Mechanisms of incomplete prezygotic reproductive isolation in an intertidal snail: testing behavioural models in wild populations

Authors


Emilio Rolán-Alvarez Departamento de Bioquímica, Genética e Inmunología, Facultad de Ciencias, Universidad de Vigo, 36200 Vigo, Spain. Tel: +34 986 812578; fax: 34 986 812556; e-mail: rolan@uvigo.es

Abstract

Two morphs (ecotypes) of the marine snail Littorina saxatilis coexist along Galician exposed rocky shores. They hybridize, but gene flow is impeded by a partial prezygotic reproductive barrier, and we have earlier suggested that this is a case of incipient sympatric speciation. To assess the mechanisms of prezygotic reproductive isolation, we estimated deviations from random mating (sexual selection and sexual isolation) of sympatric snails in 13 localities on the shore, and performed mate choice experiments in the laboratory. We also investigated the microdistribution of both morphs over patches of barnacles and blue mussels in the hybridization zone. We used computer simulations to separate the mechanisms contributing to reproductive isolation.

On the shores sampled, male–female pairs were strongly assortative both with respect to morphs (mean Yule's V = 0.77) and size (mean Pearson's = 0.47). In the laboratory, males of both morphs mounted other snails and mated other males and juveniles at random. However, mature females of equal sizes mated assortatively with respect to morph. The two morphs were nonrandomly distributed over barnacle and mussel patches in the hybridization zone. Monte Carlo simulations showed that this microdistribution could explain about half the morph and size relationships in male–female pairs, while a simple rejection mechanism, rejecting the first 1–3 mates if they were of contrasting morphs, accounted for the remaining part of the reproductive isolation, and for parts of the size relationships found between mates. A size discriminant mate choice mechanism may also, to a lesser extent, contribute to the sexual isolation. Sexual selection was observed for female size (larger ones being favoured) and among certain morphs, but distinct biological mechanisms may cause these processes.

Introduction

Speciation is, undoubtedly, a fundamental process in evolution, but still we have nothing like a predictive theory of speciation (Coyne & Orr, 1998). Even details of the commonly accepted allopatric model of speciation are under intense discussion (see Futuyma & Mayer, 1980; Barton & Charlesworth, 1984; Orr & Orr, 1996). Possibly, general mechanisms of speciation differ between groups of organisms such as marine and terrestrial, or animals and plants, which makes it even harder to suggest ubiquitous rules. While mechanisms such as geographical isolation among sexually reproducing taxa, and Haldane's rule among animals, have gained wide acceptance (Coyne & Barton, 1988; Coyne & Orr, 1989, 1998), the importance of stasipatric, sympatric and founder event speciation, as well as speciation by reinforcement, are still intensely discussed (Futuyma & Mayer, 1980; Barton & Charlesworth, 1984; Butlin, 1987; Coyne & Barton, 1988; Howard, 1993; Bush, 1994; Moya et al., 1995; Dall, 1997; Coyne & Orr, 1998).

Speciation is mostly defined as the evolution of reproductive isolation through prezygotic or postzygotic mechanisms (Mayr, 1963; Coyne & Barton, 1988), although other views exist (e.g. Templeton, 1989). Prezygotic mechanisms are essentially mating behaviours, or mechanisms involved in what Paterson (1980) termed the ‘mate recognition system’. One way of quantifying prezygotic reproductive isolation is to compare the number of between-species and within-species matings of two closely related species (Spieth & Ringo, 1983; Gilbert & Starmer, 1985; Coyne, 1992). However, neither putative cases of incipient speciation nor naturally occurring cross-mates are easily found in nature, which makes this approach useful in but a few organisms (see Littlejohn & Watson, 1985; Butlin, 1995). Indeed, there are no cases of incipient species in which all the steps of evolution of the mate recognition system or a reproductive barrier are known (Gibbons, 1979; Butlin, 1987, 1995).

The marine prosobranch snail, Littorina saxatilis (Olivi), reproduces sexually, with separated sexes and internal fertilization. It has direct development, giving birth to 0.5-mm crawl-away miniature snails. On exposed rocky shores of Galicia (NW Spain), L. saxatilis lives at densities of 10–1000 m−2 from the splash zone to below mid-water level. Over at least tens of kilometres the species is strikingly polymorphic with two main ecotypes (sensuTuresson, 1922): one ridged and banded (RB), living in the upper-shore zone of barnacles, and one smooth and unbanded (SU), found in the lower-shore zone of blue mussels. The two ecotypes overlap and hybridize in the midshore mosaic of barnacle and mussel patches, hybrids being present at low frequencies (usually 5–20%) (Johannesson et al., 1993). We define hybrids on a few inherited traits (ridges and bands; see Johannesson et al., 1993); however, it is presently impossible to know if these are true F1 or a complex mixture of intergrading genotypes. In the midshore, all morphs appear and even copulate with each other (Johannesson et al., 1995), although the microdistribution tends to be somewhat patchy (Johannesson et al., 1995; Otero-Schmitt et al., 1997; Erlandsson et al., 1998, 1999). Furthermore, snails from the upper shore are able to migrate to the lower shore, and vice versa, over a time scale of days or weeks (Johannesson et al., 1993; Erlandsson et al., 1998), which supports the conclusion of truly sympatric ecotypes (sensuFutuyma & Mayer, 1980).

Low survival rates of alien as opposed to native morphs at specific shore levels, and hybrids having similar survival rates as the parental forms in the midshore (Johannesson et al., 1995; Rolán-Alvarez et al., 1995a, 1997) supports the idea of strong selection gradients promoting the Galician polymorphism in Littorina saxatilis (Rolán-Alvarez et al., 1997). Thus the Galician hybrid zone can be considered a contact zone (sensuBarton & Hewitt, 1985), although we will refer to it as a hybrid zone following the broader and more extensive view found in the literature (see Barton & Hewitt, 1985). In addition to differences in ornamentation and colour, a number of other characters show nondiagnostic differences between the two morphs, for example size, temperature tolerance, migration, attachment ability, embryo size and radular morphology, and in most of these traits we have support for genetic, presumably adaptative, differentiation (Johannesson et al., 1993, 1997; Rolán-Alvarez et al., 1996, 1997). Despite this differentiation in adaptive traits, the two morphs do not fall into two clades when using allozymes as genetic markers, and morphological and genetic differences are not correlated (Johannesson et al., 1993). This suggests parallel evolution of similar ecotypes from different sites, and thus a sympatric origin of the two morphs as a whole. This is not surprising since a number of different ecotypes of this species are found over its whole distribution (Reid, 1996).

Despite the presumed sympatric origin of the two Galician ecotypes, we have earlier found strong assortative mating between them in two midshore populations (Johannesson et al., 1995). The assortative mating obstructs gene flow between the morphs (Rolán-Alvarez et al., 1996) and promotes morph cohesion in midshore (Johannesson et al., 1993, 1995), which as a result may be considered a case of incipient speciation. This species is capable of multiple copulations and breeds year-round (Fretter & Graham, 1980; Reid, 1996). Copulating pairs are easily detected directly on the shore (Raffaelli, 1977; Saur, 1990). During mating, the filament of the penis is inserted in the bursa copulatrix of the female and sperm is transferred. Active sperm can be stored in the seminal receptacle for more than 6 months in L. saxatilis and other littorinids (Rolán-Alvarez, 1992; Johannesson, personal observation).

The aims of the present study were to test the generality of the reproductive isolation over a number of sites and to analyse the causes of nonrandom mating. Therefore, we assessed mating behaviours in 13 populations of snails along 20 km of Galician shore. We furthermore performed mate choice experiments in the laboratory and used computer simulations to test different mechanisms of reproductive isolation. Our results suggest at least two main components of isolation: behavioural mate choice and microhabitat selection.

Materials and methods

Sampling procedures

We sampled mating pairs from 13 populations, each separated by at least 50 m of rocky shore (Fig. 1), which is enough for them to be considered discrete genetic populations (see, for example, Johannesson & Tatarenkov, 1997). Samples were from two seasons and from three different years (Table 1). Only mating pairs in which the male penis was clearly inserted in the mantle cavity of the partner were sampled. We searched only for pairs in midshore areas where both parental morphs were present.

Figure 1.

  Sampling localization at the Baiona coast at two seasons (summer in open circles; autumn in filled circles). The four localities at Silleiro are, from bottom to top, 1, 2, 3, 4.

Table 1.  Frequency of three morphs (RB, HY and SU) in all snails sampled of L. saxatilis and pair types found in midshore at the different samples and seasons. n is the number of snails or the number of pairs sampled. Thumbnail image of

In 10 of the populations (BAR94, POR94, CAN94, ESC94, SEN94, SI494, SI191, SI291, SEN95 and CEN95) we picked noncopulating individuals from the same micropatch (diameter about 15 cm) as the mating pairs. In SEN95 and CEN95, noncopulating individuals were sampled from somewhat larger areas. All individuals were sexed and the morphs and shell heights recorded. Further sampling details can be found elsewhere (see Johannesson et al., 1995; Rolán-Alvarez et al., 1995a).

Statistics, sexual selection and sexual isolation estimators

We used a classical approach to assess mating behaviour in ‘true’ (male–female) and ‘false’ (male–male and male–juvenile) pairs (Merrel, 1950; Spieth & Ringo, 1983), decomposing mating behaviour into two independent and additive components: sexual selection and sexual isolation (i.e. assortative mating). To estimate sexual selection we compared the morph frequencies of noncopulating and copulating males and females using the cross product estimator (Cook, 1971; Spieth & Ringo, 1983; Knoppien, 1985; Partridge, 1988). This fitness estimator (w) ranges from zero to infinity and estimates the sexual fitness of one morph (we used RB or SU) relative to a reference morph (we used the hybrid morph, HY). Accordingly,

inline image

where Cop is the number of copulating individuals of one sex, and Noncop the number of noncopulating individuals of the same sex. The significance of the cross product estimator was obtained by bootstrapping the morph frequencies (simultaneously) in copulating and noncopulating data (after 1000 resamplings), using a compiled BASIC program (see Johannesson et al., 1995; Rolán-Alvarez et al., 1995b, for details).

Sexual isolation of morphs was assessed by comparing observed frequencies of different pair combinations to those expected under random mating, including only the copulating snails in the calculation. Yule's V statistic was calculated in order to quantify the degree of deviation from random mating (Pielou, 1977; Gilbert & Starmer, 1985). This index can be considered as one of the best available alternatives to estimate sexual isolation (Gilbert & Starmer, 1985; Rolán-Alvarez & Caballero, unpublished observation). This index is estimated according to

inline image

where RBRB is the number of homotypic (within morph) ridged and banded pairs, SUSU is the homotypic pairs of smooth and unbanded snails, RBSU and SURB are the heterotypic pairs, and MRB, FRB, MSU and FSU are the number of RB male, RB female, SU male and SU female, respectively. The index ranges from 1, for maximum assortative mating, to − 1, when maximum disassortative mating occurs. We did not use parametric tests based on known sampling variances for this estimator (see Gilbert & Starmer, 1985) because they usually underestimate type I errors (Rolán-Alvarez & Caballero, unpublished observation). Instead we used contingency χ2 tests, generating pseudo-probabilities by Monte Carlo simulations with 1000 resamplings (see Zaykin & Pudovkin, 1993), or classical t-tests when independent estimates were available. Sexual isolation for size was tested by analysing the correlation (Pearson's r) between male and female sizes in each pair (Moore, 1987).

Classical parametric tests were carried out with the SPSS/PC package, while for unbalanced data we used a compiled BASIC randomization ANOVA program (4999 randomizations) (Manly, 1991).

Monte Carlo simulations

We proposed that two factors were likely to contribute to sexual isolation of morphs or snails of different sizes: nonrandom microdistribution and mating behaviour. We used Monte Carlo sampling (Manly, 1991) to assess how much a nonrandom microdistribution of morphs or sizes among the studied microareas would contribute to the observed sexual isolation of these traits. A male/female pair was picked at random with replacement from the group of mating and nonmating snails within each microarea. The degree of assortative mating in this sample of pairs was estimated, either as the correlation coefficient (rm), indicating size-assortative mating, or as Yule's V (Vm), indicating sexual isolation of morphs. This process was repeated 5000 times to generate the empirical resampling distribution of rm and Vm. Thereafter, we estimated the frequency of observed coefficients (Vo, ro) equal or larger than the resampling coefficients (Vm, rm, respectively). If this frequency was smaller than 2.5% (≤ 0.05 for a two-tail test), we concluded that a nonrandom microdistribution could not account for all the sexual isolation observed. This test was carried out in all samples and separately in pooled samples from summer 1994 and autumn 1995. We did not pool all data to increase sample size, as we could not a priori exclude the possibility of seasonal differences in mating behaviour and sexual selection (see below for further arguments a posteriori). However, we could pool samples within season as we have previously found that the sexual isolation is neither frequency- nor density-dependent (Rolán-Alvarez et al., 1995a), and because morph frequencies, and sexual isolation, were rather homogeneous (see seasons below).

In a second set of simulations, we evaluated the contribution of the nonrandom microdistribution and different mate choice mechanisms. The degree of assortative mating was also estimated, either as the correlation coefficient (r(m+c)), indicating size-assortative mating, or as Yule's V (V(m+c)), indicating sexual isolation of morphs. We used the same design of Monte Carlo sampling as before, except that this time only pooled samples were used (to maximize statistical power) and mating pairs from each microarea were sampled according to any of five different ‘mating algorithms’. The first algorithm assumed a size-discriminant choice (Kelly & Noor, 1996). In this a mating pair was picked at random but rejected if sizes of mates differed by more than x mm, where x = 0.25, 0.50, 0.75, 1.00, 1.50, 2.00 or 3.00 mm. The second algorithm assumed a morph-discriminant choice and a pair was only accepted if the mates were of certain morph combinations. We used two different discriminant rules; in the first (case 1), pairs were only accepted if the mates were of the same morph and hybrids were considered a separate morph. In the second (case 2), all pairs were accepted except those of a smooth and unbanded and a ridged and banded snail. However, if a certain number of trials did not result in an accepted pair, the rules were given up and the next pair accepted. The number of trials allowed (‘the discriminant rate’) indicated the strength of the discriminant mechanism (sensuKelly & Noor, 1996). Thus, for example, a discriminant rate of zero corresponded to complete random mating within each microarea, while a discriminant rate of 10 indicated a choice mechanism which failed only after 10 unsuccessful attempts to obey the rule.

The third mating algorithm was a mate-size preference mechanism, in which one individual (female or male) was first randomly drawn from the microarea, and thereafter a mate was picked at random which, however, had to be larger than the partner size (by 0.25, 0.50, 0.75, 1.00, 1.50, 2.00 and 3.00 mm) to be accepted (the rules were given up after no success). In the fourth and fifth mating algorithms one individual was drawn at random, and then the largest and the smallest partners were chosen, respectively, within the microarea. In all these simulations, we estimated the sampling distributions of the statistics by 5000 resamplings.

Mate choice in the laboratory

In a laboratory mate choice experiment, 10 males of one parental morph were placed together with 10 RB and 10 SU females in a small glass tray. The experiment was repeated four times within 2 weeks, twice using SU-males and twice using RB-males. Mountings and copulations were noted during 2.5 h. A mounting was one male climbing onto the shell of another individual and remaining there for more than 30 s. A copulation was defined as a pair being in position for copulation for more than a few seconds (Saur, 1990). However, at the end of the experiment we noticed two very different approximate copulation times in the samples: a few copulations lasted a few seconds only, while the rest were maintained for at least 10 min. The former were defined a posteriori as ‘failed copulations’ and the latter as ‘successful copulations’. It seems probable that no sperm was transferred in the ‘failed copulations’. After any copulation, the individuals involved were removed and replaced with new individuals of the same morph, size and sex. All snails involved in an experiment were dissected afterwards and sexual status and parasite infestations were assessed.

Results

The proportion of RB-morphs, hybrids, and SU-morphs varied substantially over the 18 midshore samples picked during summer and autumn seasons (Table 1). The proportion of RB individuals was smaller in summer than in autumn (randomization ANOVA; Fr = 9.78, d.f.1 = 1, d.f.2 = 16, = 0.009). Also the frequencies of different pairs varied over samples; male–male pairs, for example, were more frequent in the autumn than in the summer (Fr = 8.42, d.f.1 = 1, d.f.2 = 16, = 0.004) (Table 1).

Sexual isolation and sexual selection in field populations

Sexual isolation between morphs, as estimated by Yule's V (Vo), indicated strong deviations from random mating among male–female pairs in all but two of the 18 samples (Table 2). Three of the samples even had complete assortative mating (Yule's Vo = 1.0). In male–male and male–juvenile pairs, however, sexual isolation was roughly halved compared with female–male pairs (randomization ANOVA; summer: Fr = 19.12, d.f.1 = 1, d.f.2 = 21, = 0.001, autumn: Fr = 5.28, d.f.1 = 1, d.f.2 = 10, = 0.041). This pattern was compatible with a model of important contributions from both nonrandom microdistribution (explaining the sexual isolation in all pair types) and mating behaviour (explaining the part of the sexual isolation unique to the male–female pairs). Both true and false pairs revealed significant sexual isolation for size; mean r (ro) over the 12 summer samples was 0.466 (SD = 0.307) for true and 0.439 (SD = 0.222) for false pairs.

Table 2.    Sexual isolation observed index (Yule's Vo see text) for true (male/female) and maladaptative (male/male and male/juvenile) pairs in 12 samples from summer and six from autumn are presented. The estimated parametric standard deviations (SD) were obtained following Gilbert & Starmer (1985). The significance of each sexual isolation estimate was obtained by pseudoprobability chi-square statistics (χ2, see Zaykin & Pudovkin, 1993). The averages across-pair types and seasons and the significance of the t-test (for averges) ae also shown. See Table 1 for number of pairs studied in every sample. Thumbnail image of

Cross product estimates of morph sexual selection in six summer samples showed that males of the RB morph had, on average, lower sexual fitness than hybrids and males of the SU morph (Table 3). A general trend was detected in males from pooled data, in which the sexual fitness of RB and SU morphs differed from that of the hybrids in summer, but not in autumn (Table 3). Females did not show any significant sexual fitness estimate among samples, although both pure morphs showed equal or larger sexual fitness than hybrids (Table 3).

Table 3.    Cross-product estimates of male and female fitnesses in six samples from summer and four from autumn. The fitness of the hybrid morph (HY) is set to 1 and the values of the ridged and banded (RB) and smooth and unbanded (SU) morphs are related to this. Fitness values higher or lower than 1 represent more fit or less fit, respectively. Bootstrap standard deviations are given in parentheses. Single and pooled values are tested against the reference value (w = 1) using bootstrapping. The significance of averaged values was evaluated by a parametric t-test. Thumbnail image of

Sexual selection on size was evaluated by a one-way ANOVA comparing mean sizes of copulating and noncopulating snails within each morph, using pooled samples from summer. Significant differences were found among the SU females (= 7.3, d.f.1 = 1, d.f.2 = 592, = 0.007), with copulating snails being larger than noncopulating ones (4.1 mm and 3.9 mm, respectively). However, none of the other groups revealed any difference between sizes of copulating and noncopulating snails (SU males: = 0.1, d.f.1 = 1, d.f.2 = 754, = 0.77; RB males: = 2.6, d.f.1 = 1,  d.f.2 = 373,   = 0.106; RB   females: = 4.9, d.f.1 = 1, d.f.2 = 217, = 0.48).

Mechanisms of sexual isolation

We used Monte Carlo simulations, assuming random mating within each microarea, and calculated the sexual isolation using the simulated data set. This indicated how much a nonrandom microdistribution of morphs and sizes contributed to the sexual isolation. In most samples, the nonrandom microdistribution of morphs and sizes explained a large part of the sexual isolation observed, but in some samples, as well as in all pooled samples, Vm and rm were significantly smaller than Vo and ro, respectively (Table 4). On average, the nonrandom microdistribution of morphs and sizes explained roughly half the amount of sexual isolation observed during autumn, and somewhat less during summer. Thus although a nonrandom microdistribution is not enough to explain the sexual isolation, it is an important factor.

Table 4.  Monte Carlo simulation to test if snail microdistribution could account for observed morph and size relationships in pairs. Mating pairs were randomly resampled from different microareas in every sample (only samples with copulating and noncopulating individuals) and pooled samples. Vo and ro are the observed Yule's Vrs (see Table 3) and Pearson's r, respectively. Significance of averaged estimates was evaluated by t-test. Vm and rm are the averaged Yule's Vrs and Pearson's r coefficients obtained from the Monte Carlo distribution (1000 resamplings) in every locality and pooled samples. n is the number of microareas studied per locality, and Prob. is the probability that the observed Yule's Vrs be included in the Monte Carlo distribution generated by nonrandom snail microdistribution. Thumbnail image of

The remaining part of the sexual isolation may possibly be due to mate choice. The size-discriminant choice mechanism could, however, not explain the observed Yule's V for morphs in the summer samples. Indeed, all simulated estimates of V(m+c) were significantly smaller than the observed one (Fig. 2). In autumn samples, on the other hand, this mechanism could explain the observed sexual isolation between morphs for strong enough discriminant rates in those cases where mates of slightly larger sizes where accepted (x = 0.25–1 mm, for discriminant rates of 5–20; Fig. 2). These results suggest that the size-discriminant choice mechanism could not, at least not alone, be the mate-choice mechanism responsible for the remaining part of the observed sexual isolation. However, this mechanism could, under some of the conditions applied, explain the assortative mating with respect to size (threshold value of 2 mm for most discriminant rates; Fig. 2). These simulations likewise explained sexual isolation among parental morphs and hybrids (RB/HY and HY/SU mating pair patterns), but could not account for sexual selection on size or morph (results not shown).

Figure 2.

  Results of the simulations for the size-discriminant choice mechanism are shown for pooled summer and autumn samples. Two parameters were varied during simulations: the threshold size rejection choice mechanism (from 0.25 to 3.00) and the discriminant rate (from 1 to 20). Symbols represent the average estimate of the statistics (V(m+c) and r(m+c)) in the simulation (after 5000 resamplings). Symbols appear as open if the observed statistic (Vo and ro) can be explained by the resampling distribution (a two-tailed 95% confidence interval of resampled statistics included the observed one), and as filled if the observed statistic cannot be explained by the resampling distribution (a two-tailed 95% confidence interval of resampled statistics did not include the observed one).

A morph-discriminant choice mechanism explained the observed sexual isolation between morphs if adopting a less strong discriminant mechanism, that is a low discriminant rate and a rule tolerating all kinds of pairs except RB/SU pairs (Case 2; Fig. 3). Size assortative mating could be explained in the autumn data while not in the summer data by all combinations of discriminant rates and rejection rules. Sexual isolation with respect to other pairs of morphs (RB/HY and HY/SU pairs) was explained in both cases and seasons, while sexual selection on morphs and size was usually rejected (results not shown).

Figure 3.

  Results of the simulations for the morph-discriminant choice mechanism are shown for pooled summer and autumn samples. Two parameters were varied during simulations: the morph rejection choice mechanism (Case 1 and Case 2; see text), and the discriminant rate (from 1 to 20). Symbols represent the average estimate of the statistics (V(m+c) and r(m+c)) in the simulation (after 5000 resamplings). Symbols appear as open if the observed statistic (Vo and ro) can be explained by the resampling distribution (a two-tailed 95% confidence interval of resampled statistics included the observed one) and as filled if the observed statistic cannot be explained by the resampling distribution (a two-tailed 95% confidence interval of resampled statistics did not include the observed one).

The three remaining discrimination criteria, mates of a certain larger size, and the largest or the smallest snail available within each microarea, could not explain the sexual isolation of morphs and the size assortative mating. Furthermore, sexual selection on size or morph could not be explained by any of these behavioural mechanisms. Thus, in conclusion, the morph-discriminant choice mechanisms and a nonrandom microdistribution seemed the best combination of factors explaining the observed sexual isolation with respect to morph, while the sexual isolation with respect to size was not completely explained. Furthermore, none of the behavioural mate choice mechanisms could explain the patterns of sexual selection observed in these populations.

Mate choice in laboratory

In the mate choice experiments, females of both morphs were approximately equally sized, and slightly larger than males (as in natural populations). No snails were parasitized and nearly half of the RB females were without eggs or embryos in their brood pouch. A total of 153 mountings and 10 copulations were recorded during the four experiments, of which 30% were male–male pairs. Both RB and SU males mounted females at random, with respect to female morph (Table 5). If distinguishing between ‘successful’ (copulation time >  10 min) and ‘failed’ (copulation time < 1 min) copulations, however, we found that all of the successful copulations were between snails of the same morph, while all the failed copulations were between morphs (Table 5). This suggests a discriminant mate choice mechanism which leads to the possible rejection of mates of a contrasting morph.

Table 5.  Mating behaviour experiments in the laboratory, in which 10 males of one morph were placed with 20 females (half of each morph). This was done four times. A mounting is when a male is on the shell of another snail for more than 30 s. A copulation is defined in this context when a typical copulation position is maintained for longer than a few seconds, while ‘successful copulation’ was used for those pairs maintaining the copulation position for longer than 10 min and ‘failed copulation’ used for those pairs interrupting copulation after less than 1 min (see Materials and methods). The null hypothesis of similar frequencies of mountings within and between morphs of male–female pairs is checked by a chi-square of goodness of fit (χ2). Thumbnail image of

Discussion

In sexual and dioecious organisms mate choice is often a crucial prezygotic isolation mechanism separating closely related taxa (Spieth & Ringo, 1983; Coyne, 1992). Recent data suggest that mate choice may even lead to the build up of reproductive isolation between forming conspecific groups (Galis & Metz, 1997). Postzygotic genetic barriers may be as effective as prezygotic ones, but postzygotic mechanisms are burdened by the waste of reproductive effort. Thus an appropriate mate recognition system (Paterson, 1980) will probably be a more efficient evolutionary strategy than hybrid inviability.

Physiological or morphological constraints may, however, restrict the evolution of perfect prezygotic barriers. In many species of littorinid snails, for example, the mate recognition system is imperfect at distance. This allows for low frequencies of interspecific mountings (Raffaelli, 1977; Saur, 1990) as well as varying frequencies (up to 30%) of male–male mountings (Struhsaker, 1966; Raffaelli, 1977; Saur, 1990; Johannesson et al., 1995; Erlandsson & Rolán-Alvarez, 1998). While in one species, Littorina littorea, males have some ability to distinguish between male and female mucous trails (Erlandsson & Kostylev, 1995), males of L. saxatilis do not seem to have this ability (Erlandsson et al., 1999). Male copulation attempts with females of other species or with other males are, furthermore, interrupted after a few minutes in L. littorea (Saur, 1990), and this prezygotic mate choice mechanism may possibly be present in L. saxatilis as well. Similarly, males spend a longer time copulating with more profitable (i.e. large) females. This may explain the higher mating success of large females of Littorina littorea (Erlandsson & Johannesson, 1994). Possibly, sexual selection and assortative mating in some populations of L. saxatilis may be explained by the same mechanism (Erlandsson & Rolán-Alvarez, 1998; this study).

Mating behaviour is a complex life history trait involving numerous different biological mechanisms. In this study we aimed to estimate the contribution of individual mechanisms of mating behaviour that create reproductive isolation (sexual isolation). We found that sexual isolation between RB and SU morphs in the overlapping midshore zone is caused both by a nonrandom microdistribution and by a morph discrimination behaviour. The nonrandom microdistribution of morphs in the midshore contributed considerably to sexual isolation, explaining about 40% of the isolation of male–female pairs in autumn and 60% in summer samples, on average. We have earlier reported a nonrandom microdistribution of the parental morphs in these populations (Johannesson et al., 1995; Otero- Schmitt et al., 1997), but we did not know why snails of the same morph aggregated. One reason may be that the patches of barnacles and mussels are not equally attractive for the two morphs, as they differ in different physical parameters and epiphytic flora (Otero-Schmitt et al., 1997). Besides, surface complexity differs between barnacles and mussels, and the former seem to offer more favourable refuge sites than the latter (Kostylev et al., 1997). Density of the SU-morph decreases with increasing surface complexity while that of the RB-morph remains constant, and this may result in a nonrandom habitat association of the two morphs (Kostylev et al., 1997). On the other hand, size aggregation has also been observed in these morphs in midshore (Johannesson et al., 1995), which might partially be a consequence of intrinsic differences in size between these morphs and search for suitable refuges (see Erlandsson et al., 1999, for further discussion on mechanisms of microhabitat choice).

The nonrandom microdistribution does not, however, explain all the sexual isolation observed, and indeed we have indications of female–male interactions which produce more sexual isolation than is observed in the male–male and male–juvenile pairs. As the simulations show, a morph-discriminant mechanism would explain at least some of the observed patterns. For example, a mechanism interrupting copulation between extreme morphs, but allowing snails to mate at random after a few unsuccessful attempts, would account for the sexual isolation in summer and autumn samples, as well as the correlation between mate sizes in summer. Indeed, the laboratory mate choice trials suggested copulation time to be crucial in morph discrimination, with between-morph copulations being interrupted after a short time while within-morph copulations lasted for several minutes. The discriminant rate was rather low (1–3 attempts before mating with any partner), and this impedes the completion of sexual isolation. In addition, the morph-discriminant mechanism can, in certain cases, be frequency-dependent, as a low frequency of one morph would produce a pattern indistinguishable from random mating. This would perhaps explain the absence of sexual isolation found in lower shores where the RB morph is very rare or almost absent (see Butlin, 1995; Johannesson et al., 1995).

Size and morph are partially correlated factors in our data set, and it is difficult to study the effects of each factor separately. The simulations showed, however, that although a size-related mate choice mechanism may contribute to the sexual isolation, the rules we used could not explain more than a small part of the observed sexual isolation among different morphs (Fig. 2). However, we believe that this minor contribution may be real, as we need this mechanism to explain all the size assortative mating observed. In fact, rejecting mates of extreme size may explain the size relationship not explained by the discriminant morph mechanism (summer samples of Fig. 3). A similar mechanism may be responsible for the size assortative mating observed in some Swedish populations of L. saxatilis (Erlandsson & Rolán-Alvarez, 1998).

In earlier studies we found suggestions of sexual selection against hybrid females which may contribute to the reproductive isolation in these populations, since hybrids act as an important genetic bridge between the parental morphs (Johannesson et al., 1995; Rolán- Alvarez et al., 1995a). In the present study we found a nonsignificant trend of sexual selection against female hybrids in the autumn, and indeed the size of the selection coefficient was rather similar to what we found earlier (Johannesson et al., 1995; Rolán-Alvarez et al., 1995a). We also found sexual selection favouring SU and hybrid males over RB males in the summer samples (Table 3). Sexual selection between these female or male morphs may be a consequence of sexual competition rather than mate choice (Johannesson et al., 1995; Rolán-Alvarez et al., 1995a). The simulation studies support this, as no mate choice mechanism could explain the sexual selection coefficients observed in this study.

Prezygotic reproductive isolation can evolve directly by natural selection. If it develops when postzygotic reproductive isolation is not already complete, it is called reinforcement (sensuButlin, 1987, 1995). The idea of prezygotic barriers evolving by reinforcement is classical and theoretically feasible, but criticism has been raised recently because of theoretical constraints as well as lack of putative examples from nature (Butlin, 1987; but see Butlin, 1995, and Hostert Ellen, 1997). In fact, the hybrid zone of Galician L. saxatilis studied here has been suggested as a possible example of reinforcement (Butlin, 1995). The apparent random mating of the few RB individuals present in lower shore (Johannesson et al., 1995) may support reinforcement (Butlin, 1995), but may equally well be a consequence of frequency-dependent mate choice (Johannesson et al., 1995; this study). The low discriminant rates estimated for the morph-discriminant mechanism makes further evolution towards complete speciation difficult. Moreover, fitness estimates of hybrids compared to pure forms in this hybrid zone do not support the reinforcement model (Rolán-Alvarez et al., 1997). Alternative mechanisms have been proposed to reinforcement which seem theoretically just as feasible (Kelly & Noor, 1996). We believe reinforcement is less likely to contribute to the evolution of reproductive isolation among Galician morphs of L. saxatilis.

So what is the alternative to reinforcement explaining the evolution of reproductive isolation on Galician shores? Strong disruptive natural selection affects the L.  saxatilis populations on the studied shores (Johannesson et al., 1995; Rolán-Alvarez et al., 1995a, 1997). The shift in reproductive and behavioural traits (habitat or mate choice) may be a secondary consequence of selection on other traits adapting snails to upper and lower shore levels. Habitat choice is considered a critical promoter of sympatric speciation leading, in the first place, to polymorphic species (García-Dorado, 1986), and later, possibly, to sympatric speciation (Bush, 1994; Johnson et al., 1996; Kawecki, 1996). Although we are far from a predictive theory for the evolution of reproductive isolation, sexual isolation has probably evolved here in sympatry, as a secondary consequence of differential selection in some other characters.

Acknowledgments

We thank E. Torroja, J. De La Torre, C. García, A. Caballero and N. H. Barton for general discussion and suggestions on earlier versions, and D. Reid for corrections on a later version. L. M. Díaz helped with fieldwork and with preparation of figures and C. García contributed to the experimental design and sampling of two populations. This work was partially supported to E.R.-A. by funds from the XUNTA de Galicia (XUGA 30105B98) and the University of Vigo (64102C605), to K.J. by funds from the Swedish Natural Science Research Council, and to R. Cruz by DGICYT (PB94–0649).

Ancillary