively, amplifying an amylase intron in Crassostrea gigas, and an aldolase intron in several fish species. However, no sequence similarity was detected (BLAST on GenBank), either with the DNA sequences, or with the protein sequences derived from the three possible open-reading frames. Sequences of the primers, numbers of alleles observed, and approximate length of those alleles are summarized in Table 1. For PCR, typically 0.4 μCie of radioactive reagent (α33P) was added to 10 μL reaction volumes with final concentrations of 300 μm each dNTPs, 1.8 mm MgCl2, 0.4 μm nonlabelled primer, 0.03 μm labelled primer, about 30 ng of template DNA, 1X Taq buffer and 0.25 units of Taq polymerase. PCR products were run in 6% denaturing sequencing polyacrylamide gels and visualized by autoradiography. Ambiguities in reading a genotype were checked by either re-running the same PCR product, or repeating the PCR reaction.

Table 1.  Details of the primers used for amplification and sequencing of mitochondrial and nuclear loci, GenBank accession for the allelic sequences, number of alleles, approximate size of the amplified fragment (data for the five anDNAs from Arnaud et al., 2002 ).
LocusPrimer sequencesGenBank Acc. no.Number of allelesApp. size
Pinucl 1/ AF 414 061–0666 
R2AAACTTTTGTATTCTTATCAT  220–240
F2CAGCTATGACCAGACACGTG   
Pinucl 2/ AF 414 067–0715 
R2TAGGGTACAGTCCGAGGTGTTCCG  260–330
F2TCATTAATTGTCGAAAACTTGTC   
Pinucl 3/ AF 414 072–0735 
RACATTTCTTCAACACATTGG  170–190
FCCCAAAATTGTTCGGGGAGC   
Pin U4/ AF 414 055–0575 
RCATTGACTTGGCTACCGAGAC  80–100
FGCTGGTGGTGATGCATTTTGG   
Pin aldo/ AF 414 058–0605 
RCTTTGCAACGGCTGTATAGC  200–220
FTAGGCCACCCAATCAGAGTTC   
Mitochondrial  11 
COX I AF 374 320–330  
HCO2198TAAACTTCAGGGTGACCAAAAAATCA  710
LCO1490GGTCAACAAATCATAAAGATATTGG   
rDNA 12S AF 374 331–336  
RTTTCCCGCGAGCGACGGGCG  420
FGAAACCAGGATTAGATACCC   

mtDNA

Two mitochondrial genes were amplified for RFLP analysis (Table 1), a 710-bp fragment of the mitochondrial cytochrome c oxydase subunit I gene (COI) (Folmer et al., 1994), and a fragment of c. 410-bp of the mitochondrial 12S rRNA gene (12S) (Mokady et al., 1994). Digestion of the PCR products was initially performed with a set of 21 enzymes on a subsample of 48 individuals (12 individuals from each location Suwarrow, Takapoto, Mangareva and Hiva Oa). Six restriction enzymes showed polymorphism (DdeI, HaeIII, HindIII, Sau3A, TaqI, VspI) in the subsample and were subsequently used on the whole sample. Separation of the digestion products was performed by electrophoresis in 2% agarose gels, stained with ethidium bromide, and visualized under UV light. Eleven composite haplotypes were defined by possession of a unique set of restriction sites across all enzymes, and assigned to each individual.

Haplotype sequences were determined using both radioactive and automatic fluorescent sequencing. Primers used were the original amplification primers, but for several cytochrome-oxydase I haplotypes, the forward primer was replaced by an internal sequencing primer defined on the basis of the first alignments, as a poly A-T repetition caused polymerase errors making sequencing of the first 100 bp almost impossible. Sequences were aligned using BioEdit 4.7.1 (Hall, 1999), and identity of the fragments amplified with universal primers (Cytochrome Oxydase subunit I and rDNA12S) was verified by using the BLAST algorithm of the translated protein sequence in GenBank, before submitting the complete alignment to GenBank (Table 1). The consensus tree of haplotypes was searched with the neighbour-joining algorithm as implemented in the Phylip 3.5c package (Felsenstein, 1993) after bootstrap resampling of the sequences (1000 replicates). Bootstrap values above 70% were reported on the tree constructed from the original data set with the neighbour-joining algorithm. Trees were rooted with haplotypes of Pinctada mazatlanica from Mexico (Arnaud et al., 2001).

Population genetic data analysis

For each of the three types of markers, genetic diversity within populations was estimated by the unbiased gene diversity (H) (Nei, 1987). Nucleotide diversity (π) of the mitochondrial genes was estimated using RFLP data (Nei & Tajima, 1983) including all enzymes (polymorphic and monomorphic ones). Tajima's D-test of neutrality or fluctuation of the effective population size (Tajima, 1989a,b), which compares numbers of segregating sites and average numbers of pairwise differences among mtDNA haplotypes, was performed with sequence data using Arlequin version 1.1 (Schneider et al., 1997).

Linkage disequilibrium between nuclear loci was tested with the procedure of Black & Krafsur (1985) and a 1000 permutation test. Deviation from Hardy–Weinberg expectation was measured by FIS according to Weir & Cockerham (1984), and significance tested by permutations of alleles within populations (1000 permutations). Multilocus FST was estimated globally per pair of populations and per pair of archipelagos with the estimator θ of Weir & Cockerham (1984). The significance of the θ-values was assessed by 1000 permutations, and Bonferroni correction was applied (Rice, 1989). A Mantel test was performed in order to test for any correlation between genetic and geographical distances. As suggested by Rousset (1997) in a two-dimension model, for each population pair we plotted inline image (1 − inline image) against the logarithm of the geographical distance in kilometres. These analyses and tests were performed using the Genetix 4.0 package (Belkhir et al., 1996–2002).

Additionally, a multiallelic monolocus neutrality test was performed to test for neutrality using the distribution of allelic frequencies at nuclear loci in the metapopulation. As this test is not yet published (N. Raufaste & F. Bonhomme, submitted), we describe its principle in brief. It is based on the properties of two multiallelic estimators of FST, those of Weir & Cockerham (1984) and of Robertson & Hill (1984), whose respective sampling biases have been recently reassessed (Raufaste & Bonhomme, 2000), and which weight rare and frequent alleles differently as they combine differently the bi-allelic estimates (i.e. the contribution of each allele taken separately against all the others), the RH estimator giving more weight to the rare alleles than the WC estimator. Because of this property, these two estimators have been shown to respond differently to various selection and dominance regimes (N. Raufaste & F. Bonhomme, submitted), because rare and frequent alleles are affected in a different way: for instance, a slightly deleterious allele will see its frequency kept low in most population and thus more homogeneous than a neutral allele able to drift freely. It will thus contribute to reduce the θRH estimator whereas θWC, giving most weight to the more frequent alleles, will be almost unaffected. Symmetrically, an allele maintained in higher (and more homogeneous) frequency by some sort of balancing selection will contribute to lower the θWC estimator as compared with θRH which will be sensitive to neutrally drifting low-frequency alleles and remain higher. The difference between these two estimators follows a distribution that can be approximated by simulations under the null hypothesis of neutrality, in an infinite island model, at migration–drift equilibrium. The parameters of the simulations (initial allelic frequencies and migration rate) are chosen to fit, at best, the observed values. The test statistic (the difference between θWC and θRH) is then calculated with the ^θ-value for each locus in the dataset, and a significant departure of the observed value from the simulated distribution could be interpreted as evidence for non-neutrality of the distribution of allele frequencies at the locus concerned. It has been shown that selection coefficients as small as s = 0.001 with migration rates as large as m = 0.01 could cause a detectable effect (N. Raufaste & F. Bonhomme, submitted). This test was applied on all nuclear loci with the programme Neutrallelix (http://www.univ-montp2.fr/genetix/labo.htm), except on two allozymes and an anDNA locus (Car, Pgi and pinaldo) on which it could not be performed as the first one exhibited only two alleles (in this case θWC and θRH have exactly the same expression) and the last ones showed negative values of FST (hence no simulation could be performed).

Results

Marker diversity

The numbers of individuals successfully analysed for each kind of marker in each sampling location appear on Fig. 1. The levels of gene and haplotype diversity at the mitochondrial and nuclear marker loci employed are shown in Table 2. Generally low mtDNA haplotype diversities were observed, between only 0.09 and 0.23, except for Maupihaa (0.43) and Marquesas (0.60). The tree in Fig. 2 shows a low level of divergence among haplotypes (from 0.36 to 1.83%), suggesting a short coalescence time, but no congruence was observed with their geographical distribution. This is correlated with negative and significant values for Tajima's D calculated on the basis of mtDNA haplotype sequences and frequencies, except in the case of Marquesas (Table 2). However, this observation is in contrast with the average allozyme diversity (including the eight monomorphic loci from Durand & Blanc, 1986), which was comparable with the amount commonly reported in other bivalve species, between He = 0.24 and 0.29 (Table 2). The average anonymous nuclear gene diversity ranged between He = 0.38 and 0.46.

Table 2.  Haplotype and nucleotide diversity estimated for mitochondrial and nuclear markers in Pinctada margaritifera samples. Results of Tajima's D values are reported after haplotype and nucleotide variability.
Collection sites (West to East)Haplotype diversity (±SE)Nucleotide diversityTajima′D-valuesGene diversity – allozymes- (±SE)Gene diversity anDNAs (±SE)
  • *

    Significant,

  • nonsignificant.

Cook archipelago
 Suwarrow0.229 (±0.095)0.0018−1.88*0.404 (±0.196)
 Penrhyn0.211 (±0.089)0.0019−1.78*0.412 (±0.134)
Society archipelago
 Maupihaa0.427 (±0.122)0.0032−1.80*0.276 (±0.008)0.383 (±0.176)
 Raiatea0.100 (±0.088)0.0006−2.03*0.292 (±0.006)0.398 (±0.182)
 Manuae0.087 (±0.078)0.0004−2.08*0.255 (±0.006)0.396 (±0.232)
Tuamotu archipelago
 Takaroa0.218 (±0.103)0.0013−1.56*0.238 (±0.005)0.483 (±0.190)
Gambier islands
 Mangareva0.087 (±0.078)0.0005−2.07*0.251 (±0.008)0.400 (±0.21)
Marquesas archipelago
 Hiva Oa0.600 (±0.078)0.0044−0.280.244 (±0.007)0.518 (±0.198)
Figure 2.

Neighbour-joining tree illustrating nucleotide divergence between haplotypes of Pinctada margaritifera , rooted with the closely related taxon P. mazatlanica . Boostrap values obtained after 1000 permutations are reported when exceeding 70%.

Hardy–Weinberg equilibrium, linkage disequilibrium and neutrality

Results of the tests of Hardy–Weinberg equilibrium for each marker in each location are shown in the Appendix. After correction for multiple tests (Rice, 1989), strong and significantly positive values of FIS were found at six allozymic loci for two to four populations. Only one of the anDNA exhibited significant deviation from Hardy–Weinberg equilibrium: the locus pinucl1 in the Takaroa, Mangareva and Raiatea samples. Deviations from linkage equilibrium estimated between all possible pairs of loci according to Black & Krafsur (1985) were not significant.

The null hypothesis of neutrality was not rejected by the multiallelic test for four of the five anDNA loci (the fifth could not be tested), nor for four of the six allozyme loci that could be tested, as the differences between Robertson & Hill's and Weir & Cockerham estimators were included in the random distribution simulated under the null hypothesis. However, neutrality was rejected at the 5% confidence level for the two remaining allozyme loci: Amy-1 (P = 0.02) and Lap-2 (P = 0.03).

Genetic structure among islands

Pairwise differences observed with each nuclear locus are detailed in Table 3. It must first be noted that a large variance was observed between loci. Actually, most loci showed a genetic differentiation between some archipelagos (Car, Mdh-1, pinucl2, pinucl3, pinU4) and some between most islands (Ao, Adh, Odh, pinucl1) whereas four loci (Pgi, Amy-1, Lap-2 and pinaldo) showed almost no genetic structure at all. These differences among loci deserve further attention, especially because neutrality was rejected for two of them, Amy-1 and Lap-2, and the remaining in the last group, Pgi and pinaldo, could not be tested with the neutrality test as the global FST value obtained for these loci were negative. Accordingly, Amy-1 and Lap-2 were not considered further to illustrate patterns of gene flow, their contribution to the global FST being in any case very small and congruent with the other loci (not shown). The possible reasons for the homogeneity of their distribution are discussed further below.

Table 3. θ Values between eight Pacific pearl oyster populations. Tables present FST /population pair and global inline image value for each locus. FST values as estimated according to Weir and Cockerham are indicated.
  1. * Significant value ( P  < 0.05) after 1000 permutations, these are presented on a grey background when they remained significant after Bonferronni correction for multiple tests.

  2. Ma, Manuae; Mo, Mopelia; R, Raiatea; T, Takaroa; G, Gambiers; M, Hiva Oa; S, Suwarrow; P, Penrhyn.

inline image
inline image

Significance of the inline image-values between all pairs of populations from the four archipelagos are detailed for each type of marker in Tables 3 and 4. In almost all instances, the sample from the Marquesas Islands was differentiated from all other archipelagos. The mitochondrial marker did not show significant haplotype frequency differences among other archipelagos, whereas some anDNAs and allozyme loci showed genetic heterogeneity between nearly all archipelagos, except between Tuamotu and Gambier.

Table 4. θ Values between all Pacific pearl oyster populations, obtained on the basis of the nuclear markers retained after neutrality test (i.e. when excluding Amy-1 and Lap-2 ), and compared with the results obtained with mitochondrial DNA. FST values as estimated according to Weir and Cockerham are indicated.
  1. * Significant value ( P  < 0.05) after 1000 permutations, and these are presented on a grey background when they remained significant after Bonferronni correction for multiple tests .

  2. Ma, Manuae; Mo, Mopelia; R, Raiatea; T, Takaroa; G, Gambiers; M, Hiva Oa; S, Suwarrow; P, Penrhyn.

inline image

For the Society and Cook archipelagos, several samples from different localities were analysed. Estimates of inline image between each pair of localities are shown in Table 4 for mitochondrial and nuclear loci. This permitted an analysis of genetic differentiation at smaller geographical scales, giving a more precise picture, as some globally distinct archipelagos contain samples that are not differentiated for all marker loci. It can be noted, in particular, that the Takaroa sample from Tuamotu is not strongly different from Mangareva (Gambier) and Suwarrow (Cook) samples with the nuclear markers.

Reynolds genetic distances [D = –ln(1 − FST); Reynolds et al., 1983] were used to construct a neighbour-joining tree illustrating genetic distance between populations (Fig. 3). No significant result was obtained when applying the Mantel test to these data, suggesting that geographical distance is not the principal factor influencing dispersal.

Figure 3.

Neighbour-Joining tree illustrating genetic distances (Reynold's D ) between populations and archipelagos, on the basis of nuclear data (allozymes and anDNA).

Discussion

Heterozygote deficiencies

The cause of heterozygote deficiencies is a problem that has been largely debated in the marine bivalve literature (reviewed in Zouros & Foltz, 1984; Gaffney, 1994; David, 1998; see also Del Rio-Portilla & Beaumont, 2001). First observed with allozymes, they have been interpreted as a result of either null alleles (i.e. nonstaining of some alleles; Foltz, 1986), imprinting, aneuploidy, selection against heterozygotes, Wahlund effect, homogamy or inbreeding. It is now widely accepted that the two principal causes that have some generality are associative overdominance because of the existence of a high genetic load in bivalves (Hedgecock et al., 1996; Bierne et al., 1998, 2000) and nondetectable (null) alleles. Associative overdominance would result in heterozygote deficiency when partial inbreeding occurs in the population. While not impossible, this is probably uncommon in bivalves and not powerful enough to create strong Hardy–Weinberg desequilibria (David et al., 1995). Non-detectable alleles seem more likely for both technical and biological (post-transcriptional modifications) reasons, and capable of producing appreciable heterozygote deficits even at moderate frequency (Beaumont & Pether, 1996). For PCR-based markers, null alleles occur as a result of primer mismatch resulting in amplification failure in some variants. Hare et al. (1996) for example, working on the oyster Crassostrea virginica, demonstrated the importance of such technical artefacts, and showed experimentally that heterozygote deficits could be removed by redesigning primers. This was also performed in P. margaritifera for two of the markers used in the present study by Arnaud et al. (1999). In our case, FIS values were highly variable depending on the locus considered, and generally higher with allozymes than with anDNAs. This allows us to reject the category of hypotheses that imply similar levels of FIS whatever the marker analysed, i.e. Wahlund effect, homogamy or inbreeding. Considering those arguments, technical artefacts are probably the main cause of the high levels of heterozygote deficiencies observed at some allozymic loci in our data. Anyhow, if null or nonstaining alleles are randomly associated with all other detectable alleles, this should not affect the estimates of FST, a parameter that has been devised precisely to remove the effects of local heterozygote deficiencies on the between-sample differentiation.

Nucleotide divergence, mitochondrial and nuclear diversities, and recent history of the populations

Together with the low levels of haplotype diversity, the significant negative values of Tajima's D suggest either a recent reduction in population size, or directional selection on mitochondrial haplotypes. Few reports implying direct or indirect (hitchhiking) selection on the mitochondrial genome in natural populations are available. These are mostly concerned with incompatibility in hybrid systems (Kilkpatick & Rand, 1995; Assmussen & Basten, 1996; Goodisman et al., 1998), or association with introgression phenomena (Wilson & Bernatchez, 1998). On the other hand, several studies have provided support for the hypothesis of a drastic reduction or disappearance of populations of most species living in lagoons during the last glaciations, because of sea level drop and lagoon drainage (Paulay, 1990; Planes et al., 1993), and to the hypothesis of a recent re-colonization from refuge areas.

Our molecular data, supported by the extremely rare observations of oysters on reef external slopes, validate the hypothesis that present populations of P. margaritifera of Polynesia come from a recent colonization or re-colonization event from refuge areas, apart from those in the Marquesas. According to Paulay (1990), sufficiently large populations of inner-reef specialists could have persisted during glacial episodes in two major potential refuges: (i) the coral formation of Western Pacific, from the Great barrier reef to the Melanesian islands, (ii) the Australs (Rapa) and Marquesas islands, which lack developed reef systems of typical Central Pacific islands and instead have extensive embayments dominated by unconsolidated sediments and which harbour several typical inner-reef species. We do not have any samples from the Western Pacific to properly test these hypotheses. Allozyme studies performed on populations of the Western Pacific (Benzie & Ballment, 1994) showed a slightly higher level of mean expected heterozygosity among polymorphic loci (0.60–0.73) than those observed in Polynesia (0.52–0.65), however, no mitochondrial data are available from populations from this region. On the other hand, the most common mtDNA haplotype in Marquesas is the same as the almost unique haplotype observed in Central Polynesian populations, which make this archipelago a good candidate as a source population for an initial group of colonists of low effective maternal population size.

Comparison of population structure inferred from different classes of markers

Under the neutral model, a large variance of the FST values is expected at each locus (Lewontin & Krakauer, 1973; Beaumont & Nichols, 1996; Baer, 1999). In the present study, observed metapopulation-level FST values ranged from 0.140 for Adh to −0.004 for Pgi. Although the number of available loci is not sufficient to conduct a test as proposed by Beaumont & Nichols, we nevertheless feel that the contrasting patterns observed among our 11 segregating loci deserve further attention. Three allozymic and one anDNA loci (Lap-2, Amy-1, Pgi, pinaldo) and mtDNA show almost no genetic differentiation, while the other loci (Car, Mdh-1, pinucl2, pinucl3, pinU4, Adh, Ao, Odh and pinucl1) allowed differentiation of the Marquesas sample and to detect genetic structure between most populations pairs.

At this point, two scenarios can be considered (i) anDNAs pinucl1, 2, 3 and U4 and allozyme loci Adh, Ao, Car, Mdh-1 and Odh give a clear image of a somewhat restricted gene flow between locations (overall average FST on these nine loci is 0.086), in which case we need to find alternative explanations (balancing selection or interlocus variance because of recent re-colonization) for the remaining nuclear loci and mtDNA. Conversely (ii) mitochondrial and nuclear data at Amy-1, Lap-2, Pgi and pinaldo are representative of the action of rather high gene flow between populations and archipelagos. In this case, the situation observed for the remaining allozyme loci and anDNAs (Adh, Ao, Car, Mdh-1,Odh, pinucl1,2, 3 and U4) would necessarily imply some sort of disruptive selection on them.

We have two reasons to favour the first scenario: (i) the multiallelic test of Raufaste & Bonhomme indicates no rejection of the null hypothesis when applied to anDNA and the second group of allozymes (Adh, Ao, Car, Mdh-1 and Odh), whereas it does for two allozymes of the first group (Amy-1 and Lap-2). Pgi and pinaldo cannot be tested because of a negative inline image-value. One could claim that this represents a departure from the conditions required for the test (migration–drift equilibrium in an infinite island model), but this seems improbable given that the parameters used were good enough to predict correct values of the test statistic (θRH − θWC) at six loci, and considering the large number of generations elapsed since probable recolonization. (ii) In the case of mtDNA, on the other hand, even if neutral, the lack of efficient detection of genetic differentiation could be the result of its very low variability. Actually, stronger differentiation on nuclear than on mitochondrial loci has been reported in Atlantic char (Brunner et al., 1998) and in wallabies (Pope et al., 1996). The explanation given was, as in our case, the very low level of mtDNA haplotype variability. The same explanation could be given to explain the lack of differentiation observed with the anDNA locus pinuclaldo, in which heterozygosity was as low as the mtDNA.

In the case of Pinctada, directional selection on anonymous anDNAs and those allozymic loci showing genetic differentiation is considered unlikely, balancing selection or some other sources of interlocus variance on other nuclear loci is considered possible, and the level of mitochondrial diversity is too low to provide evidence for population differentiation, the actual level of gene flow is probably better reflected by the anDNAs pinucl1, 2, 3 and U4 and allozyme loci Adh, Ao, Car, Mdh-1 and Odh.

Restriction of gene flow between populations of Pinctada margaritifera

The estimate of Wright's FST obtained on the basis of anDNA and allozyme allele frequency data, when excluding Amy-1 and Lap-2 suspected of responding to selection (qualitatively, the same results are obtained when including all loci) showed that the samples of P. margaritifera do not form a homogeneous group, as each archipelago is discernible from the other (Table 4). Such differentiation suggests there is some restriction to gene flow, despite a pelagic larval stage of approximately 3 weeks (Saucedo & Monteforte, 1997). These results contrast with the report of relative genetic homogeneity observed with allozymes in populations of P. margaritifera from North-western Australia (Benzie & Ballment, 1994). However, other cases of structuring among populations of marine organisms with a long pelagic larval stage have already been reported. Examples are the case of the purple sea urchin Strongylocentrus purpuratus in California (Edmands et al., 1996), the queen scallop Aequipecten (Chlamys) opercularis in Great Britain (Lewis & Thorpe, 1994), and, in the same area as our studies, the convict surgeonfish Acanthurus triostegus (Planes et al., 1996). In the latter, allozymic variation was analysed on samples from almost the same archipelagos as in ours, except that no Cook sample was analysed, and a sample from the Australs was included. The pattern of population structure reported is similar to our observations on P. margaritifera (Fig. 3), and can in the same manner be partly explained by the prevailing currents in that area (Fig. 1; Apel, 1987; Rancher & Rougerie, 1992). The regular Marquesas counter-current, in opposition with the Southern equatorial current, can account for the strong divergence of the Marquesas population, whereas the weaker differentiation shown between the Tuamotu–Gambier group and the Society archipelago can be explained by the unstable situation of the border between the Southern equatorial current and the counter-current.

Acknowledgments

We would like to thank N. Raufaste, N. Bierne, L. Chikhi, M. Goyard, D. Nolan and H. McCombie and Gareth Pearson for their useful comments and suggestions on this article. Special thanks to B. Ballment and J. Benzie for providing samples from Cook islands, to E. Desmarais for advice when applying the DALP methodology, and to J. De Barry and L. Valera for technical help. This work was supported in part by contract IFREMER-URM 16 to F.B.

Supplementary material

The following material is available from http://www.blackwellpublishing.com/products/journals/suppmat/JEB/JEB549/JEB549sm.htm

Table A1 Restriction profiles obtained with the six enzymes detecting length polymorphism. The composite haplotypes were designed a six letter code according to the alphabetical order of enzymes (Ddel, Haell, Hind II, Sau3A, Taql, Vspl). Details of allelic frequencies are exposed in table A2.

Table A2 Allelic and haplotype frequencies for 6 allozymic loci, 5 nuclear loci and mtDNA in eight samples of Pinctada margaritifera from Polynesia. The number of haplotypes on which estimates are based are shown in bold. For allozymes and nuclear loci, significant excesses of homozygotes above Hardy-Weinberg equilibrium proportions after a 1000 permutations test are indicated with a *, and are in bold when remained significant after the Bonferonni correction.

Received 24 September 2002;revised 4 February 2003;accepted 14 March 2003

Ancillary