Introduction

Littorina striata King & Broderip 1832 is a strictly Macaronesian (Azores, Madeira, Canary Islands and Cape Verde Islands) periwinkle (Littorinidae) inhabiting the upper eulittoral and littoral fringes of rocky shores (Reid, 1996). Despite the species' planktonic development, it displays a number of shell polymorphisms that do not appear to be randomly distributed at either micro- or macrogeographical scales (e.g. De Wolf et al., 1997, 1998a). This variation mainly involves sculpture (nodulose vs. smooth), shell and aperture size, and shell weight (i.e. thickness) (Reid, 1996; De Wolf et al., 1997). Formerly, the sculpture polymorphism was even interpreted as a taxonomic difference, with nodulose specimens being assigned to a separate species, namely L. affinis d'Orbigny 1839 (in Reid, 1996). Because of this taxonomic uncertainty and because species with planktonic development are not expected to show extensive population differentiation and structuring (Chapman, 1995 and references therein; but see Palumbi, 1996), it has been questioned whether L. striata represents a single, homogeneous gene pool and if so, which factors cause the nonrandom distribution of its shell variation. In order to address these issues, allozyme data have been used to assess the degree of genetic exchange (i.e. gene flow) between L. striata populations and morphotypes at various geographical scales (De Wolf et al., 1997, 1998b). These studies revealed no significant genetic structuring, even though there were signs of a macrogeographical N–S differentiation with respect to numbers of alleles, heterozygosity estimates and esterase band counts (De Wolf et al., 1998c). In addition, there was a tendency for isolation by distance, but this pattern was not statistically significant (De Wolf et al. unpubl. data).

However, these results were based on only four to five polymorphic enzyme loci and nonspecific esterases. Moreover, allozymes may be a biased sample of a genome (e.g. Karl & Avise, 1992; Lu & Rank, 1996) and/or may be subject to selection (e.g. Hilbisch & Koehn, 1985; Johannesson et al., 1995). Hence the allozyme-based population genetic data of L. striata needed to be verified by other, supposedly unbiased and neutral, markers. Random Amplified Polymorphic DNA (RAPD) (Welsh & McClelland, 1990; Williams et al., 1990) may provide such data. Although this technique has many weaknesses (e.g. Clark & Lanigan, 1993; Lynch & Milligan, 1994; Backeljau et al., 1995; Fritsch & Rieseberg, 1996), it also offers some important advantages such as its relative ease and speed, the high degree of (artificially simple) polymorphisms and the virtually inexhaustible pool of possible genetic markers (e.g. Clark & Lanigan, 1993; Lynch & Milligan, 1994; Fritsch & Rieseberg, 1996). Nevertheless, RAPDs have hitherto only marginally been applied in littorinid (Crossland et al., 1993, 1996) or gastropod population genetics (e.g. Jacobsen et al., 1996; Larson et al., 1996; Stothard & Rollinson, 1996).

In the present study we used RAPDs to assess population genetic differentiation in L. striata, i.e. differentiation between nodulose and smooth specimens and between five macrogeographical populations which had been previously analysed by allozyme electrophoresis.

Materials and methods

Sample collection and DNA extraction

A total of 320 specimens of L. striata were collected at wave-exposed shores in Macaronesia, including two populations from the Azores (AZ1: Vila de Nordeste, São Miguel; AZ2: Faial da Terra, São Miguel), two populations from the Cape Verde Islands (CV1: Baia das Gatas, São Vicente; CV2: Mindelo, São Vicente) and one population from the Canary Islands (CA1: Playa de America, Tenerife). Populations were separated by distances ranging from 10 km (over land) between CV1 and CV2 to 2000 km between CV1,2 and AZ1,2.

Genomic DNA was extracted from individual columellar muscles following the CTAB protocol of Winnepenninckx et al. (1996), modified so that tissues were not ground under liquid nitrogen, but instead, directly incubated in CTAB buffer.

RAPD amplification

Prior to PCR, DNA concentrations were adjusted to 50 ng/μL. PCR reactions were performed in a final volume of 25 μL, containing 25 pmol RAPD primer, 50 ng template DNA and a standard quantity ‘Ready To Go’ RAPD analysis mixture (Pharmacia Biotech). Amplifications were performed with the Hybaid Omn-E thermocycler (Hybaid Ltd, Teddington, Middlesex, UK). PCR conditions were: 5 min at 95°C followed by 45 cycles, each of 1 min at 95°C, 1 min at 36°C and 2 min at 72°C. After the last cycle, a final step of 5 min at 72°C was added to allow complete extension of all amplified fragments. Amplification products (10 μL) were separated for 90 min at 120 V on 0.8% agarose TBE gels (1.78 mM Tris; 1.78 mM Boric Acid; 0.4 mM EDTA). These were stained with 10 μL ethidium bromide added to the gel and 0.25 μL ethidium bromide/mL TBE buffer to the TBE running buffer. Gels were photographed under UV light.

Forty random, decamer primers were screened (Kits H and G of Operon Technologies), two of which, OPH-12 (5§-ACGCGCATGT-3§) and OPG-08 (5§-TCACGTCCAC-3§) were retained for further analysis. RAPD bands were scored as present/absent (p/a=1/0) and only well-resolved profiles were considered. The reproducibility of RAPD profiles was tested for each primer, by reanalysing 20 individuals and comparing both profiles.

Data analysis

Assuming that each RAPD band represents a single diallelic locus in Hardy–Weinberg (H–W) equilibrium, the p/a data set can be converted into allele frequencies (e.g. Apostol et al., 1996; Lu & Rank, 1996). However, because RAPDs behave as dominant markers (i.e. 1=AA or Aa and 0=aa), allele frequency estimates are less accurate than with codominant markers (Lynch & Milligan, 1994). Using the fraction of double-recessive genotypes (x) (i.e. band absence at a given locus), the biased frequency of the null allele is q=x1/2 and its unbiased estimate is given by eqn (2a) in Lynch & Milligan (1994).

Unbiased allele frequencies were calculated by hand and were subsequently entered into the program BIOSYS-1 (Swofford & Selander, 1989) to compute heterozygosity levels (H). Unbiased H-values for the ith locus and jth population were obtained according to Lynch & Milligan (1994) (i.e. eqns 4a and 2b). However, because unbiased and biased allele frequencies and H estimates only differed very slightly (Table 1) and as the available software [RAPDFST (Apostol et al., 1996) and POPGENE v. 1.1 (Yeh & Boyle, 1996)] did not allow entering unbiased allele frequencies, all further analyses used biased allele frequency and H estimates.

Table 1 Number of Littorina striata scored (N), and biased allele frequency of the amplified fragment and heterozygosity levels (H). Unbiased allele frequency and heterozygosity levels, according to Lynch & Milligan (1994), are given in bold
Full size table

The homogeneity of allele frequencies between (i) the five populations, (ii) both Azorean populations, (iii) both Cape Verde populations, and (iv) nodulose and smooth Canarian specimens, was tested with the contingency chi-squared analysis option of POPGENE. FST-values and FST-based gene flow (Nm) estimates, for these comparison levels, were estimated with the RAPDFST program. The comparison of nodulose and smooth specimens was restricted to the Canarian population because that was the only one in which both morphotypes were sufficiently represented (nodulose=19; smooth= 21). The numbers of nodulose specimens in the other populations were too small (<7).

The BIOSYS-1 program was used to calculate Prevosti distances between the five populations. For comparative purposes, we did the same for the available, unpublished, allozyme data of these populations. Both distance matrices were subjected to UPGMA clustering and matrix correlations were tested with Mantel's statistic (permutation level=1000) as implemented by the NTSYS v. 1.8 program (Rohlf, 1993).

Results

The two primers retained for analysis yielded six polymorphic, reproducible and consistently scorable RAPD-bands (i.e. six loci: four with OPH-12 and two with OPG-08), whose size ranged from 220 bp to 1636 bp (Fig. 1). None of these RAPD bands was diagnostic for either nodulose or smooth specimens.

Fig. 1
figure 1

RAPD profiles of Littorina striata, using the (a) OPG08 and (b) OPH12 primers of the Operon kits H and G (Operon Technologies). [White arrow represents first locus, respectively OPG08-1 in (a) and OPH12-1 in (b); white and black arrows represent remaining loci in numerical order, respectively OPG08-2 in (a) and OPH12-2, OPH12-3 and OPH12-4 in (b); left lane represents a kb DNA ladder in both (a) and (b) ].

Full size image

Although mean H estimates were more or less comparable among populations, the two lowest values were observed in the Azores (Table 1).

There were six cases of significant genetic heterogeneity. However, after sequential Bonferroni correction only one locus remained significant (locus OPG08–02, all populations compared) (Table 2). In contrast, no allele frequency heterogeneities were observed between nodulose and smooth specimens (Table 2).

Table 2 Probabilities of a genetic heterogeneity test, performed at each locus, by means of contingency chi-squared analysis, for comparison of smooth vs. nodulose shells and Azorean, Cape Verde and all populations, of Littorina striata
Full size table

FST-values were small, ranging between 0.001 and 0.076 (Table 3). The highest mean FST-value was observed among the five populations (mean FST=0.044), although it was more or less halved among Cape Verdian or Azorean populations and decreased even further between the two morphotypes (Table 3). Consequently, FST-based gene flow estimates were high, ranging from Nm=5.475 among the five populations to Nm=16.412 between both morphotypes.

Table 3 FST and Nm estimates for Azorean, Cape Verde and all populations, as well as for smooth and nodulose shells of Littorina striata
Full size table

The UPGMA trees based on Prevosti distances calculated from, respectively, RAPD and allozyme data (Table 4) were comparable in that both placed the Cape Verdian populations at the base of the tree and both showed the same cluster of Azorean and Canary Island populations (Fig. 2 and 3). Yet, in the RAPD tree the Cape Verdian populations appeared paraphyletic with CV2 as the first branch (Fig. 3), whereas in the allozyme tree the Cape Verdian populations appeared as a monophyletic sister group of the Azorean–Canary Islands cluster (Fig. 2). Hence, both distance matrices were not significantly correlated (Mantel test: r=0.267, P=0.202).

Table 4 Prevosti genetic distances between populations of Littorina striata; above diagonal based on RAPD data, below diagonal based on allozymes
Full size table
Fig. 2
figure 2

Prevosti genetic-based UPGMA tree, based on allozyme data, for populations of Littorina striata.

Full size image
Fig. 3
figure 3

Prevosti genetic-based UPGMA tree, based on RAPD data, for populations of Littorina striata.

Full size image

Discussion

Previous surveys of four to five enzyme loci in 66 populations (including the present ones) of L. striata from all over Macaronesia, revealed no consistent and, after Bonferroni correction, significant deviations from H–W equilibrium conditions (De Wolf et al. unpubl. data). This implies that the assumption of H–W frequencies may also be acceptable for the RAPD data. In contrast, none of the enzyme loci was diallelic, but instead carried at least five to seven alleles (De Wolf et al. unpubl. data). Therefore, we feel less confident about the assumption of RAPD diallelism. This issue cannot be resolved without further, labour-intensive work (e.g. Fritsch & Rieseberg, 1996). Yet, imposing a diallelic structure on RAPD data can be considered as a pooling of genotypes, making statistical tests more conservative.

Although eqn (2a) in Lynch & Milligan (1994) provides an unbiased estimate of the null allele frequency (q) by considering small sample sizes, it still may be biased downwards if the null allele is rare (Lynch & Milligan, 1994). Therefore, RAPD analyses should be restricted to amplified bands that are not too common, i.e. those with a frequency of less than 1−(3/N) in each population (Lynch & Milligan, 1994). This was the case in the present study. Anyway, as already noted in Materials and methods, biased and unbiased allele frequencies were very similar (Table 2) because the correction term in eqn (2a) of Lynch & Milligan (1994) differed only slightly from 1 (range: 1.00007–1.005837). Hence the outcome of our analyses is not expected to be affected by whether or not our allele frequency estimates were biased. The same applies to our heterozygosity estimates.

The RAPD heterozygosity values for L. striata (0.244–0.306) were slightly higher than those observed using allozyme loci (0.151–0.216), but tentatively suggested the same N–S differentiation, viz. Hnorth<Hsouth (De Wolf et al. unpubl. data). However, testing the significance of this difference will require more data. Likewise, the RAPD FST values of L. striata (0.002–0.076) were of comparable magnitude to the allozyme-based FST-values (0.001–0.032 De Wolf et al. unpubl. data) and indicated only very limited differentiation between populations. Hence, Nm estimates were high (Nm>≲1). Both the RAPD H- and FST-values in L. striata were comparable to allozyme-based figures in other mollusc species with planktonic development (e.g. Johannesson, 1992; Dayan & Dillon, 1995).

Because none of the RAPD markers studied was diagnostic for shell morphotype, and because there was no significant genetic differentiation between nodulose and smooth specimens, we conclude that only one biological species is involved. This result is consistent with previous allozyme data (De Wolf et al., 1998b) and suggests that the sculpture polymorphism is indeed intraspecific.

In conclusion, the present RAPD data nearly perfectly fit previous allozyme data on L. striata (except perhaps for the basal branches in the UPGMA trees, which produced a nonsignificant correlation between the distance matrices involved), even to the point that tendencies suggested by allozymes remain tendencies with RAPDs (i.e. the N–S difference of H). This congruence between our allozyme and RAPD data strongly contrasts with previous studies, which reported that allozymes underestimate genetic diversity compared to RAPDs (e.g. Apostol et al., 1996; Crossland et al., 1993, 1996). As pointed out by Schaal et al. (1991) and by Karl & Avise (1992), the differences between RAPD- and allozyme-based estimates of genetic diversity suggest that allozymes may not be a representative sampling of the entire genome. Yet, this does not follow from the present work. Indeed, given that there was no reason to assume that our data were non-neutral, we regard the numerical congruence between our allozyme and RAPD data as an indication that in the case of L. striata the allozyme data provide a fairly good picture of the population genetic structure of this species.