Contrasting patterns of body shape and neutral genetic divergence in marine and lake populations of threespine sticklebacks



    1. Ecological Genetics Research Unit, Department of Biological and Environmental Sciences, PO Box 65, FI-00014, University of Helsinki, Helsinki, Finland
    Search for more papers by this author
  • J. M. CANO,

    1. Ecological Genetics Research Unit, Department of Biological and Environmental Sciences, PO Box 65, FI-00014, University of Helsinki, Helsinki, Finland
    Search for more papers by this author

    1. Ecological Genetics Research Unit, Department of Biological and Environmental Sciences, PO Box 65, FI-00014, University of Helsinki, Helsinki, Finland
    Search for more papers by this author

    1. Ecological Genetics Research Unit, Department of Biological and Environmental Sciences, PO Box 65, FI-00014, University of Helsinki, Helsinki, Finland
    Search for more papers by this author

T. Leinonen, Ecological Genetics Research Unit, Department of Biological and Environmental Sciences, PO Box 65, FI-00014 University of Helsinki, Helsinki, Finland.
Tel.: +35819157801; fax: +35819157694


Comparisons of neutral marker and quantitative trait divergence can provide important insights into the relative roles of natural selection and neutral genetic drift in population differentiation. We investigated phenotypic and genetic differentiation among Fennoscandian threespine stickleback (Gasterosteus aculeatus) populations, and found that the highest degree of differentiation occurred between sea and freshwater habitats. Within habitats, morphological divergence was highest among the different freshwater populations. Pairwise phenotypic and neutral genetic distances among populations were positively correlated, suggesting that genetic drift may have contributed to the morphological differentiation among habitats. On the other hand, the degree of phenotypic differentiation (PST) clearly surpassed the neutral expectation set by FST, suggesting a predominant role for natural selection over genetic drift as an explanation for the observed differentiation. However, separate PST/FST comparisons by habitats revealed that body shape divergence between lake and marine populations, and even among marine populations, can be strongly influenced by natural selection. On the other hand, genetic drift can play an important role in the differentiation among lake populations.


Uncovering the causes behind patterns of phenotypic and genetic differentiation is a crucial step towards answering some of the most intriguing questions of evolutionary biology – how fast do adaptive differentiation and radiations occur, and what are the relative roles of selective and neutral processes behind the observed differentiation? Patterns of divergence in any organism are likely to exhibit both shared and unique features (Langerhans & DeWitt, 2004). Shared features can result either from a shared selective regime or common ancestry. Unique features, on the other hand, reflect the specific evolutionary history of each organism, within certain boundaries set by historical and functional constraints (Langerhans & DeWitt, 2004). It is thus imperative that phylogenetic relationships are taken into account in studies of adaptive divergence (e.g. Taylor & McPhail, 2000), and that the studies are done on a proper scale (Volis et al., 2005).

Phylogenetic relationships can be inferred from neutral markers, divergence in which can also be utilized as a neutral expectation, when testing for natural selection on quantitative traits (e.g. Lande, 1976; Lynch, 1990; Merilä & Crnokrak, 2001). Complex quantitative traits have traditionally presented a number of challenges to the study of evolution (e.g. Lande, 1979; Lande & Arnold, 1983), but recent advances in the study of body shape by geometric morphometrics (Bookstein, 1991; Rohlf & Marcus, 1993) have helped to overcome some of these challenges. Geometric morphometrics is rapidly gaining popularity in its application to a variety of biological contexts, such as quantitative genetics (e.g. Klingenberg & Leamy, 2001; Monteiro et al., 2002; Reimchen & Nosil, 2004) and phylogeny reconstruction (e.g. Rohlf, 1998, 2001; Cardini & O'Higgins, 2004; Caumul & Polly, 2005; Couette et al., 2005).

The threespine stickleback species complex provides a good example of the importance of considering historical factors in the study of evolution (Taylor & McPhail, 2000). Originally a marine species, threespine sticklebacks have colonized freshwater habitats across the Northern Hemisphere. As a result of the last ice age and the following land uplift, numerous freshwater populations have become isolated and subsequently diverged both morphologically and genetically from their marine conspecifics (Bell & Foster, 1994). Phenotypic divergence in threespine sticklebacks has been found to have an adaptive basis (e.g. Bell & Foster, 1994; Walker, 1997; McKinnon & Rundle, 2002; Hendry & Taylor, 2004), but formal tests of the effect of genetic drift as an alternative explanation for the observed differentiation are scarce at best (but see Hendry et al., 2002; Hendry & Taylor, 2004; Schluter et al., 2004; Raeymaekers et al., 2005). Moreover, most of the studies of morphological variation either focus mainly on the divergence of bony armour and less on body shape (but see Walker, 1997; Taylor et al., 2006), compare sympatric morphs in freshwater habitats (Kristjansson et al., 2002; Taylor et al., 2006), or are restricted to the North American range of distribution (but see Kristjansson et al., 2002 for Icelandic populations).

This study aims at uncovering the patterns of body shape divergence among marine and lacustrine threespine stickleback populations in Northern Europe. The goal is twofold. First, to test whether body shape divergence is higher among populations inhabiting lakes than among marine populations, given the strong habitat differences and reproductive isolation among the lake populations, and the apparent lack of barriers to gene flow among the marine populations. In turn, we expect that the greater degree of divergence will be found between marine and lake populations due to the strong environmental differences imposed by marine and freshwater habitats (i.e. salinity, habitat complexity, food availability and predation pressure). Secondly, to assess the relative roles of genetic drift and natural selection for the observed population differentiation in body shape by comparing the degree of phenotypic differentiation measured by the PST index, analogous to QST (Spitze, 1993) but affected by environmental effects, against the neutral expectation set by allelic divergence in microsatellite markers (FST).


Study populations and sampling

Threespine sticklebacks were collected from 10 locations during summer 2003 (Fig. 1). Of these, four were Baltic coastal and five were landlocked lake populations (Table 1). In addition, one marine population from the North Sea (coastal Norway) was used as a reference for the ancestral marine form (Münzing, 1963; Walker & Bell, 2000; evidence for marine ancestry also reviewed by Bell & Foster, 1994). Samples were collected using minnow traps and a seine net, all with a mesh size of 6 mm. All sampling was done from the shore with depth not exceeding 2 m, with the exception of the North Sea population (MV), which was obtained as a by-product of trawling. Fifty to 100 threespine sticklebacks were collected from each location. The fish were killed with an overdose of MS-222 (tricane methanesulphonate) and stored in individually in 96 % ethanol. The tubes were stored horizontally to prevent bending of the specimens.

Figure 1.

 Locations of sampling sites. Sea populations are marked with filled circles, lake populations with open circles. Abbreviations are explained in Table 1.

Table 1.   Sampling sites, their coordinates and habitat characteristics.
Sampling siteCodeCoordinatesSample size (females /males)HabitatHeight from sea level (m)Distance to sea (km)Surface area (ha)Maximum depth (m)Possible stickleback predators* HE
  1. *CL, Coregonus lavaretus; ST, Salmo trutta; TT, Thymallus thymallus; EL, Esox lucius; SA, Salvelinus alpinus; PF, Platichthys flesus; SS, Salmo salar; SL, Stizostedion lucioperca.

  2. Niemelä & Vilhunen (1987).

  3. ‡Introduced.

  4. §Degerman et al. (2001).

  5. Mäkinen et al. (2006).

Kotka, MussaloKO60°27′N, 26°55′E47 (24/23)Sea00n/an/an/a0.81
Pyhäjoki, KaukorantaPJ64°28′N, 24°13′E97 (53/44)Sea00n/an/an/a0.80
Hanko, TvärminneTV59°50′N, 23°12′E46 (23/23)Sea00n/an/an/a0.81
Tornio, Peräpohjolan opistoTO65°51′N, 24°09 E50 (31/19)Sea00n/an/an/a0.79
Utsjoki, KevojärviUK69°45′N, 27°00′E53 (30/23)Lake75.370116 35CL, ST, TT, EL†0.60
Utsjoki, MieraslompoloML69°34′N, 27°14′E41 (19/22)Lake204757<10CL†, Odonata larvae0.29
Utsjoki, KarilampiKL69°33′N, 27°14′E47 (40/7)Lake2047514<10CL, ST†, Odonata larvae0.27
Nuorgam, PulmankijärviPU69°58′N, 27°58′E46 (24/22)Lake14.6251623 35CL‡, ST, TT, EL, SA, PF†0.62
Sweden, Vättern, VisingsöRV58°54′N, 14°24′E82 (59/23)Lake88100191 200127EL, SA, SS, SL, ST§0.74
Norway, OrrevannetMV58°44′N, 05°31′E50 (33/17)Sea00n/an/an/a0.80

After 2 months in alcohol, the right pectoral fin was removed from each fish for DNA analysis. The fish were then fixed in 10 % formalin for a minimum of 2 weeks. After the fixation the fish were stained allowing for external bones to be seen with a naked eye. The staining was done with Alizarin red S using a procedure that has been used previously by Pritchard & Schluter (2001). Digital photographs of the fish were taken when the staining was complete.

Data acquisition

The left side of each fish was photographed with a Nikon Coolpix 4500 digital camera (with 4.0 Mega Pixels resolution; Nikon Corp., Tokyo, Japan). All the photographs were taken from a standard angle, and a ruler was placed in each photograph for scaling. After photographing, the fish were sexed by gonad inspection. Landmark based geometric morphometrics were used to acquire shape data. Sixteen landmarks (Fig. 2a) were digitized on each image using tpsDig 1.37 (Rohlf, 2002a). The landmarks were chosen so that they would capture the overall body shape with as few variables as possible, because the power of geometric morphometric analyses is inversely proportional to the number of landmarks for a given number of specimens. Another criterion for the choice of landmarks was that they should capture morphological traits that have shown variation in earlier studies of threespine sticklebacks, preferably studies that have used geometric morphometric methods (Walker, 1997; Caldecutt & Adams, 1998; Walker & Bell, 2000; Schluter et al., 2004).

Figure 2.

 Schematic illustration of a threespine stickleback with 16 landmarks (a) (see Methods section for a description of the landmarks), and shape changes along the first (b) and second (c) relative warp axes. Arrows indicated the magnitude and direction of change at each landmark. The arrow direction is from positive to negative sea to lakes in (b), small lakes to larger lakes in (c).

The landmarks digitized were (Fig. 2): (1) anterior tip of upper lip; (2) posterior edge of angular; (3) posterior tip of ectocoracoid; (4) caudal tip of posterior process of pelvic girdle on ventral midline (VML); (5) base of first anal ray on VML; (6) insertion of anal fin membrane on VML; (7) origin of caudal fin membrane on VML; (8) caudal border of hypural plate at lateral midline; (9) origin of caudal fin membrane on dorsal midline (DML); (10) insertion of dorsal fin membrane on DML; (11) base of the first dorsal fin ray at DML; (12) anterior junction of the second dorsal spine with DML; (13) anterior junction of the first dorsal spine with DML; (14) posterior extent of the supraoccipital; (15) lachrymal – lateral ethmoid suture at orbit; and (16) anterioventral extent of sphenotic at orbit.

Analysis of population effects

Digitized landmarks were superimposed with tpsrelw 1.31 (Rohlf, 2002b), which aligns, scales and rotates the landmark configurations using generalized orthogonal least squares Procrustes procedures (Rohlf & Slice, 1990). Partial warp scores (including uniform components) were obtained with the same program from the superimposed specimens, and were used as shape variables in statistical analyses.

To test whether the habitat (sea or lake) of origin had an effect on shape, a multivariate analysis of covariance (mancova) was performed on the partial warps scores. Habitat was considered as a fixed factor, whereas population (nested within habitat) was treated as a random effect and body size (i.e. centroid size) as a (fixed) covariate. Although the effects of isometric growth are removed in the superimposition, the effects of allometry remain (Rohlf & Marcus, 1993). Because of this, interactions between centroid size and population as well as centroid size and habitat were also tested. Centroid size is a reliable size measure, not only theoretically (e.g. Bookstein, 1991), but also on the basis of earlier fish studies which have found strong correlations between centroid size and standard length (e.g. Walker, 1997; Douglas et al., 2001; Langerhans & DeWitt, 2004). Furthermore, in a random sub-sample of 202 fish in our data, the centroid size and standard length of the fishes were strongly positively correlated (r2 = 0.95, F1,204 = 3478, P < 0.001). The mancovas were performed with s-plus® 6.1 (Insightful Corp., Seattle, WA, USA).

General patterns of shape differentiation were investigated by performing a relative warps analysis on the partial warps scores. The relative warps analysis with α = 0 (i.e. equal weights for all partial warps at all spatial scales) is in essence a principal component analysis. Thin plate spline grids together with vectors were used to see how the results of the relative warps analysis were reflected on shape. The shape visualizations, as well as the relative warps analysis were done using program tpsrelw, version 1.31 (Rohlf, 2002b). Orthogonal least squares superimposition was used with scaling to unit centroid size.

A canonical variates analysis (CVA) was done to assess shape differentiation among populations. Shape changes associated with canonical variates axes were visualized by projecting the specimens onto canonical vectors and regressing the partial warps scores onto each canonical axis (e.g. Rohlf et al., 1996; Adams & Funk, 1997). The CVA was done using ntSYS-pc 2.11s (Rohlf, 2000) and tpsregr 1.30 (Rohlf, 2004) was used to obtain the shape visualizations.

An overall measure of shape differences between individuals and populations was needed for comparison with molecular genetic data. Procrustes distance gives such an overall measure of difference between landmark configurations, and was therefore used to build a neighbour joining tree, which was then compared with the tree drawn from the microsatellite data. A two-way Mantel's z-test with 1000 permutations test was computed with ntsyspc 2.11s (Rohlf, 2000) to make a linear comparison of the distance matrices that were used to draw the trees. Procrustes distances were calculated with tpssmall 1.20 (Rohlf, 2003) and the neighbour joining tree was built with values from ntsyspc 2.11s(Rohlf, 2000) and plotted with treeview1.6.6 (Page, 1996).

All Tps-software was obtained from, which is a service to the scientific community, maintained by Prof. F. James Rohlf.

Molecular data

The divergence in neutral markers was estimated based on previously published allele frequencies at 18 microsatellite loci (Mäkinen et al., 2006). On average 32 individuals, including those used in the morphological analyses, were analysed from each population. FST values were estimated according to Weir & Cockerham (1984) and were calculated as implemented in fstat (Goudet, 2001). A neighbour-joining tree was assembled from FST distance matrix using the neighbour subprogram in the Phylip 3.6 program package (Felsenstein, 2004).

Index of phenotypic divergence

The degree of phenotypic divergence (PST) was used to evaluate the relative roles of selection and genetic drift as explanations for divergence in phenotypic traits. PST is analogous to the index of divergence in genes coding for quantitative traits (QST), but can be influenced by environmental and nonadditive genetic effects (e.g. Merilä & Crnokrak, 2001). Nevertheless, the use of PST as a surrogate for QST may be justified when QST estimates are not available [see Meriläet al. (1997); Storz (2002) and Saint-Laurent et al. (2003) for examples of studies based on PST estimates].

PST was estimated as


where σGB2 is the variance between populations, σGW2 variance within populations and h2 the heritability. Since PST is affected by environmental effects, the PST values were calculated with two hypothetical heritability (h2) values, in order to account for two contrasting scenarios. First, assuming heritability of 0.5, meaning that environmental and nonadditive genetic effects account for half of the observed phenotypic variation; second, assuming that all variance is genetic (h2 = 1) and therefore all the phenotypic variation is considered to be genetic and additive. Because the scores from the first two relative warps were used as surrogates for body shape, PST does not encompass all shape variation, but only the shape variation associated with the first two relative warp axes.

For PST estimation, the body shape traits were assumed to be normally distributed and a linear model was fitted to each trait (response variable) separately. Population was entered in to the model as a random effect and body size (i.e. centroid size) as a covariate. The models were fitted to the data using a Bayesian approach (Gelman et al., 2004) and the full posterior distribution of all the parameters of interest was estimated (i.e. σGB, σGW and PST). Prior distributions were uninformative. Specifically, the inverses of all the variances were given exponentially distributed priors with mean of one. The resulting prior distribution of PST was flat. All the models were fitted using a Gibbs sampler with winbugs 1.4 package (Spiegelhalter et al., 2003). For each analysis, the posterior distributions were obtained running two chains (10 000 iterations after a burn-in of 5000 iterations and convergence reached); every second of the 2 × 10 000 iterations was taken to give 2 × 5000 draws from the posterior distribution.


General patterns in body shape and size differentiation

The body shape variation presented here can be considered at three different levels, each of which has its own purpose. First, the nested mancova of all partial warps and uniform components represents all the body shapes captured by the chosen landmarks. It revealed significant body shape differences among populations and habitats (Table 2). The significant differences in body size among (F1,549 = 904, P < 0.001) and within habitats (sea: F4,289 = 28.1, P < 0.001; lakes: F4,268 = 19.4, P < 0.001) imply different covariate ranges for the shape analysis, and therefore the potential allometric effect of body size on shape cannot be statistically accounted for.

Table 2.   The results of a nested mancova of body shape variation.
Pop (habitat)3.25312.8972244216<0.001

The second block of results focuses more on those specific features of body shape that diverge the most among populations. The first relative warp axis (RW1) divided the populations according to their habitat, whereas on the second relative warp axis (RW2) lake populations are divided according to the size of the lake the samples originate from (Fig. 3; lake characteristics listed in Table 1). The remaining RWs, representing 52 % of variation in body shape, were unrelated (completely scattered and mixed) with respect to habitat or population of origin (results not shown). For this reason, we decided to estimate the indices of population differentiation (PST) using the individual scores on the two first RWs. The measures of phenotypic divergence (PST) thus comprise only the shape differences related to the first two relative warp axes. The shape visualizations along the first relative warp axis show that the sea populations are deeper bodied, have shorter caudal peduncles and smaller eyes than the lake populations (Fig. 2b). Shape visualizations along the RW2 indicate that the populations from the largest lakes (PU and RV) are characterized by longer caudal peduncles than the rest of the lake populations (Fig. 2c). Another difference can be seen in the distance between the caudal tip of the pelvic girdle (landmark 4) and the first anal fin ray (landmark 5; Fig. 2a). This distance is longer in the samples from the large lakes, where fish also have on average smaller mouths, bigger heads, and longer snouts (distance from landmark 1–15) than fish from the rest of the populations (Fig. 2c).

Figure 3.

 The first two relative warps showing 95 % confidence intervals for the means of each population. See Table 1 for population abbreviations.

The CVA (Fig. 4) confirmed the main features of body shape differentiation among populations, which are mostly differentiated in body depth and the length of the caudal peduncle, as revealed by the first CVA axis (46 % of shape variation explained; Fig. 4). Marine populations are located at one end of the first CVA axis clearly differentiated from the lake populations. The marine populations have deep bodies, whereas the lake populations tend to have more slender bodies. The shape visualizations along the second CVA axis show that this tendency is more pronounced in the populations inhabiting larger lakes and probably have larger population sizes as indicated by their levels of heterozygosity (Table 1). The CVA also highlighted the trend that there is more differentiation among lake populations than among sea populations. Among the sea populations, the ancestral reference population (MV) is the only one that shows differentiation on any of the first three canonical variates axes. The second and the third canonical variates axes also show that the large lake populations are differentiated among themselves (Fig. 4).

Figure 4.

 Canonical variates ordination of threespine stickleback populations. The first three canonical variates axes represent 46 %, 21 % and 11 % of the shape variation, respectively. Lake populations are marked with empty circles, sea populations with filled circles. Shape visualizations are from the extremes of each axis. Deformations of the grid indicate differentiation from the overall consensus configuration. See Table 1 for population abbreviations.

Comparison of genetic and morphological divergence

In this third block of the results, the absolute magnitude of variation in the average shape is considered without providing a direction for the variation in morphospace. Comparison of the neighbour joining tree of the overall shape differences with the tree based on FST-values concurs with the overall pattern described above (Fig. 5). In other words, both trees show more divergence among the lake than among the sea populations. The main difference between the trees is that in the FST tree Lake Vättern (RV) population clusters with sea populations, while in the tree with Procrustes distances it clusters with the lake populations (Fig. 5). The relative degree of divergence is higher among the sea populations for body shape than for neutral marker variation (Fig. 6). Comparison of FST and Procrustes distance matrices with two-way Mantel z-test (Mantel, 1967) showed that there was a significant correlation between the two matrices (r = 0.34, P = 0.027), indicating that differentiation in neutral genetic markers is linked with shape differentiation. Note that when comparing the branch lengths of the trees, relative rather than absolute lengths should be looked at, because both trees use different indices as units (Fig. 5).

Figure 5.

 Neighbour joining trees obtained from FST values (left) and Procrustes distances (right). See Table 1 for population abbreviations.

Figure 6.

PST values for shape assuming two different heritability (h2) values along with FST values. All values are with 95 % Bayesian confidence intervals. Variance in shape is adjusted for variance in size and sex by using the latter two as covariates.

Comparison of phenotypic (PST) and neutral genetic marker (FST) differentiation confirms what was already evident in the neighbour joining trees. Overall differentiation in shape (PST) along the RW1 (i.e. representing variation in body depth and length of the caudal peduncle) is larger than the overall differentiation in neutral genetic markers (FST), but within the two different habitats, PST is higher than FST among the sea populations (confidence intervals do not overlap), whereas FST does not significantly differ from the PST among the lake populations (Fig. 6). The shape differentiation along the RW2 (i.e. representing position of the mouth and bending of the body) follows the same trend, although the differences from the FST values are not significant.


The results of this study revealed similarities and dissimilarities in the patterns of population differentiation in phenotypic traits and neutral marker genes among different threespine stickleback populations and habitats. Pairwise phenotypic and neutral genetic distances were correlated, suggesting a role for genetic drift in explaining divergence in phenotypic traits. However, the average degree of differentiation in phenotypic traits (PST) clearly exceeded that in neutral marker genes (FST) suggesting a predominant role of selection over drift in divergence among marine and lake populations. Furthermore, comparisons within habitats revealed that the relatively modest morphological differences among sea populations appear to be the result of natural selection, whereas those among more divergent lake populations could have been achieved by random genetic drift alone. In what follows, we will discuss these interpretations in detail, and in particular, the possibility that the differing patterns of morphological and neutral differentiation among habitats could be attributable to differences in effective population size.

PST values for body shape across marine and lake populations exceeded FST, and thus, divergent natural selection is a likely explanation for population differentiation among habitats. This is the common pattern seen is most comparative studies of quantitative trait and neutral marker differentiation (reviews in Merilä & Crnokrak, 2001; McKay & Latta, 2002). However, the role of genetic drift cannot be completely ruled out since there was also a significant correlation between pairwise phenotypic and neutral genetic distances. Despite this correlation, there was significant uncoupling in the pattern of differentiation among neutral markers and body shape. This uncoupling is illustrated by the fact that the Lake Vättern population clustered together with the Baltic Sea populations in the FST tree, but not in the tree based on body shape differentiation (Procrustes distances; Fig. 5). The considerably larger effective population size of the Lake Vättern population compared with the other lake populations (Table 1) could explain this pattern. In large populations the effects of drift on neutral markers are reduced, and there are no constraints to natural selection, as discussed below.

Separate analyses of marine and lake populations helped to better disentangle the relative importance of drift and selection on phenotypic differentiation. For instance, comparisons of PST and FST values (cf. Fig. 6) revealed that the patterns and processes behind population divergence differed clearly between habitats. The similar degree of phenotypic and neutral marker differentiation among the freshwater populations (PSTFST) indicates that genetic drift may play a major role in their divergence. Since the population sizes in freshwater populations are much smaller than those in marine populations (see Table 1), genetic drift can be expected to be a much more potent force in freshwater when compared with marine populations. In fact, both theoretical (Wright, 1931; Robertson, 1960) and empirical studies (Jones et al., 1968; Weber & Diggins, 1990; England et al., 2003) suggest selection may become inefficient relative to drift in small populations. The effect of population history should also be considered. Although the Fennoscandian freshwater populations have diverged only recently (c. 10 000 years ago) from a common ancestral marine population (Mäkinen et al., 2006), the patterns of subsequent colonization history can differ significantly between populations. For instance, the Lake Vättern population (RV) is likely to have been connected to the sea longer than the smaller Northern Fennoscandian populations (Eronen et al., 2001). Thus, constraints imposed by low genetic diversity and inefficiency of selection on the one hand, and historical constraints on the other hand, can together explain the observed lack of differentiation between the indices of neutral genetic and quantitative trait divergence among lake populations. However, while similarity between the indices of quantitative and neutral population divergence does not support selective divergence, it does not rule out the possibility that natural selection may still have some influence on the body shape of the freshwater threespine sticklebacks. Environmental factors known to influence body shape divergence among freshwater populations include predation regime and amount of relative littoral area (Walker, 1997).

Although differences in body shape are more apparent in the among habitat comparisons, similar trends can be seen when comparing the freshwater populations among themselves. Threespine sticklebacks from the largest and deepest lakes (PU and RV) are more streamlined and have longer caudal peduncles than the threespined sticklebacks from the smaller and shallower lakes (KL, ML and UK). This is again concordant with the idea that deeper bodied threespine sticklebacks with shorter caudal peduncles should be found in lakes with a higher relative littoral area, although the shortage of native predatory fish in the smallest lakes (see Table 1) would predict a similar outcome (Walker, 1997). Overall, the shape differences between the sea and the lake populations reflect the patterns described in earlier studies using traditional morphometric methods (reviewed by Bell & Foster, 1994). The sea populations are deeper bodied with longer caudal peduncles and relatively smaller eyes than the lake populations. It has been suggested that the shallower body in most freshwater sticklebacks may be a function of reduced predation pressure (Walker & Bell, 2000), but formal tests of this idea are still lacking. Furthermore, a vast majority of the studies on threespine stickleback body shape have thus far compared either sympatric freshwater morphs or freshwater populations among themselves (Walker, 1997; Caldecutt & Adams, 1998; Schluter et al., 2004; Taylor et al., 2006).

In addition to the higher degree of neutral genetic differentiation, there is also more phenotypic divergence among the freshwater populations in comparison with the coastal sea populations. The greater variability among the freshwater populations can be explained by the higher degree of isolation of freshwater environments compared with marine environments. Freshwater environments are also more heterogeneous than marine environments, which further facilitates diversification (McPhail, 1994). Among the sea populations, neutral genetic divergence (FST) is virtually nonexistent, whereas phenotypic divergence is significantly higher (PST). The CVA also reflects clear differences in the mean shapes of the sea populations (Fig. 4). Especially, the reference population from the North Sea (MV) shows that despite the putative unrestricted gene flow among the sea populations, there seems to be still room for divergence based on selection. This result is in agreement with other studies in this species that have provided evidence for adaptive differentiation in the presence of significant levels of gene flow (Hendry et al., 2002; Hendry & Taylor, 2004), although caution should be exercised while comparing these results to ours. Our PST estimates are not free of environmental and nonadditive genetic effects (e.g. Merilä & Crnokrak, 2001; Lee & Frost, 2002). Even when considering h2 = 0.5, we are still assuming that half of the phenotypic variation is genetic and additive. However, two lines of evidence suggest that it is unlikely that population specific maternal or environmental effects would have biased our inference to a significant degree. First, population differences in body shape (especially in body depth) similar to the ones observed here have been shown to have a genetic basis in other threespine stickleback populations (McPhail, 1984). Secondly, preliminary results from common garden experiments conducted with a subset of populations used in this study indicate that body size and shape differences among populations are actually genetically determined (J.M. Cano, T. Leinonen, J. Merilä, unpublished data).

In conclusion, this study confirms that there is more body shape divergence among freshwater than among marine stickleback populations in Fennoscandia. The same applies to the neutral genetic divergence. While the results suggest that genetic drift has contributed to patterns and magnitude of differentiation in both habitats, the main factor influencing morphometric divergence among the sea populations and across habitats appears to be directional natural selection. In contrast, although the degree of morphological differentiation among lake populations exceeds that among marine populations in absolute terms, genetic drift cannot be excluded as an explanation for morphological differentiation among lake populations.


We thank Per Sjöstrand and Arne Levsen for providing the samples from Sweden and Norway, as well as Ari Haikonen and Laura Buggiotti for helping with the sampling in Finland. We also thank two anonymous referees, whose comments helped to improve this manuscript. This study was supported financially by the Academy of Finland and LUOVA-graduate school funded by the Ministry of Education (Finland).