The close relationship between estimated divergent selection and observed differentiation supports the selective origin of a marine snail hybrid zone

Authors


Carlos García, Departamento de Xenética, Facultade de Bioloxía, Universidade de Santiago de Compostela, 15782 Santiago de Compostela, Galicia, Spain.
Tel.: +34 981 563100, ext: 13259; fax: +34 981 596904;
e-mail: bfcarlog@usc.es

Abstract

To study the role of divergent selection in the differentiation of the two morphs in a hybrid zone of the intertidal snail Littorina saxatilis, we compared the strength of the divergent selection acting on a series of shell characters (as estimated by the viability of snails in a reciprocal transplant experiment) with the contribution of these characters to the phenotypic differences between the morphs. We found a close correlation between selection and differentiation, which suggests a cause–effect relationship, i.e. that all present differentiation is the result of past divergent selection. In addition, divergent selection was a very important component of the total natural selection acting on shell measures. These novel results support previous evidence, based on allozyme analysis, of a parapatric origin for this hybrid zone. We discuss possible limitations of this interpretation and the circumstances under which allopatric differentiation would produce the same results. Phenotypic analysis of divergent selection may be a useful method of investigating the evolutionary mechanisms involved in differentiation processes.

Introduction

The ecological hypothesis of speciation is that reproductive isolation ultimately evolves as a consequence of divergent selection acting on subpopulations or populations living in different environments (Schluter, 2001; Via, 2002). Field studies using tests able to distinguish divergent selection from other possible causes of differentiation and speciation, such as genetic drift, have shown that ecological speciation may be an important evolutionary process (Schluter, 2001). These tests are based on correlations of the strength of divergent selection with the degree of reproductive isolation between populations (Fitzpatrick, 2002; Nosil et al., 2002), or with the known causes of this isolation (Macnair & Christie, 1983; Podos, 2001); on selective reductions of hybrids’ fitness (Mallet et al., 1998; Hatfield & Schluter, 1999), or on finding cases of parallel evolution of mating incompatibilities over similar environmental gradients (reviewed in Johannesson, 2001). However, none of these studies alone provide entirely conclusive evidence, and the joint results of different tests may be required to clarify the role of ecological speciation. Some of the tests are also difficult to carry out in many situations, as they require a series of paired populations to make comparisons, or knowledge of the causes of the reproductive isolation. For these reasons, new tests based on alternative sources of evidence may be very useful in the study of the processes of differentiation of populations. In the present study we show that determining the correlation between the pressure of the divergent selection acting on a series of characters and the contribution of the characters to the observed phenotypic differentiation may provide one of these tests. If the correlation is high, we can assume a cause–effect relationship, and that divergent selection is the cause of the observed differentiation. Characters would have diverged to an extent dependent on the strength of the selection acting on them (Kondrashov & Kondrashov, 1999). This test of ecological speciation does not require a series of paired populations, a single pair being sufficient. Clearly, estimation of the action of divergent selection on a series of individual traits may also serve to determine which characters are the targets of selection, thus providing a more profound and functional understanding of the mechanisms involved in adaptive differentiation.

Evidence supporting divergent selection as the cause of the observed differentiation does not enable us to determine if the differentiation took place in allopatry (i.e. in geographically separated populations), in sympatry or parapatry (in geographically overlapping or adjacent populations respectively), which is another important point in understanding differentiation processes in nature (Turelli et al., 2001; Via, 2001). However, measuring the importance of divergent selection relative to selection acting in parallel in all populations could help to assess the likelihood of nonallopatric differentiation. In this kind of differentiation, divergent selection must be strong enough to overcome the homogenizing effect of gene flow between subpopulations (Felsenstein, 1981; Coyne, 1992), and therefore it must be an important component of total natural selection. This restriction would disappear in situations of allopatry, in which even weak divergent selection would result in differentiation. Thus, finding divergent selection to be an important component of total natural selection would be consistent with nonallopatric differentiation.

We carried out a reciprocal transplant experiment to study the relationship between the strength of divergent selection and the observed phenotypic differentiation, as well as the relative importance of divergent selection, in a hybrid zone of the intertidal marine snail Littorina saxatilis (Olivi), and used the results to discuss the conditions under which the observed differentiation developed.

The Galician Littorina saxatilis hybrid zone

Littorina saxatilis has direct development, and does not produce pelagic larvae (Reid, 1996), which may reduce dispersal and thus increase polymorphism and speciation rates (Gavrilets et al., 2000). Two genetically distinct morphs of this species are found on the exposed shores of Galicia (NW Spain). The first lives on the upper shore and has a ridged and banded shell; the second lives on the lower shore and has a smooth un-banded shell. The morphs also differ in size and in overall shell shape. A detailed description of the hybrid zone is provided by Johannesson et al. (1993). Divergent selection is involved in the maintenance of the differentiation between morphs, as each morph is more viable when living in its usual habitat (Rolán-Alvarez et al., 1997). The hybrid zone is located in the environmental transition of the mid-shore, where both heteromorphic mating pairs and phenotypically intermediate individuals, showing a range of intermediately shaped and incompletely ridged or banded shells, are found. It is stabilized by prezygotic restrictions to gene flow, such as incomplete assortative mating between morphs (Rolán-Alvarez et al., 1999; Cruz et al., 2004), and also by post-zygotic restrictions, such as reduced male mating success (Cruz et al., 2001), embryo size and female fecundity (Cruz & García, 2001), for the most phenotypically intermediate. The corresponding reduction in fitness may be related to a disrupted pattern of response of these intermediate individuals to environmental variation (Cruz & García, 2003). Sharp changes in allozyme frequency across the hybrid zone indicate some degree of genetic isolation on a microgeographic scale, but on a wider scale there is no evidence for parallel genetic divergence between morphs, as expected if different populations of the same morph shared a common allopatric origin (Johannesson et al., 1993). This suggests that the hybrid zone resulted from parapatric differentiation, rather than from secondary contact between two allopatrically differentiated populations, and is thus one of the few existing examples of such a situation (Jiggins & Mallet, 2000).

Materials and methods

Sampling

The populations studied were located in Corrubedo (42°32′N, 9°2′W) and Portecelo (41°59′N, 8°53′W), on the western coast of Galicia. The distance between the sites is 63 km, and they are separated by long tracts of protected shore in which at least the lower shore morph is absent. Experiments were carried out in the summer of 2001.

At each location, we marked five sampling points on the rock surface in a linear transect perpendicular to the shoreline. Each of these points (designated from one to five) corresponded to different environments: lower shore; lower shore–mid-shore border (where the frequency of typical upper shore phenotypes became approximately <5%), mid-shore (where the frequency of upper and lower shore typical phenotypes was approximately equal), mid-shore–upper shore border (where the frequency of typical lower shore phenotypes became approximately <5%), and upper shore. The distances (in m) between the points were 3.5, 3.5, 4.4 and 4.8 in Corrubedo and 3.5, 1.8, 3.9 and 5.9 in Portecelo. The transects covered most of the width of the shore band occupied by the lower shore morph, but only a section of the wider band occupied by the upper shore morph. We collected 125 adult individuals at random, at each point. The size of the area sampled depended on the local population density – the maximum width was 1 m in a vertical direction and 2 m in a horizontal direction (relative to the shore line); the corresponding area was much smaller when the population was dense, as in the lower shore of Portecelo. We transported the snails to the laboratory in wet receptacles, took individual digital photographs of them using a video camera, and marked them individually by attaching a small (approximately 2.5 × 2.5 mm), white paper label to their shells, with fast-acting, waterproof glue; each label was marked (using a laser printer and font size 4) with a different three-digit number. Once dampened by the glue, the labels became rather inconspicuous, and also adapted to the surface curvature of the shell. We maintained the snails at 15 °C overnight and returned them to their sampling locations the following day. All individuals survived this treatment and were reintroduced to the field alive.

Each of the five samples was divided at random into five groups of 25 individuals, and each group was transplanted to one of the five sampling points, which then became release points. We therefore had equal-composition mixtures of individuals from every sampling point considered in the experiment, at each release point, and were able to estimate the viability of similar ranges of phenotypes when exposed to the different environmental conditions encountered on the shore. Transplants were made when the release points were above the water level at low tide; the transplanted individuals were dispersed along bands, parallel to the shoreline, of no longer than 30 cm and no wider than 5 cm. The resulting population density in the release areas was high, but not too extreme for this species. Similar or even closer inter-individual distances are observed in dense natural populations on the lower shore or in crevice refuges on the upper shore. To stimulate the individuals to become active as soon as possible and to attach themselves firmly to the rock surface, we sprinkled seawater on the animals just after release. We made sure that all snails were attached before the tide rose.

We recaptured the marked snails 1 week later. We used the capture–recapture data to estimate snail viability. We considered recaptured snails as alive and nonrecaptured ones as dead. To use only the data from individuals likely to have been exposed to their release environments throughout the experiment, we excluded those snails recaptured outside their release shore levels i.e. those found crossing the lower shore-mid-shore or the upper shore-mid shore border (eight individuals in Corrubedo and six in Portecelo), as well as those individuals recaptured with water-damaged and unreadable labels (12 in Corrubedo and 17 in Portecelo), from the logistic analysis. We therefore included in our analyses 96 recaptured individuals and 509 nonrecaptured individuals, for the Corrubedo site. The corresponding numbers for the Portecelo site were 130 and 472, respectively. The reduction in data did not alter the conclusions obtained in the analyses.

Measured characters

Using the stored digital images of individuals and an image analyser (PC_Image VGA 24 Version 2.1; Foster Findlay Associates, Newcastle upon Tyne, UK), we made seven linear measurements of the shell of every snail (Fig. 1). All measurements were made relative to shell height (except shell height itself), and were logarithmically transformed. We also recorded the number of bands and ridges on the shells of all individuals that were not ‘perfect’ pure morphs; i.e. only partially ridged or banded.

Figure 1.

Linear measures of shell shape made on upper (right) and lower (left) shore individuals. Shell height (SH), shell aperture height (SAH), shell aperture width (SAW), first whorl width (W1W), second whorl width (W2W), first whorl height (W1H) and shell width (SW). The same measures are indicated on the lower shore snail. The position of SAW is different because of differences in aperture shape. To represent the shells using the same scale, we took a below-average sized upper shore snail and an above-average lower shore snail.

Data analysis

Given that there were only two possible values for viability (1 = recaptured, 0 = not recaptured), we used logistic regressions to study the effect of every shell measure on the viability estimates (Brodie & Janzen, 1996). We analysed each measure separately, using a regression model that included as regressors the measurement, the release point (both standardized), their product and their squares. The measurement enabled us to estimate selection gradients; the release point, environmental effects; the squared terms, nonlinear effects of the phenotype or the release point, and finally, the product term made possible to test for differences in habitat-specific selection gradients. For example, if selection favoured high values for a given measurement at high release points (upper shore) and low values at low points (lower shore), then individuals with positive products would be highly viable, whereas negative products would correspond to the less viable combinations, i.e. high phenotypic values on the lower shore and low phenotypic values on the upper shore. The regression coefficient of viability on the product of phenotype × release point estimated the sign and strength of divergent selection. The standardization of shell measurements made it possible to compare directly the regression slopes obtained for different variables. To avoid difficulties associated with collinearity, linear effects were estimated in simple models that did not include the quadratic terms (Lande & Arnold, 1983).

We estimated the across-shore phenotypic differentiation in every shell measurement as the standardized difference between the means of the two extreme original samples, and compared these estimates with those for the strength of divergent selection. A positive relationship between the two estimates would indicate that selection was reinforcing the observed differentiation. One limitation of this comparison was that the shell measurements were correlated, so that they did not provide seven independent observations for studying the relationship between selection and differentiation. As an alternative method of analysis, we applied principal component analysis to the data for each location to summarize the observed morphological variation, and repeated the above comparison using the principal components obtained. This eliminated any dependence among characters, as the correlation between principal components is zero.

Effects of shell shape vs. ridges and bands

The presence of shell ridges and bands is the most conspicuous difference between the two morphs in this hybrid zone. We did not use these variables in the global analyses of viability because their variation was very discontinuous (most snails were either entirely ridged and banded or entirely smooth and un-banded). However, we attempted to determine if shell pattern (i.e. ridges and bands) rather than shell shape was the direct target of divergent selection by carrying out separate logistic regression analyses for the viability of the partially ridged (44 individuals in Corrubedo and 83 in Portecelo) and for that of the partially banded (181 individuals in Corrubedo and 225 in Portecelo) snails. Numbers of ridges and bands showed more continuous variation in these data subsets. The regression model for the partially ridged individuals included as independent variables the release point, the number of ridges, the first principal component of shell shape obtained above (all standardized) and their product terms. In the regression model for the partially banded individuals, we used the number of bands instead of the number of ridges.

Results

We show the distributions of the shell measures across the five sampled shore levels in Fig. 2, and the estimates of the degree of differentiation in Table 1. The logistic regression of viability on phenotype and release point revealed significant effects for the product terms, and therefore evidence for divergent selection, for several traits (Table 2). Graphical representations of predicted viability (not shown) confirmed that all significant product terms in Table 2 corresponded to situations of divergent selection. Of the remaining independent variables in the regression model, release point was often significant, but the squared terms were not (not shown). Divergent selection was an important component of overall selection. Of the 13 significant regression coefficients shown (Table 2), four corresponded to directional selection and nine to divergent selection. Divergent selection was clearly dominant in Portecelo.

Figure 2.

Phenotypic distribution in the five sampled shore levels, from lowest (left) to highest (right), for each shell measure. The box plots for each sample show the 10, 25, 50, 75 and 90 percentiles. The left axis represents measures in millimetres and corresponds to shell height, whereas the right axis represents proportions of shell height and corresponds to the remaining six shell measures.

Table 1.  Mean values (in millimetres for SH and as proportions of SH for the remaining variables) at the two extreme sampling points. The standardized differences (STD) given are those actually used in the analyses, and correspond to logarithmically transformed variables. Each mean is based on 125 observations.
LocationSHSAHSAWW1WW2WW1HSW
  1. SH, shell height; SAH, shell aperture height; SAW, shell aperture width; W1W, first whorl width; W2W, second whorl width; W1H, first whorl height; SW, shell width.

Corrubedo
 Highest shore7.1750.6750.5140.7180.3190.1720.977
 STD1.760−1.000−1.1810.8951.376−0.558−1.376
 Lowest shore4.0920.7240.5730.6860.2080.1911.024
Portecelo
 Highest shore7.0840.6600.5260.7260.3180.1640.985
 STD1.587−1.292−1.0941.3961.404−0.342−1.198
 Lowest shore4.2700.7420.5760.6570.1990.1781.021
Table 2.  Logistic regression coefficients of viability on shell shape measures (bm), which provide a measure of directional selection, and on the product between measure × release point (bmr), which provide a measure of divergent selection.
LocationSHSAHSAWW1WW2WW1HSW
  1. *P < 0.05, **P < 0.01, ***P < 0.001.

  2. SH, shell height; SAH, shell aperture height; SAW, shell aperture width; W1W, first whorl width; W2W, second whorl width; W1H, first whorl height; SW, shell width.

Corrubedo
 bm0.454***−0.186−0.358***0.1200.395***0.015−0.374***
 bmr0.341*−0.109−0.2250.375*0.322−0.210−0.288*
Portecelo
 bm−0.0530.1400.141−0.025−0.0760.0870.136
 bmr0.463***−0.289***−0.286***0.269***0.534***−0.062−0.310***

When we compared this divergent selection with the differentiation measurements shown in Table 1, we found a close relationship; divergent selection had a stronger effect on characters that contributed most to differentiation between morphs (Fig. 3). Thus, divergent selection reinforced the differentiation observed at present. The similarity of results from the two locations was remarkable.

Figure 3.

Relationship between strength of divergent selection, measured as the regression coefficient of viability on the product shell measure × release point, and differentiation across seven measures of shell shape. rS, Spearman's correlation.

Although the correlations between shell measures were not very high (Table 3), we condensed the shell shape variation into principal components to study further the divergent selection. We used three principal components in each location; each component accounting for at least 10% of the total variation in shell shape (Table 4). Their structures were very similar at both locations. We followed the same procedures as described above to estimate selection (Table 5) and differentiation for these three independent measures of shell shape and again found a close relationship between them (Fig. 4). Results were very similar when using the first seven principal components (data not shown). This supported the previous observation that morphological differentiation was proportional to the strength of the divergent selection acting in this hybrid zone. Again the results for the two locations were very similar.

Table 3.  Phenotypic correlations between shell measures for Corrubedo (above the main diagonal) and Portecelo (below the main diagonal). Each value is based on 625 observations. The two matrices were very similar (Mantel test R = 0.975, 10 000 replicates, P < 0.001).
 SHSAHSAWW1WW2WW1HSW
  1. *P < 0.05, **P < 0.01, ***P < 0.001.

  2. SH, shell height; SAH, shell aperture height; SAW, shell aperture width; W1W, first whorl width; W2W, second whorl width; W1H, first whorl height; SW, shell width.

SH −0.420***−0.427***0.288***0.470***−0.136***−0.467***
SAH−0.463*** 0.523***−0.197***−0.358***−0.409***0.411***
SAW−0.461***0.619*** −0.043−0.335***−0.177***0.685***
W1W0.364***−0.542***−0.163*** 0.288***−0.044−0.069
W2W0.455***−0.602***−0.462***0.551*** −0.308***−0.469***
W1H−0.013−0.219***−0.317***0.012−0.201*** 0.180***
SW−0.400***0.596***0.717***−0.256***−0.556***0.088 
Table 4.  Eigenvectors including the coefficients of the first three principal components.
LocationSHSAHSAWW1WW2WW1HSWCPV
  1. SH, shell height; SAH, shell aperture height; SAW, shell aperture width; W1W, first whorl width; W2W, second whorl width; W1H, first whorl height; SW, shell width; CPV, cumulative percentage of variance accounted for.

Corrubedo
 PC10.438−0.415−0.4540.1980.412−0.015−0.4680.416
 PC2−0.147−0.416−0.252−0.182−0.3390.7670.0750.626
 PC3−0.1230.181−0.306−0.809−0.083−0.216−0.3880.777
Portecelo
 PC10.365−0.462−0.4200.3270.4290.045−0.4260.492
 PC2−0.078−0.130−0.384−0.261−0.3320.8070.0040.673
 PC3−0.0120.184−0.343−0.692−0.039−0.372−0.4790.806
Table 5.  Logistic regression of viability on phenotypic principal components and release point. We show the regression coefficients for principal components (bp) and for the products of principal component × release point (bpr).
LocationPC1PC2PC3
  1. *P < 0.05, **P < 0.01, ***P < 0.001.

Corrubedo
 bp0.418***0.0210.058
 bpr0.380***−0.243−0.186
Portecelo
 bp−0.0730.006−0.114
 bpr0.447***−0.150−0.019
Figure 4.

Relationship between strength of divergent selection, measured as the regression coefficient of viability on the product shell measure × release point, and differentiation across the first three principal components of shell shape. Parametric Pearson's correlations (rP) are given because we could not apply significance tests for Spearman's correlations based on only three observations.

Most of the observed divergent selection acted on the first principal component, PC1 (Table 5 and Fig. 4), which contrasted shell height and second whorl width with aperture height, aperture width and shell width (Table 4). Comparison of PC1 with the mean values for characters (Table 1) and with the shell shapes shown in Fig. 1 indicated that it was very similar to a discriminant function between morphs (in fact the Pearson correlation of PC1 with a discriminant function calculated to separate the snails taken at the extreme sampling points 1 and 5 was 0.866, n = 625, P < 0.001 for Corrubedo, and 0.838, n = 625, P < 0.001 for Portecelo). This further emphasized the relationship between divergent selection and differentiation.

In the analysis of the effects of PC1 and number of ridges on viability, only PC1 was significantly related to divergent selection in Portecelo (bPC1×release point = 0.887, P < 0.05; bridges×release point = 0.328, n.s.) whereas no significant effects were detected in Corrubedo (bPC1×release point = 0.311, n.s.; bridges×release point = 0.107, n.s.). Results were similar for the regression model including PC1 and number of bands. We found divergent selection for PC1 (bPC1×release point = 0.476, P < 0.05) but not for number of bands (bbands×release point = 0.412, P = 0.056) in the Portecelo snails, and no significant effects in the Corrubedo snails (bPC1×release point = 0.007, n.s.; bbands×release point = 0.407, n.s.). Thus, we had no clear evidence that divergent selection acting on ridges or on bands was stronger and could explain that acting on shell shape. The correlations between PC1 and numbers of ridges and bands were not high (Table 6), which suggested that their genetic basis may differ to some extent.

Table 6.  Phenotypic correlations between the first principal component, the number of ridges and the number of bands, in snails from Corrubedo (above the main diagonal) and in snails from Portecelo (below the main diagonal). All individuals used for these calculations were partially ridged or partially banded (n = 181 in Corrubedo and n = 228 in Portecelo). All estimates were significant (P < 0.005).
 PC1RidgesBands
PC1 0.4960.224
Ridges0.503 0.281
Bands0.4720.547 

Discussion

The differential adaptation of the two morphs to their respective habitats, and therefore the operation of divergent selection in this hybrid zone, was already well known (Rolán-Alvarez et al., 1997; Cruz et al., 2001), but the adaptive value of the different aspects of the morphological differentiation had not been studied. In this experiment, we found not only that selection did not act to the same degree in all shell shape characters, but also that there was a close correspondence between this degree and the contribution of the character to the differentiation between morphs. This suggested that the observed differentiation was the outcome of divergent natural selection pressures similar to those acting at present, rather than drift or founder events, and that this selection may account for all the observed differentiation. Such a cause–effect relationship between selection and differentiation would require that the relation between divergent selection and observed response (i.e. the heritability) is similar for all characters studied. The estimates available for the heritability of different shell shape characters in Littorina (Boulding & Hay, 1993; Carballo et al., 2001) are not markedly heterogeneous, and would thus not generate large disruptions in the relationship between selection and differentiation. However, the corresponding experiments were not large enough to allow a detailed comparison between estimates.

Although our results clearly provide evidence of divergent selection as the cause of the observed differentiation, therefore favouring ecological speciation, they are compatible with both allopatric and nonallopatric differentiation. Although the situation is parapatric at present, we cannot discount the possibility that the observed differentiation occurred during a period of allopatry, the present parapatry resulting from secondary contact between already differentiated populations. Despite the fact that allopatric differentiation may result from random processes such as genetic drift, and not only from divergent selection, we would still find a correlation between divergent selection and differentiation after secondary contact, because gene flow would erode all genetic differences except those involved in the differential adaptations of the populations (Barton & Bengtsson, 1986; Kim & Rieseberg, 1999). Such a situation of differential introgression, involving AFLP markers instead of shell shape characters, has been described for a Littorina hybrid zone in eastern England (Wilding et al., 2001). However, the opportunities for spatial isolation and allopatric differentiation seem limited in the case of the lower shore morph. Subpopulations of snails may have colonized a very exposed rock islet near the low shore in the past, adapting themselves to the low shore conditions free of gene flow from the upper shore, and then reinvaded the upper shore. But no such islets populated by lower-shore-phenotype snails have ever been observed. It is easier to imagine the spatial isolation of subpopulations colonizing protected shores. There are long stretches of protected shore in the Galician Rias, in which there are large populations of L. saxatilis with upper-shore-like shells whilst snails with lower-shore-like shells are completely absent. This may be consistent with lower wave action in protected shores, which makes the physical environment somewhat similar to that in the upper areas of exposed shores. After having adapted themselves to protected shores by evolving the upper-shore phenotype, such subpopulations may have recolonized the upper levels of exposed coasts, thus originating the presently observed hybrid zone.

Nonetheless, the strong divergent selection found in this experiment rather suggests a nonallopatric origin. Divergent selection makes up an important component of total natural selection in these populations, and was even dominant in Portecelo, where no directional selection was detected for any character. This may be an indication of nonallopatric differentiation, because in the presence of gene flow, divergent selection may have to play a dominant role for the progressive development of genetic differentiation and reproductive isolation to occur. In situations of secondary contact between previously separated subpopulations that may have diverged, at least in part, by genetic drift, divergent selection may be of little importance, depending on what proportion of the differentiation between subpopulations was adaptive. Indeed, divergent selection must be strong to maintain polymorphism, along with the observed incomplete assortative mating (Rolán-Alvarez et al., 1999; Cruz et al., 2004) and the reduced reproduction of intermediate individuals (Cruz et al., 2001; Cruz & García, 2001), in this narrow Littorina hybrid zone, in which both the shore band occupied by the lower shore morph and the midshore band of phenotypic transition are only a few meters wide. This is not a large distance in terms of the dispersal ability of individual snails (Erlandsson et al., 1998), so that gene flow would constitute a serious hindrance to local adaptation (Rice & Hostert, 1993; Doebeli & Dieckmann, 2003).

Differentiation in parapatry is a viable and simple explanation for the origin of this hybrid zone, because, if divergent selection is strong enough to maintain the observed differentiation in the presence of gene flow, there is a good chance that it was strong enough to initiate it (Fry, 2003). The action of divergent selection is at present helped by assortative mating and reduced reproductive fitness for phenotypically intermediate individuals, but assortative mating at least may have taken place from the start of the differentiation process. This is because assortative mating is generated in these populations by differences in spatial microdistribution and in shell shape and size (Rolán-Alvarez et al., 1999; Cruz et al., 2004), so that it would have begun to take place as soon as these differences developed. Thus, our results based on the phenotypic analysis of selection were consistent with a parapatric origin for this hybrid zone, and therefore also with previous interpretations based on allozyme analysis (Johannesson et al., 1993).

The estimates of divergent selection across shore levels (see the regression coefficients for the phenotype × release point products in Table 1) for some of the characters were easy to interpret in terms of the biology of intertidal gastropods (Vermeij, 1993, pp. 63–76). Small shell size (or negative values for SH) and relatively wide shell aperture (or positive values for SAH, SAW and SW) are favoured on the lower shore because they reduce the risk of detachment by strong water flow or wave action, whereas the alternative combination for the same measures (i.e. large shell sizes and narrow apertures) reduces the risk of overheating and water loss by exposure to the sun. The role of other measures subject to considerable divergent selection, such as W1W and W2W, is less obvious.

The adaptive value of shell shape, as measured by PC1– the most important principal component, was higher than that of the more conspicuous ridges and bands. Therefore, the presence of ridges and bands does not appear to be the main target of divergent selection in this hybrid zone. However, these ridges and bands may also play a role in the adaptive differentiation between morphs, as the coefficients for their products were almost significant in the Portecelo population. Perhaps this effect was not large enough to be detected in the separate analyses of partially banded and ridged snails. The separate analyses were less powerful than the overall analysis because they used fewer individuals and included a narrower range of shell shapes.

Phenotypic analysis of natural selection, along with use of genetic markers (e.g. McCune & Lovejoy, 1998; Pfenninger & Posada, 2002; Rico et al., 2003) and biogeographic studies (e.g. Endler, 1977; Barraclough & Vogler, 2000; Coyne & Price, 2000) may provide interesting information about the history and mechanisms involved in speciation processes. In the present study, two features were revealed – a close relationship between divergent selection and differentiation, and a high degree of divergent selection. These clearly provide evidence of a decisive role of divergent selection in the development of the observed differentiation. They are also consistent with parapatric differentiation, but not with many situations of allopatric differentiation, thereby suggesting a primary origin for the Galician L. saxatilis hybrid zone.

Acknowledgments

This work was funded by Spain's DGICYT with grant PB94-0649.

Ancillary