Sexual selection on phenotypic traits in a hybrid zone of Littorina saxatilis (Olivi)


  • R. Cruz,

    1. Departamento de Bioloxía Fundamental, Facultade de Bioloxía, Universidade de Santiago de Compostela, 15706 Santiago de Compostela, Galicia, Spain
    Search for more papers by this author
  • E. Rolán-Álvarez,

    1. Departamento de Bioquímica, Xenética e Inmunoloxía, Facultade de Ciencias, Universidade de Vigo, 36200 Vigo, Galicia, Spain
    Search for more papers by this author
  • C. García

    1. Departamento de Bioloxía Fundamental, Facultade de Bioloxía, Universidade de Santiago de Compostela, 15706 Santiago de Compostela, Galicia, Spain
    Search for more papers by this author

Raquel Cruz, Departamento de Bioloxía Fundamental, Facultade de Bioloxía, Universidade de Santiago de Compostela, 15706 Santiago de Compostela, Galicia, Spain. Tel.: +34-981 563100, ext: 13349; fax: +34-981 596904; e-mail:


Step clinal transitions in inherited character(s) between genetically distinct populations are usually referred to as hybrid zones. An example is found in the population of the intertidal snail Littorina saxatilis in Galicia (NW Spain). We studied the shape of the overall fitness surface for sexual selection in this hybrid zone, and the position of hybrids and pure morphs on this surface. We found that sexual divergent selection acted on a combination of phenotypic traits separating the pure morphs, and therefore that sexual selection contributed to morph differentiation. The average fitness of hybrids as a group was not significantly different from that of the pure morphs, but they did show divergent sexual selection in some traits. These results are in agreement with a model of divergent selection favouring both the pure morph as well as those hybrids most resembling each morph. The finding of divergent selection is remarkable because quadratic selection gradients are usually weak in nature.


Most models of sympatric speciation have assumed strong disruptive natural selection leading to genetically based ecological diversification (Tauber & Tauber, 1989). Following the biological species concept of Mayr (1942), the two diverging populations will become different species when there is complete reproductive isolation between them. Isolation may evolve as a correlated response to the different action of natural selection on each population, or may be the result of selection acting through a process of reinforcement (Dobzhansky, 1940). The latter can be defined as the evolution of prezygotic isolating barriers in zones of overlap or hybridization as a response to selection against hybridization (Howard, 1993). Therefore, natural selection may be indirectly or directly involved in both steps of sympatric speciation: the establishment of a stable polymorphism and the evolution of reproductive isolation.

Hybrid zones are invaluable natural laboratories in which to study the role of natural selection in sympatric speciation (Hewitt, 1988; Barton & Hewitt, 1989; Harrison, 1993; Arnold, 1997). They are defined as places where two or more population of individuals, distinguishable on the basis of one or more heritable characters, overlap spatially and temporally, and cross to form viable and at least partially fertile offspring (Arnold, 1997). Studies of natural selection in hybrid zones have traditionally been based on estimates of viability or fecundity (see Arnold, 1997 for a review of hybrid zones), and sexual selection has received only limited attention (Butlin, 1989; Parsons et al., 1993). However, sexual selection is also a potentially strong mechanism contributing to the maintenance of hybrid zones and to the possible evolution of reproductive isolation (Endler, 1986; Vamosi & Schluter, 1999). Sexual selection can amplify spatial patterns of variation in male traits (Lande, 1982) and can lead to reproductive isolation, either by favouring assortative mating in a Fisher runaway-type process (Lande, 1981; Turner & Burrows, 1995; Payne & Krakauer, 1997) or acting against the adult hybrids (Vamosi & Schluter, 1999), or both. Most studies of the role of sexual selection in sympatric speciation have involved laboratory studies or theoretical analysis of models, and few estimates of sexual fitness have been obtained in the field, where the evolutionary consequences of the results are clearer (Vamosi & Schluter, 1999).

The organism and the problem

Littorina saxatilis (Olivi) is an intertidal snail that shows an extensive habitat-associated morphological variation (Reid, 1996), in some cases even at a microgeographical scale (Janson, 1983; Johannesson et al., 1993; Johannesson & Johannesson, 1996). It has direct development, separate sexes and internal fertilization, and mating pairs can be found all year round (Johannesson et al., 1995). The male climbs his partner’s shell in a counter-clockwise manner before inserting his penis into the mantle cavity of his partner (Saur, 1990).

Populations of L. saxatilis on the Galician coast (NW Spain) provide an extreme example of within-shore dimorphism in the shell, anatomical and behavioural characters (Johannesson et al., 1993; Rolán-Álvarez et al., 1996, 1997). There are two very different morphs, associated with different shore levels. The smooth and unbanded (SU) morph has a small body size and inhabits the lower, more wave-exposed shore, dominated by mussels (Mytilus galloprovincialis); the ridged and banded (RB) morph is larger and inhabits the upper, more sun-exposed shore, dominated by barnacles (Chthamalus stellatus) (Johannesson et al., 1993).

In the midshore, which is covered by a mixture of mussel and barnacle patches, the two pure morphs live and reproduce sympatrically, remaining phenotypically distinct even when they occur together under the same environmental conditions. There is much additional evidence for genetic differences between the pure morphs: they ‘breed true’ in the laboratory (Johannesson et al., 1993), there is a partial genetic barrier between them, revealed by allozymic polymorphisms (Rolán-Álvarez et al., 1996), and there is also a clear genetic basis for traits that differentiate between the two morphs, such as growth rate (Johannesson et al., 1997) as well as several quantitative shell traits (Carballo et al., 2001). The pure morphs hybridize in the midshore (Rolán-Álvarez et al., 1997) where both heteromorphic mating pairs and snails with intermediate phenotypes are found in varying frequencies. However, despite the finding of heteromorphic matings, the two pure morphs mate assortatively in the midshore (Johannesson et al., 1995; Rolán-Alvarez et al., 1999), partially because of the nonrandom microdistribution over the mussel and barnacle patches, and partially because of differences in behavioural discrimination (Johannesson et al., 1995; Otero-Schmitt et al., 1997; Rolán-Alvarez et al., 1999). In spite of the distribution of morphs along the shore gradients, it is possible to observe clinal variation for many traits, even within morphs.

This hybrid zone is likely of primary origin, as, although changes in allozyme frequency show that there is some genetic isolation between the morphs on a microgeographic scale, there is no evidence of correlation between morphological and genetic differences on a broader scale, as expected if different populations from the same morph shared a common allopatric origin (Johannesson et al., 1993). Thus, the polymorphism in the Galician population of L. saxatilis probably evolved by strong differential selection between upper and lower shore. In fact, different components of natural selection have already been studied, namely: viability (Rolán-Álvarez et al., 1997), fecundity (Cruz et al., 1998), male fertility (Johannesson et al., 2000) and sexual selection (Johannesson et al., 1995; Rolán-Álvarez et al., 1995, 1999). These studies have shown that the polymorphism seems to be maintained by natural selection, as one morph has the highest viability in the upper shore and the other in the lower shore (Rolán-Álvarez et al., 1997). The role of the other fitness components in the distribution of morphs was not as clear and, in general, hybrids were not inferior. All of these studies involved comparisons between average values of morph traits and until now, none of the selection components has been studied in relation to the continuous phenotypic variation that exists between and within morphs, which is required to understand how selection models the observed clinal variability.

In this study, we have investigated the possible role of sexual selection in the maintenance of the polymorphism and in the evolution of reproductive isolation in the hybrid zone. The existence of such a role would be indicated by the inferiority of hybrids or divergent sexual selection acting on some phenotypic trait that separates the pure morphs. To study the strength and form of natural selection and to identify the traits on which it is acting, we carried out multiple regressions of an estimate of sexual selection (probability of mating) on phenotypic traits (Lande & Arnold, 1983).

Materials and methods

Sampling sites

The studied hybrid zone of L. saxatilis encompasses a 30 km stretch of coast between Baiona and A Guarda (Galicia, NW Spain). We collected samples at two sites (Centinela and Senín), where there were especially wide hybrid zones. Samplings were carried out in autumn 1995 (from September 21 to November 24) and in summer 1996 (from June 13 to July 19).

The sampling points were determined by the presence of a mating pair. We picked up the copulating pair and placed each individual in separate subdivisions of a plastic box. We also collected the 15 nearest nonmating individuals. As the population density varied over the gradient, the size of the area in which we could collect this number of individuals varied (a circle, of radius 5–20 cm in the lower or midshore and of radius up to 60 cm in the upper shore). At each sampling point, we also delimited an experimental unit of 20 × 20 cm in which to measure nine environmental variables: barnacle (C. stellatus) and mussel (M. galloprovincialis) abundance and degree of cover continuity, biological shore level (a qualitative measure of habitat zonation, based on the presence of different species of animals and plants), the proportion of the sample area that was occupied by bare rock, by crevices, or by the water of a tide pool and sample population density (see Cruz et al., 1998; for a detailed description of these measures). We also considered the geographical locality and the season as two more environmental variables to be included in the analysis.

We collected individuals in 510 sampling points. All sampled individuals were taken to the laboratory, placed in boiling water for a few seconds and then conserved at –22 °C until their dissection. In everyone of the samples, we determined the sex and morph of both mated individuals and the 15 nonmating individuals. We considered as pure morphs those snails that had their shell ridged and banded (RB morph) or smooth and unbanded (SU morph) and we considered as hybrids (HY) (as did Rolán-Álvarez et al., 1997) those snails that had a complete set of bands but lacked ridges, or vice-versa, or those that, having both ridges and bands, had at least two incomplete bands.

When both pure morphs were present in the same sample in a frequency >5%, we considered it as a midshore sample. Only such samples were used to obtain an estimate of sexual selection in every pure morph and in the hybrids (see data analysis).

Measurement of phenotypic traits

To study the sexual selection on phenotype, we dissected all mated individuals and also one male and one female sampled randomly from the nonmating individuals of each sample. Samples from the upper, RB typical shore, the lower, SU typical shore and the midshore of sympatry were used in this study. We dissected and measured phenotypic traits in 459 mated males, 442 mated females, 469 nonmating males and 454 nonmating females (the number was not exactly the same because we excluded those individuals with damaged shells).

In each individual, we counted the number of complete and incomplete ridges and bands, the characteristics used to define the morphs (see above). We measured 14 longitudinal shell traits and, in males, six penis traits using an image analysing system (PC_Image VGA 24 Version 2.1 (Foster Findlay A.L.)) connected to a binocular microscope. We also defined four shell traits and five penis traits calculated from some of the longitudinal measurements. Given that preliminary analyses had shown significant differences between morphs for all these longitudinal measurements (the RB morph showing the highest values and SU the lowest ones) and trying to facilitate a direct interpretation of the analyses results, we made all linear shell measurements (except shell height itself, of course) relative to shell height, thus converting them into more size-independent measures of shell form. This was important because some of the most remarkable (and potentially biologically significant) differences between the pure morphs, beyond those involving average size, are differences in shell form. For example, the large width and height of the aperture of the SU individuals relative to their small size (see Fig. S1 in Table S1) may promote a firm attachment in the wave-exposed sites where they inhabit (Johannesson, 1986).

Other variables, such as whole body weight, radular length, an index to measure coloration of the head and the foot, and some penis characteristics were also recorded. All measured variables are listed in Table S1. Measurements were log-transformed to enhance the normality of the distribution and the homogeneity of the variances. Weight was cubic root transformed before the log transformation. We also counted the number of embryos of normal appearance carried in the embryo brood pouch of females and used this as an estimate of their fecundity.

Many of the phenotypical variables measured (39 in males and 28 in females) were highly correlated. To reduce multicollinearity problems in the regression analyses of natural selection, and as recommended by Lande and Arnold, we made a reduction in the number of characters by dividing the list of characters in several highly correlated clusters and retaining only one character from each cluster. We, therefore, used the VARCLUS procedure (SAS, 1989), which clusters highly correlated variables and then we choose the variable showing the highest correlation with its cluster as being representative of each cluster. We made a separate selection of variables for males and females. In males, cluster formation was stopped when we had seven clusters accounting for 65% of the total variation (see Table S2). The seven variables chosen were: shell height (SM1), shell width/SM1, shell aperture height/SM1, shell second whorl diameter/SM1, penis lower perimeter/SM1, penis base-extreme length/SM1, and number of complete ridges. In females, we obtained six clusters accounting for 69% of the total variation. The chosen variables were the same shell variables than in the male case plus the number of incomplete bands.

All our analysis of sexual selection on phenotype only included these chosen variables or a linear combination of them (see below).

Data analyses

Estimates of sexual selection

To estimate sexual selection in every pure morph and in the hybrids, we compared morph frequencies in copulating and noncopulating midshore individuals, by using the cross product estimator (Spieth & Ringo, 1983; Knoppien, 1985; Partridge, 1988). We obtained an estimate of fitness for each morph relative to the SU morph, which was chosen for reference because it was the most commonly occurring morph. For instance, the fitness of the hybrids was obtained by:

inline image

where W is the fitness estimate (ranging from zero to infinity), cop is the number of copulating individuals and noncop is the number of noncopulating individuals. This was calculated separately for males and females. The significance of the estimates was obtained by simultaneously bootstrapping the morph frequencies, recorded from 1000 resamplings of copulating and noncopulating individuals, using a compiled BASIC program (for details see Johannesson et al., 1995; Rolán-Álvarez et al., 1995).

These analyses were restricted to data from the midshore samples (i.e. the zone of sympatry).

Relationship between sexual selection and phenotypic variability

The simplest measure of selection on a phenotypic trait is the selection differential which estimates the total directional selection (both direct and indirect) on that trait (Lande & Arnold, 1983; Brodie et al., 1995). We calculated the linear selection differential as the coefficient of an univariate linear regression of the fitness on everyone of the standardized phenotypic traits. Then, we calculated the nonlinear selection differential as the coefficient of an univariate quadratic regression.

To study the direct action of sexual selection on each phenotypic trait we made a multiple regression of fitness on phenotype (Lande & Arnold, 1983), in which the partial linear regression coefficients are estimates of selection gradients, indicating the intensity of the selection acting directly on each trait. Usually, linear multiple regression is used first to estimate the forces of directional selection (linear selection gradients) and then a quadratic multiple regression is made to estimate the forces of stabilizing, disruptive or correlational selection (nonlinear selection gradients) (Lande & Arnold, 1983; Brodie et al., 1995). In this quadratic multiple regression, the number of coefficients to be estimated is n (the number of characters) for linear coefficients, plus n(n + 1)/2 for nonlinear coefficients. Thus, for example in our males data the total number of the coefficients would be 35: the seven phenotypic traits and the 28 product variables (shell height × shell height, shell height × aperture height…n° complete ridges × penis base-extreme length, n° complete ridges × n° complete ridges). Endler (1986) and Mitchell-Olds & Shaw (1987) advocate careful consideration of the relation between sample size and number of characters studied, because for a given fitness-phenotypic trait covariance, the more characters considered, the more difficult it is to detect a significant selection effect. An enormous number of observations would be required to carry out a regression analysis with such a high number of independent variables included in the model (without statistical problems). Therefore, to make the study of nonlinear selection we reduced the number of independent variables by performing the regression analysis on the first few principal components of variation (PCA, PRINCOMP procedure, SAS, 1989). We obtained the first two principal components (PC1 and PC2 which accounted for 51 and 24% of morphological variance, respectively) with the log transformed phenotypic variables, and used them as independent variables in the regression model. The use of PCs may complicate the interpretation of the results, because both selected and unselected characters may be associated within a PC (Lande & Arnold, 1983; Mitchell-Olds & Shaw, 1987). However, the advantage of our PCs was that the first PC acted as a discriminant function between the pure morphs (see Results for details), allowing us to test for divergent selection on traits related with morph differentiation.

All independent variables included in the regression models were standardized to mean 0 and variance=1 to allow direct comparisons among the different data groups (Lande & Arnold, 1983).

Because our fitness measure was dichotomous (mating: 1, nonmating: 0), the assumptions required for inference in multiple linear regression were violated (Neter et al., 1990). We, therefore, applied logistic regression which relates the probability of mating for an individual to that individual’s traits as follows:

inline image

where W(z) is the probability of mating for an individual described by the column vector of traits z, αT is the vector of logistic regression coefficients and α0 is the intercept.

Then, we transformed the logistic coefficients (Janzen & Stern, 1998) so that we could consider them as selection gradients.

A common problem in multivariate selection methods is that environmental effects may generate spurious correlations between traits and fitness measures (Rausher, 1992; Van Tienderen & de Jong, 1994). Our sampling design almost completely avoided this problem, as we collected one mating and one nonmating individual at the same sampling point and therefore under the same environmental conditions (Mitchell-Olds & Shaw, 1987). However, the environment can also influence the relationship between fitness and phenotype because of the existence of phenotype–environment interactions. To test for the existence of these interactions, we repeated the logistic regression by including the independent phenotypic variables (PC1 and PC2) and their products with every environmental variable in the model. We tested for the interactions with all environmental variables simultaneously by including all the products in the same regression model.


Fitness estimates

The cross-product estimates of sexual selection in males showed that the RB males showed significantly lower fitness (W=0.62, < 0.05) than SU (W=1) and HY (W=1.07, > 0.05) males. We found a similar trend, when we analysed both seasons separately (although in summer the RBs’ inferiority was not significant, 0.05 < < 0.10). Hybrids did not show a fitness disadvantage in any season. In females, there was no significant difference in the fitness of RB and HY relative to the reference morph (RB, W=1.34, > 0.05; HY, W=1.28, > 0.05; SU, W=1); this result did not change when the seasons were analysed separately.

Female sexual selection on phenotypic variables

We did not find a clear relationship between sexual selection and phenotypic variability in females. The only variable that showed a significant relationship with the probability of mating was the aperture height. We found a significant positive linear selection differential (b=0.0718, SE=0.0170, < 0.0001) and a negative quadratic selection differential (b=–0.0258, SE=0.0126, P=0.0412) so indicating that there is an intermediate-high optimal value for this variable. The multiple regression of the probability of mating on the individual phenotypic traits suggested that selection was in fact acting directly only on this trait (selection gradient b=0.1029, SE=0.0227, < 0.0001). We did not find significant divergent or stabilizing selection on linear PC (results not shown).

As we also had an estimate of the females fecundity (the number of embryos in the inner of their embryo pouch), we explored the relationship between this fitness component and the phenotypic variables that we had included in the sexual selection study. The most remarkable result was the finding of a significant and negative effect of aperture height on fecundity (selection differential b=–37.36, SE=3.27, < 0.0001, selection gradient b=–11.75, SE=2.91, < 0.0001). Thus, natural selection through fecundity seems to be favouring females with smaller aperture height (the opposite that we found for sexual selection). Thus, this trait could be involved in a fitness trade-off, negative values tending simultaneously to increase fecundity and to decrease mating success.

Male sexual selection on phenotypic variables

Table 1 shows the selection gradients obtained from the multivariate logistic regression on standardized phenotypic traits in a joint analysis of the whole male population. We found positive selection on shell height and negative selection on shell width. A one-way analysis of variance (Table 2) revealed that the RB morph had the highest average shell height and the lowest shell width relative to shell height, indicating that sexual selection is favouring RB-typical values for these traits. However, there were also positive significant selection gradients for aperture height and penis lower perimeter, and negative ones for number of complete ridges. The SU morph showed the highest averages for the two first variables and the lowest for number of ridges (Table 2), indicating that sexual selection is favouring SU-typical values for these traits. The selection gradients could seem misleading (simultaneously favouring SU and RB typical traits), but we must take into account that these coefficients measure the direct action of selection on a single trait, holding all other constants and therefore, independently of other measures correlated traits. If we are interested in evaluating the total action of selection on a trait, selection differentials must be calculated. Thus, Table 1 also shows the linear and nonlinear selection differentials for the seven phenotypic traits. We found a total positive action of sexual selection on aperture height, penis lower perimeter and penis length and negative action on second whorl diameter. Differences between selection differentials and gradients are because of indirect selection through some correlated trait. Thus, for example, we did not find a significant directional selection differential on the shell height (SM1) because the direct selection to increase the shell aperture height (which is negatively correlated with shell height, see Table S3) produced negative indirect selection on SM1, and this could contribute to mask the direct effect of selection to increase SM1. All these linear selection differentials indicated that directional sexual selection was favouring SU males: high values for aperture height, penis lower perimeter and penis length and low values for second whorl diameter.

Table 1.   Results of logistic regression of probability of mating on phenotype. The dependent variable was ln (W(z)/(1−W(z))) and the independent variables were the seven individual phenotypic traits. Linear and nonlinear selection differentials were obtained from linear and quadratic univariate regressions on every one of the seven standardized phenotypic traits to avoid the collinearity between coefficients. Selection gradients were calculated from a multiple regression on the seven standardized phenotypic traits. The number of mating and nonmating males is given in parentheses (*except in the univariate regressions on the penis measurements, in which the number of mating and nonmating males were 456 and 464). The coefficients shown were transformed following Janzen & Stern (1998). Thumbnail image of
Table 2.   Morph average values for each phenotypic trait. We also show the F statistic and its probability obtained from a one-way analysis of variance (PROC ANOVA, SAS) for each variable by using the morph as a factor. Thumbnail image of

For both penis measurements we also found positive nonlinear selection differentials, so indicating that mated males showed intermediate-high values for penis lower perimeter and penis length.

Overall fitness surface

To study the possible nonlinear relationship between phenotype and fitness (to make the quadratic multiple regression), we used the first two PCs obtained in PCA as independent variables. The composition of these PCs and their average values for each morph are shown in Fig. 1. The highest values of PC1 corresponded to diagnostic characteristics of the RB morph, such as large shells, large second-whorl diameter, small aperture height relative to shell height and high number of ridges. Similarly, the lowest values of this PC corresponded to SU-like males. Thus, PC1 acted almost like a discriminant function between pure morphs (see Fig. 1), with hybrids having in average an intermediate value. In the second PC, the most important variables were the two penis measurements and the aperture height relative to shell height. In this case, morphs were not so clearly distinguished and the hybrids showed the highest average value. The multiple logistic regression showed a negative directional effect on PC1 and a positive directional effect on PC2 (Table 3). The negative directional effect on PC1 suggests that there was a sexual selection advantage for SU males (low PC1 values), and conversely, that the RB males (high PC1 values) were at a disadvantage. This was in accordance with the selection differentials for individual phenotypic traits (see above). Analysis of the complete model showed that the three quadratic terms were significant, and the coefficients for PC12 and PC12 (the product between PC1 and PC2) were positive and that for PC22, negative.

Figure 1.

 Mean and standard deviation for PC1 and PC2 in pure morphs and hybrids. PC loading for the seven phenotypic traits studied in males are indicated as arrows from the origin. sm1: shell height; sm2’: shell width/sm1; sm3′: shell aperture height/sm1; sm9′: shell second whorl diameter/sm1; pm2′: penis lower perimeter/sm1; pm5′: penis base-extreme length/sm1; ncr: number of complete ridges.

Table 3.   Multiple logistic regression, using ln (W(z)/(1−W(z))) as independent variable and the first two standardized PC as independent variables. Number of mating and nonmating males is given in parentheses. Coefficients were transformed following Janzen & Stern (1998). Thumbnail image of

The precise interpretation of the selection on two traits is most readily obtained through graphical representation (e.g. to find out if a positive quadratic coefficient represents a mere point of inflection in the fitness surface or a true decrease and therefore evidence of divergent selection). Figure 2 shows the fitness surface obtained by applying the quadratic model to the complete range of values of the first two PCs (each line depicting a value of expected fitness), and the place of the individuals on the fitness surface. There were two different zones of high expected fitness and each pure morph appeared to be associated with one of them: the first zone was defined by low values of PC1 and intermediate values of PC2 and the other by high values of PC1 and intermediate values of PC2. Given that PC1 acted as a discriminant function between the pure morphs, the positive quadratic coefficient and its corresponding decrease in fitness for intermediate values of this PC provided evidence of divergent selection and therefore of the role of sexual selection in the phenotypic separation of the pure morphs. This divergent selection seems actually to be the result of a combination of variables because previously we did not find any significant and positive quadratic selection differentials for individual phenotypic variables.

Figure 2.

 Overall fitness surface. From the results of logistic regression we obtained an equation to predict the probability of mating in function of the phenotype: W(z)=exp(α0 + αTz)/[1 + exp(α0 + αTz)]. By applying the quadratic model to the complete range of values of the first two PC we obtained this fitness surface. Each line depicts a line of equal expected fitness. The position of the three morphs on the adaptive landscape is also shown. SU: squares, HY: triangles, RB: circles. Filled symbols: mating males; unfilled symbols: nonmating males.

The positive directional and stabilizing sexual selection found for PC2 seems to be related to penis lower perimeter and penis base-extreme length, as these two variables showed significant and positive linear and quadratic selection differentials (Table 1). Moreover, both penis variables had the two most important loadings in PC2, so that we may consider this PC as representing ‘penis characteristics’. The overall fitness surface suggested that there was a more or less intermediate optimal value for PC2, which was somewhat higher for high values of PC1 than for low values of PC1, because of the significant interaction indicated by PC12 (Fig. 2). This PC2 appeared to be related to mating success in a similar way in both pure morphs (results not shown) and in the hybrids (see below).

The importance of the environmental interactions was, in general, minimum (results not shown). We only found some interaction environmental variable – PC2, which corresponded to changes in magnitude but not in sign in the effect of this PC on fitness in different environmental conditions, and in any case the nonlinear effects that we found on the PCs were the result of different linear effects in different environments.

Fitness surface of hybrids

The hybrids are obviously the most interesting group of individuals within a hybrid zone, therefore, we analysed the hybrids separately, but in the same way as in the joint analysis (Table 3). The fitness surface of the hybrids (Fig. 3) was very similar to that of the whole population. We again found divergent selection on PC1 and stabilizing selection on PC2.

Figure 3.

 Fitness surface of hybrids. The fitness surface obtained by applying the hybrid quadratic model to the complete range of values of the first two PC. Each line depicts a line of equal expected fitness (probability of mating). Filled triangles: hybrid mating males; unfilled triangles: hybrid nonmating males.

The divergent selection found for PC1 was especially interesting in the case of hybrids, because it suggests that two different classes of hybrids are favoured by sexual selection: those phenotypically resembling the SU morph and those resembling the RB morph. This effect may be because of the differential adaptations to different environments by these hybrids, i.e. the SU-like hybrids may be favoured in an environment more typical of SU morph. In this group, we could not study the existence of interactions with individual environmental variables because of the small number of observations. Therefore, we studied that possibility by using the ‘RB morph frequency’ variable (MOFR) in the sample as a measure of the proximity of each pure morph to its typical environment. The repetition of the logistic regression in the hybrids, including PC1, PC2 and the products of PC × MOFR in the model, revealed a significant interaction between PC1 and MOFR (the regression coefficient for this product was positive and significant: b=1.02, < 0.05). Thus, SU-like hybrids were favoured in samples with low values of MOFR (in the low midshore, near the lower shore), and hybrids more similar to RB were favoured in samples that showed high values of MOFR (in the high midshore, near the upper shore).

The divergent selection on PC1 may also be associated, more or less directly, with the mating behaviour in the midshore; in this sympatric zone there is assortative mating within the pure morphs, as previously explained (65% of mating pairs in the midshore were homomorphic) but not within the hybrid group. The divergent selection for PC1 in the HY could be related to this mating behaviour if the individuals with low PC1 (i.e. more similar to SU morphs) were preferentially mating with SU females and those with high values of PC1 (more similar to RB morphs) were preferentially mating with RB females. We, therefore, analysed the differences in PC between the nonmating HY males and the HY males mating with females of different morphs, separately (Table 4). Males that were mating with HY or SU females had a significant negative effect of the PC1, favouring the ‘more SU type’ individuals. The HY males that were copulating with RB females had a positive, but not significant, effect on PC1 (to be more similar to RB it was not important to be mated with them). These results suggest that there may be different hybrid phenotypes, which were favoured in different zones within the midshore (although we failed to identify a physical environmental variable responsible for this effect), and/or in different mating classes.

Table 4.   Logistic regression, using different subgroups of hybrid mated males (those mating with SU, HY or RB females). In each case, the dependent variable was ln (W(z)/(1−W(z))) and the independent variables were the first two standardized PC. The number of mating males is given in parentheses; in the three cases all the 27 nonmating hybrid males were used in the analyses. Thumbnail image of

A more complete fitness estimate

The SU males showed a higher probability of mating than RB males and would therefore be at an advantage, in terms of fitness, in the midshore. However, this advantage could be compensated by the fact that RB males are more likely to mate with RB females, which are, on average more fecund; thus, although the RB males mate less frequently than SU males, they may produce more descendants. The fecundity of the mated female could be considered as another estimate of male fitness: the ‘quality of the mate’ obtained by the male. As we dissected both male and female mating individuals, we were able to use the number of embryos carried by the female as an estimate of its fecundity, and thus obtain a more complete measure of male fitness that combined the probability of mating and mate quality. The combination of both fitness estimates may be important in the evolutionary dynamics of the hybrid zone; although the RB males mate less frequently than SU males, they may produce more descendants. To analyse this possibility, we multiplied the two fitness components (Arnold & Wade, 1984a,b), i.e. the probability of mating of males and the fecundity of their partners. Regression of this new fitness estimate on the phenotype of the males (the PCs of Table 2) revealed positive directional selection on both PCs, divergent selection on PC1, positive correlational selection on PC1 and PC2 (coefficient for PC12 significant and positive) and stabilizing selection on PC2. Figure 4 shows the best quadratic surface of this combined male fitness estimate. The fitness surface was very similar to that of the analysis of success mating but in this case the decrease in fitness was closer to the SU morph PC1 average.

Figure 4.

 The best quadratic surface of the ‘more complete male fitness estimate’ (estimated female fecundity × probability of mating). The fitness surface was obtained by applying the quadratic model to the complete range of values of the first two PC. Each line depicts a line of equal expected fitness. The position of the morphs is the same as in Fig. 2.


Female sexual selection

All females examined (a total of 896) had embryos in their embryo pouches, indicating a low frequency of nonmated females in the population. Therefore, any differences between females sampled while mating and those sampled while not mating should be interpreted as being associated with mating frequency and not with reproductive success (given that females of both kinds carry similar numbers of viable embryos). The evolutionary consequences of differences in female mating frequency in this species is not clear at present, as there is no evidence of sperm competition, sperm scarcity or other processes that may be relevant in this context. In fact, we found no differences in estimates of female fitness between morphs when we applied the cross-product estimator. Moreover, the only variable that showed a significant relationship with the probability of mating was the aperture height and the relationship of this variable with other fitness component (fecundity) is the opposite: the most fecund females show a smaller aperture height (Grahame & Mill, 1992; this study). We think that possibly this other fitness component is more relevant to the evolution of the female phenotype because the relative aperture is one of the few traits in which females are smaller than the males (in all the morphs).

Male sexual selection

The main aim of this study was to investigate the role of sexual selection in the maintenance and evolution of the polymorphism in the Galician hybrid zone of L. saxatilis. The cross-product estimates of average male fitness in each morph were similar to those from previous studies (Rolán-Alvarez et al., 1999). These estimates did not show any decrease in fitness in hybrids, and therefore did not suggest that this fitness component had any role in maintaining the natural polymorphism on the shore. However, the conclusion was different when, instead of considering the three morphs as discrete classes, we studied the effect of the continuous morphological variation across this hybrid zone. We then found divergent sexual selection for the first PC, which acted as a discriminant function between the pure morphs. Moreover, divergent selection on PC1 was still found when we used a combination of the probability of mating with the expected number of descendants obtained from the mating as our measure of fitness. Thus, phenotypically intermediate males mated less frequently and their female mates tended to have lower than average fecundity. This supports the idea of a role of male reproductive success in morphological differentiation between pure morphs. Divergent sexual selection for size also partly explained the size differences existing between two Swedish ecotypes of L. saxatilis (Erlandsson & Rolán-Alvarez, 1998). Thus, divergent sexual selection may be common in natural populations of this species in which two different ecotypes overlap. This finding is especially interesting because quadratic selection gradients are typically quite weak and frequently nonsignificant (Kingsolver et al., 2001), and there have been few empirical studies investigating whether such a type of selection operates in wild populations (Hatfield & Schluter, 1996), in spite of the existence of theoretical studies showing that divergent sexual selection can cause speciation (e.g. Lande, 1981, 1982; Schluter & Price, 1993; Higashi et al., 1999), and therefore, that it may also be an important selection component in the dynamics of hybrid zones.

Given that it was not easy to separate the effect of phenotypic differentiation between morphs and that of environmental segregation in the intertidal zone, as is often the case in these hybrid zones associated with environmental gradients, there may be different explanations for the observed divergent action of sexual selection. First, the fitness advantage we found for both pure morphs may be independent of the environment that they occupy and, therefore, sexual selection may be caused by some kind of female choice. However, this is unlikely, because the mating behaviour of L. saxatilis seems to be rather simple. Males attempt to copulate with any female (or even with other males) and are only rejected by females in extreme circumstances (Saur, 1990). Although assortative mating for size or morph has been observed in different L. saxatilis populations (Saur, 1990; Hull et al., 1996; Rolán-Alvarez et al., 1999), it is supposed to be caused by a physical incompatibility between male–female reproductive organs (R. Cruz, M. Carballo, & E. Rolán-Álvarez, unpublished work). Secondly, the environment may be important in influencing sexual selection through male competition. Males that resemble any of the pure morphs may have a fitness advantage in the typical environment of that morph, and a disadvantage in the alternative environment. This would be not surprising, as it is already known that environmental factors that differentiate the upper and lower shore, such as exposure to sunshine and waves, fresh water submergence and predation by crabs, differentially affect the viability of the pure morphs, each morph being adapted to its own environment (Rolán-Álvarez et al., 1997). The situation in this hybrid zone would therefore be in accordance with that described in most models of sympatric speciation (Tauber & Tauber, 1977; Kondrashov & Mina, 1986) and with the more general model of ‘ecological speciation’, in which two groups of individuals are climbing different adaptive peaks that correspond to distinct ecological niches (Endler, 1977; Rice & Hostert, 1993; Schluter, 1998). The stability of a polymorphism can be increased by several factors, such as habitat preference, habitat-specific survival rates or habitat-specific consumption rates (Diehl & Bush, 1989; Johnson & Gullberg, 1998). The existence of a habitat-specific advantage in probability of mating may also contribute to the stability of the polymorphism. Future reciprocal transplants along the intertidal gradient will allow us to ascertain the role of environmental variation in this hybrid zone.

Our results indicate that sexual selection on males does contribute at least to the first step of sympatric speciation: the differentiation between the two ecotypes. Sexual selection may also be involved in the evolution of reproductive isolation by acting against the hybrids (post-mating isolation). Even if there is no evolutionary response to this hybrid inferiority (i.e. reinforcement), a mating disadvantage may be a part of the total selection against hybrids, and thus important in its own right in limiting gene flow (Hatfield & Schluter, 1996). Studies of hybrid fitness are typically carried out in the laboratory, but many recent studies (Grant & Grant, 1992, 1996; Craig et al., 1997; Feder, 1998) have emphasized the need to measure fitness of hybrids in their natural environment, where an intermediate phenotype may experience ecological disadvantages (Hatfield & Schluter, 1999). We estimated the fitness of hybrids under natural conditions and found that hybrids were not less fit in terms of probability of mating. This is surprising because if divergent selection favours the pure morphs, and given that the hybrids are on average phenotypically intermediate, we would expect them to have, on average, less success in mating (Vamosi & Schluter, 1999). However, we did not find this because the hybrids were not uniformly intermediate in PC1. In fact, examination of the hybrid fitness surface suggested that the more intermediate hybrids seemed to be at a disadvantage, whereas hybrids with PC1 values similar to those of the pure morphs were favoured by sexual selection. Moreover, we must take into account that these intermediate hybrids were not completely unfit because of the high value of PC2 (the optimal value was close to the hybrid mean value).

Genetic or ecological causes (or a combination of both) may explain differences in hybrid fitness. Hatfield & Schluter (1999) point out that, if there is a genetic breakdown of favourable gene combinations (genetic causes for inferiority of hybrids) this is usually more evident in backcrosses than in F1 hybrids, whereas the ecological model predicts that individuals that phenotypically resemble the pure morphs should have higher fitness in the wild than F1 hybrids (which would be phenotypically intermediate). Our findings appear to be similar to the latter situation, with the lower probability of mating of the phenotypically intermediate hybrids depending on ecological factors. The importance of these kinds of factors was also suggested when we studied the possible relationship of the variation in fitness of the hybrids with environmental factors. The fitness advantage of those individuals that phenotypically resembled the pure morphs was related to the variation in morph frequency and to the morph type of the mated female. This may indicate that the favoured male type varies with the morph composition of the population or with the position of the hybrids along the midshore. In fact, pure morphs are not randomly distributed in the midshore (Erlandsson et al., 1999), as they have somewhat different microhabitat preferences (Kostylev et al., 1997; Otero-Schmitt et al., 1997). Therefore, hybrids phenotypically resembling the pure morphs may be at an advantage because they have the appropriate phenotype (to compete with other males or to be chosen by a female) or because they are in the appropriate habitat, or both.

In conclusion, the overall fitness surface and the separate analysis for hybrids showed that there are some phenotypic characteristics (PC2, penis measurements), which distinguished between mating and nonmating males in a similar way in the three morphs, whereas another combination of phenotypic traits (PC1) distinguished between two groups of mating males, each phenotypically similar to the pure morphs. This indicates that, in spite of the lack of a clear sexual disadvantage of the hybrids, sexual selection is at least partially involved in the maintenance of this hybrid zone by contributing to morph differentiation.


This work was funded by the Spanish Dirección General de Investigación Científica y Técnica, grant no. PB94-0649. E. R. A. also thanks the Xunta de Galicia for grant no. XUGA 30105B98 and the Universidad de Vigo for grant no. 64502C925.

Supplementary material

The following material is available from

Figure S1 Shell measurements.

Table S1 Phenotypic measurements (Fig. S1) of the shell of the pure morphs and hybrids showing on the RB morph the shell measurements taken.

Table S2 Outcome of the VARCLUS procedure in the males. All longitudinal measurements were expressed relative to sml (see text). In each cluster, we choose as representative the variable that showed the highest correlation with its own cluster (in bold) except in the second cluster where the chosen variable was not second whorl breadth (that showing the highest correlation) but second whorl diameter (the second higher correlation) because both variables were very similar measurements (see Fig. S1) and second whorl diameter was measured with more precision (R. C. Guerrero, pers. Obs.).

Table S3 Pearson correlation coefficients (and probability) among the seven phenotypic variables chosen in the VARCLUS in the males.