Morphological and genetic divergence of intralacustrine stickleback morphs in Iceland: a case for selective differentiation?


Guðbjörg Ásta Ólafsdóttir, Institute of Biology, University of Iceland, Sturlugata 7, Reykjavik IS101, Iceland.
Tel.: +354 525 5230; fax: +354 525 4281; e-mail:


The evolutionary processes involved in population divergence and local adaptation are poorly understood. Theory predicts that divergence of adjacent populations is possible but depends on several factors including gene flow, divergent selection, population size and the number of genes involved in divergence and their distribution on the genome. We analyse variation in neutral markers, markers linked to putative quantitative trait loci and morphological traits in a recent (< 10 000 years) zone of primary divergence between stickleback morphs in Lake Thingvallavatn, Iceland. Environmental factors, especially predation, are clearly implicated in reducing gene flow between morphs. There is continuous morphological and genetic variation between habitats with a zone centre similar to secondary contact zones. Individual microsatellite loci are implicated as being linked to adaptive variation by direct tests as well as by differences in cline shape. Patterns of linkage disequilibria indicate that the morphs have diverged at several loci. This divergence shows parallels and differences with the well-studied limnetic–benthic stickleback morphs, both in phenotypic divergence and at the genomic level.


Speciation is a key subject in evolutionary biology (Coyne & Orr, 2004). Early studies focused on ecological differences between species and natural selection as a driving force in speciation. As genetics came to dominate Neo-Darwinism, there was a shift towards studies of quantitative and population genetics, with research concentrating on the analysis of species differences. Recently, there has been a resurgence of interest in the influence of ecological selection on population divergence (Schluter, 2000). This has been partly stimulated by numerous reports of ecologically diverse morphs or recently formed species found to coexist in close proximity.

Gene exchange between ecologically differentiated adjacent groups will counteract their divergence. Understanding the maintenance of divergence in sympatry or across ecotones is therefore of special interest. Theory shows that ecological divergence can lead to speciation if there is a connection between sexual isolation and diverging ecological traits, for example with the build-up of correlated genetic influences or with a single character being involved in both ecological specialization and mate choice (Kondrashov & Kondrashov, 1999; Gavrilets et al., 2000); however, field studies usually implicate direct ecological selection against hybrids in addition to assortative mating resulting from divergence in ecological niches.

The simplest and probably the most convincing example of reproductive isolation being influenced by ecological factors is when premating isolation is caused by host or habitat shift (Feder, 1998; Wood et al., 1999; Nosil et al., 2002). The apple maggot fly (Rhagoletis pomonella) has provided a textbook example of this process, although recent work indicates that a more complex mechanism may be involved (Feder et al., 2003). Their adaptation to a newly introduced host plant from its native host has created two reproductively isolated host races (Feder, 1998). Habitat choice or site fidelity can also facilitate morph formation and may have been important in the divergent morphs of the Lake Malawi cichlids. Several studies have emphasized the role of habitat complexity in promoting divergence by what has been described as microallopatry (Van Oppen et al., 1997; Arnegard et al., 1999; Markert et al., 1999). In Lake Malawi, deep waters and sandy bays restrict migration between rocky patches, even at small distances, although this does not suffice to completely isolate distinct populations (Arnegard et al., 1999). These isolated groups do not necessarily exhibit strong ecological specialization, for example in trophic resource use, but are kept isolated by site fidelity and sexual selection (Turner, 1999). A direct role for natural selection in limiting migration between habitats may also be an important factor in ecological speciation. This process has been suggested in leaf beetles (Funk, 1998), mimetic butterflies (Mallet & Barton, 1989) and sticklebacks (Schluter, 1995). Increased susceptibility to predators directly reduces movement between different colour pattern morphs of walking stick insects, consequently limiting gene flow (Nosil, 2004). In contrast, a recent study of guppies finds no effect of predation risk in intermediate habitats on gene flow between populations, where geographical features predominantly explain gene flow (Crispo et al., 2006).

The processes which maintain differences between adjacent subspecies have been extensively studied in hybrid zones, especially those resulting from secondary contact. Cline theory provides a useful approach for the study of gene flow and selection between locally adapted contiguous populations (Barton & Hewitt, 1981, 1985; Szymura & Barton, 1991; Kruuk et al., 1999; Moore & Hendry, 2005). Looking at parapatric populations or locally adapted marginal populations at a small scale allows the application of cline models, providing detailed information on the forces maintaining the populations (Hendry et al., 2002; Ross & Harrison, 2002). For instance, cline theory can provide information on differential selection on characters and the maintenance of bimodality in contact zones (Bridle & Butlin, 2002) as well as the balance between gene flow and adaptive divergence (Bell & Richkind, 1981; Bell, 1982; Moore & Hendry, 2005). Characters that have diverged in response to natural selection may show very different patterns of variation from neutral genetic markers (Durett et al., 2000). For example, different patterns of divergence of neutral microsatellite markers and quantitative traits have been used to infer divergent natural selection as a primary factor in the divergence of Drosophila melanogaster populations across an altitude cline (Gockel et al., 2001).

Icelandic lakes have recently diverged morphs of arctic charr and sticklebacks. Two stickleback morphs have been described in Lake Thingvallavatn (Kristjánsson et al., 2002). One occupies small caves and crevices in benthic lava surfaces at the north end of the lake (lava morph), whereas the other is found in abundance at depths of 10–20 m where dense stands of the green alga Nitella opaca predominates (nitella morph). The morphs differ substantially in phenotype, the nitella morph is more slender, with a smaller head and fins and longer spines (Fig. 1). The genetic basis of these morphological differences has not yet been determined in the Lake Thingvallavatn stickleback. However, in other systems of recently diverged threespine stickleback morphs population differences in these morphological traits have been shown to be genetically based (McPhail, 1984; Hatfield, 1997). Although both morphs are benthic the morphological differences are to some extent parallel to the Canadian benthic–limnetic species pairs (McPhail, 1994), with the lava morph resembling the benthic type and the nitella morph resembling the limnetic form. In Lake Thingvallavatn sticklebacks are subjected to high predation pressure by a large piscivorous morph of arctic char, a specialized stickleback predator (Malmquist et al., 1992), and some of the most notable differences among the stickleback morphs are seen in antipredator morphology and behaviour (Kristjánsson et al., 2002; Doucette et al., 2004). It has been suggested that predation is important in the divergence and maintenance of the stickleback morphs (Doucette et al., 2004). Sticklebacks predominantly occupy the two habitats where cover from predators is abundant, i.e. the stony littoral zone (including the lava habitat) and the dense nitella zone. These habitats are separated by a zone where little cover from predation is available (Fig. 2). This ‘predation belt’ could therefore act as a barrier limiting movement of stickleback between the littoral zone and nitella habitat.

Figure 1.

 The lava (a) and nitella (b) morph. Morphological measurements are shown on the Nitella individual. 1 = length of the first dorsal spine; 2 = length of the second dorsal spine; 3 = length of the pelvic spine; 4 = body depth; 5 = length of the last gillraker on the lower arch; 6 = number of gillrakers on the lower arch.

Figure 2.

 (a) Sampling localities within Lake Thingvallavatn. Site 1 represents a pond approximately 200 m north of the lake. A recent lava flow has entered the lake from the north (indicated with grey dots) forming the substrate for approximately 2 km. The lava is most rugged towards the north as the constant inflow of spring water (indicated with grey arrows) prevents accumulation of sediment. The lava habitat (rugged lava, inflow from springs) includes sites 2 and 3. A smaller, more recent lava patch is found at site 12. Grey areas indicate dense nitella opaca patches, however, Nitella opaca is found around the lake at depths of between 10 and 20 m. Sample sizes and further habitat characteristics can be found in Table 1. (b) Schematic outline of the main habitats. This figure shows how the proposed predation belt is formed between the stony littoral and the nitella (filamentous algae) zone at 10–20 m. The breadth of the predation belt varies considerably among sites as it is connected to the depth gradient.

Here small-scale geographical variation in morphological characters and microsatellite allele frequencies of neighbouring groups of threespine sticklebacks in Lake Thingvallavatn is examined to test several predictions concerning intralacustrine population divergence. (1) Multilocus assignment is used to identify the two morphs, lava and nitella, and their putative hybrids. (2) Tests based on reduction in allelic variation are used to identify genetic markers potentially linked to adaptive variation. (3) Partial Mantel tests are used to test for associations between putatively neutral and putatively selected genetic loci and ecological factors. Specifically, we ask: (i) can genetic divergence be explained by isolation by distance? (ii) does the predation risk in intermediate habitats reduce gene flow between littoral and nitella habitats? (iii) does ecological variation better explain genetic divergence at markers linked to quantitative trait loci (QTL) than at variation at neutral markers? In addition: (4) the zone of divergence between the lava and nitella morph is examined in detail and we ask: (i) is there a discernable centre to the zone, akin to hybrid zones? (ii) are there differences in clinal patterns between neutral, putatively selected loci and phenotypic traits, indicating adaptive divergence?

Materials and methods

Study system and sampling

Lake Thingvallavatn is a large oligotrophic lake situated on the Mid-Atlantic rift zone. Recent (< 10 000 years ago) geological activity in the Thingvallavatn basin has created novel freshwater habitats with a complex physical structure and high ecological diversity (Malmquist et al., 2000). At the north end of the lake a post-glacial lava flow followed by high tectonic activity created a benthic lava habitat characterized by underwater rifts, caves and crevices. This area is also characterized by numerous springs with inflowing water at constant temperatures of 4–5 °C. Apart from the lava littoral zone, a stony littoral zone of varying complexity characterizes the shallow waters (0–8 m), except in small bays and coves where occasionally a layer of fine sand or sediment has built up. With increased depth, sediment and vegetation increase. The predominant vegetation is the green alga Nitella opaca, which is found in high densities at depths of 10–20 m. See Fig. 2, Table 1 and Jónasson (1992) for further descriptions of habitats.

Table 1.   Sample size, habitat characteristics, morphological measurements and variance (Var) at the 12 sites. Morphological measures were standardized to a common body length of 4.7 cm. The central lava and nitella groups are distinguished in bold. Only sites 2–10 were used in the lava–nitella divergence zone analysis.
SiteN (males/females)Depth (m)Habitat typeBody1st SpineGillraker

For the current study, samples were taken from several sites in Lake Thingvallavatn in the summers of 2001 and 2002 (Fig. 2). The sites differ in levels of predation, substrate type, vegetation and depth (Table 1). Fish were collected with unbaited benthic minnow traps except at depths of < 1 m at the lava site, where electrofishing was used. All fish were anaesthetized, fin clipped and preserved in 10% formaldehyde. Fin clippings were preserved in 98% ethanol for genetic analysis.

Genetic analysis

Total genomic DNA was isolated from fin clippings using a standard proteinase K lysis. Samples were scored for nine microsatellite loci, Gac2111, Gac4174, Gac3133, Gac7188, Gac1097, Gac7033, Gac1125, Gac1116 and Gac5196 (Largiadér et al., 1999). Four of these (Gac2111, Gac7033, Gac4174 and Gac1125) have been linked to phenotypic variation in other populations of threespine stickleback, although there was an approximately 3 cM distance between the closest marker and the loci Gac7033 and Gac2111. The per cent of variation explained (PVE) refers to the closest mapped marker. Gac2111 and Gac7033 were linked to variation in dorsal spine length (PVE = 1.53 and 2.57 respectively) and Gac7033 to gillraker number (PVE = 15.4) in Canadian benthic limnetic stickleback pairs (Peichel et al., 2001), Gac4174 to variation in lateral plate number (PVE =76.9) and plate width (PVE = 12.9) and Gac1125 to plate width (PVE = 28.9) and height (PVE = 17.9) in crosses of marine and freshwater morphs (Colosimo et al., 2004). Spine length and gillraker number are highly variable in the Lake Thingvallavatn sticklebacks and have diverged between the lava and nitella morphs, the lava morph has considerably shorter spines but more numerous, longer gillrakers. However, there is little variation in plate number between the lava and nitella morphs. Amplification followed the methods of Largiadér et al. (1999) and the products were visualized on 6% polyacrylamide gels. genetix 4.02 (Belkhir, 2000) was used to calculate FIS and estimate heterozygote deficiencies across samples. Significance was confirmed by 1000 permutations. The same program was used to estimate Weir and Cockerham's θ between all populations and to test for significance by permutating the values 1000 times.

Morphological measurements

After being fixed in 5% formaldehyde for approximately 2 weeks, fish were rinsed and transferred to 70% ethanol then stained with alizarin red in 1% KOH (e.g. Bell, 1982) and photographed with a digital camera. Standard length and body depth were measured from the photographs. Spine length and gillraker length were measured and gillrakers and lateral plates counted under a dissection microscope. The length of the last gillraker on the lower gill arch was measured. See Fig. 1 for further information on morphological landmarks.

As a measure of antipredator morphology, the length of the first two dorsal spines and the left pelvic spine was used (Reimchen, 1994). Body depth, gillraker number and length were used to estimate phenotypic differences related to feeding. Body depth differences are commonly observed between stickleback morphs adapted to different foraging environments (McPhail, 1994). This is thought to be effected by streamlined bodies being better suited for sustained swimming, for example in a limnetic environment, whereas deeper bodies are better suited for burst swimming and are more common in benthic feeding sticklebacks (Taylor & McPhail, 1986). More gill rakers are expected to be better for feeding on small planktonic prey and fewer gill rakers for feeding on benthic invertebrates (Bentzen & McPhail, 1984; Lavin & McPhail, 1986). The effect of body size was removed by regressing each size-related character against body length and then using the residuals for further analyses, separately for each sample site (there were no significant differences in standard length between sample sites). Discriminant function analysis was used to estimate overall morphological differences between the groups, as well as between sexes within groups. anova and post hoc Bonferroni correction was used to test for effect of habitat, site, water depth and sex on morphological differences. Sample site nested within habitat type (nitella, lava and littoral) was set as a fixed factor and water depth as covariate. Three sets of Euclidian distances were calculated, the first based on all morphological traits measured, the second based on trophic morphology, body depth, gillraker number and length and a third based on spine lengths. Morphological measurements were scaled to a value between 0 and 1 before being used in the analysis of clinal patterns.

Population structure and individual assignment

We employed a model-based genetic admixture analysis to identify population structure as well as to identify possible migrants or hybrids. The algorithm is implemented in structure 2.0 (Pritchard et al., 2000). Model priors are specified and the parameters estimated using a Bayesian approach with Markov Chain Monte Carlo (MCMC) sampling to estimate the posterior probability distribution of parameters of interest (number of populations, allele frequencies in each population) without prior information of allele frequencies in the populations. The program accounts for deviations from Hardy–Weinberg and linkage disequilibrium by population structure and assigns individuals to a population or jointly to k populations so as to minimize disequilibrium. The model can also incorporate population admixture. Each individual can then have the proportion of its genome originating in different populations [individual admixture proportions (q)] estimated. We evaluated models with K taking values from 1–5, allowing admixture between populations. For each value of K, the MCMC scheme was run with a burn-in period of 100 000 steps and a chain length of 200 000.

To study admixture we use prior information of putative lava and nitella parental morphs. Individuals from sample sites 2 and 9 that were assigned with more than 95% probability either as lava or nitella, were used as parental classes. The fish from intermediate sites were then assigned to one of the parent groups, as admixed individuals or as belonging to neither of those groups.

Signs of selection

Population differences in allelic size variance and expected heterozygozity were used to identify loci possibly under selection (Kauer et al., 2002; Schlötterer, 2002). The ln RV is the natural logarithm of the ratios of the variance in allele size for two populations and ln RH the natural logarithm of variance in expected heterozygosity. It has been demonstrated that ln-transformed ratio of V and H are approximately normally distributed under various demographic scenarios (Kauer et al., 2002; Schlötterer, 2002). The probability that a given locus deviates from neutrality can be obtained from the density function of a standard normal distribution.

Mantel analysis

We used partial Mantel tests to test hypotheses about the effect of geographical and ecological factors on divergence. Mantel tests provide a powerful method to estimate association between two or more dissimilarity matrices (Smouse & Long, 1986; Manly, 1991). Partial Mantel tests allow the analysis of multiple, correlated, predictor variables in a very similar manner as multiple regression, allowing for several interacting hypotheses to be tested simultaneously, separating the often inter-correlated effects of selection and history. They have proved especially useful in testing for associations between genetic distance and environmental and historical variation of closely related populations and species (Thorpe, 1996; Ritchie et al., 2001). As with all multivariate techniques, they have been criticized regarding their ability to disentangle the effect of correlated variables (discussed in Raufaste & Rousset, 2001; Castellano & Balletto, 2002, and Rousset, 2002), but our analyses did seem to distinguish different predictor variables successfully. All matrix variables were standardized to a mean of 0 and unit variance in systat 9 (SPSS Inc., Chicago, IL, USA) before analysis. Matrix comparisons were performed with fstat 2.9.3. (Goudet, 2001), using 10 000 permutations to estimate significance.

The geographical pattern of neutral genetic variation can indicate factors that contribute to reduction in gene flow between sites. We used genetic divergence at loci not known to be linked to selected phenotypes in other studies to represent potential neutral divergence to test two predictions. First, can gene flow between sites be explained by a simple isolation by distance mechanism? Secondly, do ecological factors such as predation risk influence the level of gene flow? The same pattern is not necessarily expected at loci under direct selection or, as in this study, closely linked to QTL potentially under selection. Matrices of genetic divergence at the four loci linked to QTL in other known systems of stickleback divergence (Peichel et al., 2001; Colosimo et al., 2004) were tested separately to see if there is evidence of different patterns reflecting selection on these loci in this system also. Morphological distance matrices were also used to test for adaptive divergence. If ecological selection is shaping the pattern of phenotypic variance, we would expect ecological variation to better predict phenotypic variation than does geographical distance.

Three predictor matrices represented. (1) Geographical distance: shortest potential migration distance measured to the nearest 0.1 km. (2) Water depth: measured in metres during sampling. Water depth covaries with much of the ecological variation, e.g. food availability, substrate complexities and vegetation cover (Jónasson, 1992). As substrate complexities and vegetation are strongly linked with predation risk, in effect predation pressure is partly represented by this predictor matrix although the connection is not as direct as with the next predictor. (3) Predation barrier: the predation hypothesis predicts that fish separated by the predation barrier (intermediate habitats with little cover, see Fig. 2) are more distantly related than predicted by geographical separation and habitat differences. A simple matrix was constructed, assigning either 0 or 1 to pairs of populations based on the presence of a potential predation barrier.

Seven dependent variables were used, five representing genetic divergence and two estimates of phenotypic divergence. Genetic divergence was measured both with assumed neutral microsatellite markers and four markers linked to the QTL putatively under divergent selection. (1) Genetic distance (FST) based on five presumably neutral microsatellite markers. (2) Genetic distance based on Gac7033 (linked to gillraker number, spine length). (3) Genetic distance based on Gac2111 (linked to spine length). (4) Genetic distance based on Gac4174 (linked to lateral plate number and width). (5) Genetic distance based on Gac1125 (linked to lateral plate width and height). Additionally, two morphological matrices are used. (4) An Euclidian distance measure representing possible adaptation to varying predation regimes based on the lengths of the first and second dorsal spine, the pelvic spine and number of lateral plates. (5) A second based on body depth and the number and length of gillrakers, potentially representing trophic specialization.

Cline analysis

Cline shape can be used to infer selection on traits and the position of the cline centre can give information on the factors influencing divergence. In general, stronger selection will cause narrower clines. Therefore, if selection acts only on a few traits there can be large discrepancies in cline widths, both in phenotypic traits and genetic markers (Barton & Gale, 1993). If selection is acting on putative QTL, then linked markers (Gac7033, Gac2111, Gac4174 and Gac1125) are expected to show narrower clines compared with the neutral microsatellite markers. In the same way cline width of the phenotypic traits will depend on the strength of selection, but also on the number of loci affecting each trait. Even with the same level of selection, clines of traits affected by few loci will have narrower clines than traits with many underlying loci. Lack of cline coincidence can be influenced by ecotonal selection or epistatic effects (Butlin et al., 1991; Barton & Gale, 1993; Bridle & Butlin, 2002).

Most analyses of clinal patterns in allele frequencies have been on distinct taxa with several diagnostic alleles. Allele frequency clines between closely related species or populations are more problematic as these groups often share many common alleles that can be present in both populations in high frequencies. Several recent papers have used factorial correspondence analysis (FCA) to sort alleles into groups that contribute significantly to population divergence, and these alleles used for further analysis (Turgeon & Bernatchez, 2001; Bierne et al., 2003). In the Thingvallavatn stickleback the alleles that primarily distinguish the lava and nitella populations (as estimated by FCA) are often relatively rare, found in low frequencies in one of the population but almost absent in the other. Therefore, to try to best represent the pattern of divergence across the zone we fit clines for two types of markers (1) the most common allele at each locus and (2) the grouped diagnostic alleles at each locus (alleles contributing more than 0.01 to the divergence of lava and nitella morphs on axis 1 in the FCA were classified as diagnostic). For the Gac3133 locus a diagnostic allele was also the most common. The analysis was completed with genetix 4.02 (Belkhir, 2000). The cline analysis used samples from sites 2–10, representing the end points of the lava and nitella habitats (Fig. 2).

We used the computer program analyse (Barton & Baird, 1999) to fit maximum likelihood clines to both microsatellite and morphological data. analyse uses a Metropolis–Hastings algorithm to fit a tanh curve to the data, estimating the most likely cline centre, width and character frequencies at each end of the zone given the observed variation in characters across the zone. The shape of a cline can be modelled by combining the following three equations (Szymura & Barton, 1986, 1991):


where decaya = 2√θa, decayb = 2√θb, intercepta = tb/(ta + tb + tatb), interceptb = ta/(ta + tb + tatb), ta = Badecayb, tb = Bbdecayb and X = (x − c)/w is the distance X along the cline, scaled relative to centre and width. Eqn 1b is sigmoid in the centre and eqns  1a and 1c are exponential decay curves leading from either side of the cline centre. θaθb describe the rate of exponential decay and Bb and Ba describe the size of the barrier to gene flow in populations ‘a’ and ‘b’. Cline width is defined as the inverse of the maximum slope and the centre is where the cline slope is steepest (Barton & Gale, 1993). We used the routine ‘Fit 1D cline’ with 20 000 iterations run from a 1000 different starting points. Parameter support limits were set by finding the parameter values where the log likelihood probability drops two units below the maximum (Edwards, 1972).

Significance of differences in cline shape were tested by Bartlett's test of homogeneity as implemented in analyse. Deviations in the position of the centre were tested by fitting a curve to each character keeping the cline centre position fixed to that of the character being tested against. Significant deviations in the position of the cline centre were assumed when the test log likelihood score for the centre was > 2 when compared with the maximum likelihood cline.

Linkage disequilibria

Several evolutionary processes including selection, migration and nonrandom mating will result in disequilibrium between loci, and patterns of linkage disequilibrium have been extensively used in studies of hybrid zones (Barton & Gale, 1993; Barton, 2000; Bierne et al., 2003). The program multilocus 1.3. (Agapow & Burt, 2001) was used to calculate a multilocus linkage disequilibrium estimator rd.

Contact zones of distinct species or ecotypes are expected to result in increased variance in quantitative traits with maximum values at the centre of the zone, caused by the mixing of distinct genotypes or variation in ecological variation (Endler, 1977). The covariance between quantitative traits that vary across a cline will also increase at the zone centre because of linkage disequilibrium between the parental loci (Szymura & Barton, 1986, 1991; Barton & Gale, 1993). The increase in covariance can then be used to estimate pair-wise disequilibrium (Nurnberger et al., 1995; Bridle & Butlin, 2002). Assuming that the environment independently affects traits and that populations do not differ significantly from Hardy–Weinberg proportions, this would cause associations between genetically determined traits and linkage disequilibrium (D*) can be estimated by the equation:


where z and z′ represent the two quantitative traits and Δz and Δz′ the differences in trait frequencies across the cline. This method assumes that the underlying genetic components are additive and distributed across the genome (Barton & Gale, 1993).


Phenotypic divergence

Patterns of morphological variation are summarized in Table 1. There were significant morphological differences between groups from different sample sites (Wilks’λ =0.22, F1,220 = 5.36; P < 0.0001). We did not find significant sexual dimorphism in the morphological traits used in this study. In general, spine length increased both with depth and with distance from the lava fields in the north end of the lake, the nitella morph had the maximum observed spine length (site 9) and the lava morph (sites 2 and 3) had the shortest spines. Both habitat type and water depth significantly explained variation in spine length (Table 2). Fish caught in shallow waters were generally deeper bodied with fewer, shorter gillrakers than the groups at depths of 10 m and more. Water depth was a better predictor of variation in body depth and gillraker morphology than habitat type (Table 2).

Table 2.   Results from nested anova showing effects of sample site nested within habitat type and water depth on phenotypic variation.
Body depthHabitat (sample site)90.0300.0050.5840.743
Water depth40.1520.15217.626< 0.001
Spine lengthHabitat (sample site)90.2990.053.322< 0.001
Water depth40.1840.18412.238< 0.001
Gillraker numberHabitat (sample site)90.0300.0050.8520.531
Water depth40.0670.06711.527< 0.001

Population structure and individual assignment

The assignment test clearly detected the two lava and nitella morphs, the most likely number of groups (k) was 2 (P = 0.998). More than 90% of fish were correctly assigned to these groups with high probability (> 0.95). When using prior population information on lava and nitella morphs and including the possibility of mixed ancestry, 92.2% of the fish sampled at the main nitella site were classified as nitella fish, 2.4% as hybrid and 5.4% were unassigned (assigned to an undefined third group). At the lava end of the zone 83.3% were assigned as lava morphs and 16.7% identified as possible hybrid (Fig. 3). At intermediate sites the percentage of fish assigned as parental lava or nitella morphs did not exceed 70% and the number identified as having mixed ancestry varied from 0% to 45%. At sites 7, 11, 12 and 13 a high proportion of individuals could not be assigned to either of the defined parental morphs, suggesting that the population structure of stickleback within the lake may be more complex than the previously described lava morph–nitella morph divergence. Parental lava fish were only observed at sites close to the lava habitat (sites 2–5). The nitella morph was found at sites 4–12 and the frequency of nitella fish was high outside the actual nitella habitat or as much as 66% at site 4 (Fig. 3).

Figure 3.

 Multilocus assignment of individuals to lava morph, nitella morph or to mixed ancestry. The percentage of unassigned fish is also given, as not all individuals could be reliably classified to any of the three groups.

Signs of selection

The ln RV test indicates significantly reduced variation at locus Gac2111 and ln RH at locus Gac3133 (Fig. 4). At both loci there is less variation in the lava morph. The other three loci potentially linked to adaptive variation are not significant outliers, although both Gac7033 and Gac4174 lie near the edge of the distribution. The lack of significant differences in variation between the lava and nitella morph does not rule out selection at linked loci; symmetrical divergent selection would not necessarily be identified with these tests.

Figure 4.

 ln RV and ln RH values across all loci. Values represent the natural logarithm of the ratio of allele size variance and expected heterozygosity between the nitella and the lava sticklebacks. Values at loci Gac2111, Gac3133 lie outside the 0.99 confidence interval on a normal distribution.

Mantel analysis

The matrix comparisons are summarized in Table 3. Mantel analysis of presumed neutral genetic variation including all predictor matrices provided high partial regression coefficients for the predation barrier and geographical distance. Depth explained most of the variation at Gac7033 (partial regression coefficient =0.452, P < 0.01) and Gac2111 (partial regression coefficient = 0.306, P < 0.001). These loci have been linked to variation in spine length and gillraker number in other stickleback populations (Peichel et al., 2001). As the same morphological characters are highly divergent between morphs in Lake Thingvallavatn these results imply parallel genetic mechanisms underlying adaptation in stickleback population across continents. However, depth did not significantly explain variation at the microsatellite markers linked to variation in lateral plates (Gac4174 and Gac1125). Distance measures at these loci seem to follow a similar pattern as the presumably neutral microsatellites. Geographical distance predicted phenotypic variation in anti-predator morphology, but depth was the best predictor of variation in trophic phenotype.

Table 3.   Partial regression coefficients from Mantel analyses. Distances are based on neutral and putatively selected microsatellite variation and two morphological distance measures. The independent factors are geographical distance and two matrices representing ecological variation, predation barrier and depth.
Distance calculated fromGeographical distancePredation barrierDepth
Neutral microsatellites
 Partial regression coefficient0.3040.401−0.288
 Probability< 0.05< 0.001NS
Locus Gac7033 (Gillrakers; 2.dorsal spine)
 Partial regression coefficient0.1240.1220.452
 ProbabilityNSNS< 0.01
Locus Gac2111 (1.dorsal spine)
 Partial regression coefficient−0.0550.1700.306
 ProbabilityNSNS< 0.001
Locus Gac4174 (lateral plate number)
 Partial regression coefficient0.2840.322−0.216
 ProbabilityNS< 0.05NS
Locus Gac1125 (lateral plate width)
 Partial regression coefficient0.3190.4760.070
 Probability< 0.05< 0.001NS
Armour morphology
 Partial regression coefficient0.3760.245−0.205
 Probability< 0.05NSNS
Trophic morphology
 Partial regression coefficient0.011−0.0490.561
 ProbabilityNSNS< 0.001

Cline analysis

A summary of the fitted cline shapes and support limits for all characters are shown in Table 4. Cline centres and width could not be estimated for two of the nine loci, Gac7188 and Gac5196, presumably because their underlying pattern was not clinal. Estimates of the cline centre were comparable for the most common alleles and diagnostic alleles. Based on the most common alleles at each locus the position varied between 1550 m (∼ sample site 4), for locus Gac2111 and 6654 m (∼ sample sites 6–7), for Gac7033, measured from the north end of the lake. The most common position was approximately 3000 m south of the lava habitat, or close to sample site 5 (Table 4). For diagnostic alleles the position of the cline centre varied from 1454 m (∼ sample site 4), for locus Gac3133 to 6618 m (∼ sample site 6), for Gac7033. Diagnostic alleles at loci Gac3133 did not have a significantly reduced likelihood when keeping the position of their cline centres fixed.

Table 4.   Fitted cline centres and widths for each microsatellite loci and the two quantitative traits.
 Centre95 % CLWidth95 % CLln L
  1. 95 % confidence limits (CL) are given. The position of the centre is given in metres from site 1 (north end of lake). Log likelihood score is in the last column.

  2. *Gac2111 is linked to variation in QTL affecting dorsal spine length (1st spine) in Canadian sticklebacks.

  3. †Gac7033 is linked to variation in gillraker number and dorsal spine length (2nd spine).

  4. ‡Gac3133 was identified as potentially influenced by selection in Lake Thingvallavatn.

  5. Note that the diagnostic allele at locus Gac3133 also represents the most common allele. Missing values (–) represent traits where clines could not be fitted reliably and 95 % CL too wide to represent in the table.

Diagnostic alleles
 Gac111629112816–475497194754–12 214−5.07
Most common alleles
 Gac2111*1550834–367730 556−1.07
 Gac417430901993–365415 329−5.98
 Gac11251758928–483730 058−2.70
 Gac7033†66544981–794134 978−3.01
 Gac109731903002–486654 445−2.11
 Gac111624962342–288223 558−4.69
 Spine length41974065–477395414932–11 077−2.05

Based on the most common alleles, estimates of zone width generally lay well beyond the sampled area. The estimates based on diagnostic alleles fell within the limits of the divergence zone, although again there was considerable variation between loci. Loci Gac7033 and Gac3133 had a significantly different cline shape using the most common alleles (inline image = 3.86; P < 0.05). When comparing concordance of clines at the diagnostic alleles, all pair-wise comparisons were significantly different. The two quantitative characters had very different cline patterns. Variation in spine length had a similar cline width and centre as the microsatellite markers (Table 4). However, gillraker number, although variable across sites, did not conform to a clinal pattern (Fig. 5).

Figure 5.

 Clinal variation of morphological traits and diagnostic microsatellite alleles across the lava = nitella divergence zone. The y-axis represents allele frequencies and scaled measures of phenotypic traits.

Linkage disequilibrium

With the multilocus disequilibrium measure inline image (Agapow & Burt, 2001) four sites showed significant multilocus disequilibrium. These were at the intermediate sites 4–6 and 8. Similarly, linkage disequilibrium estimates (D*) based on the covariance of phenotypic traits were estimated for all individual sites. Covariance of phenotypic traits was significant at sites 5 and 6. Both measurements gave the highest linkage disequilibrium values at intermediate sites (Fig. 6).

Figure 6.

 Multilocus disequilibrium estimator rd, and disequilibrium between dorsal spine length and gillraker number D*.


The putative lava and nitella stickleback morphs from Lake Thingvallavatn in Iceland can be clearly identified using multilocus genotypic assignment. However, several fish were classified as having mixed ancestry and several factors were consistent with a primary transition zone between the morphs. The results indicate that dispersal of lava individuals outside the lava zone is limited, the lava genotype was not found south of site 5. Conversely, nitella fish are present in considerable density outside the central nitella habitat including sites very close to the parental lava population (Fig. 3). The Mantel analysis and examination of cline shapes demonstrate a complex pattern of variation with considerable differences amongst neutral and putatively selected traits, which respond differently to habitat, predation and geographical distance. The role of selection in the divergence of the morphs is further supported by tests using variation in allele number and heterozygosity. These identify two outliers, one of these has been linked to variation in dorsal spine length in other stickleback populations.

Predation facilitates divergence

Predation is likely to play an important role in the divergence of the Lake Thingvallavatn stickleback morphs. There are spine length differences between the morphs. The lava morph has much shorter spines than the nitella morph, suggesting that the morphological divergence has been affected by reduced predation pressure in the lava habitat. However, there is evidence that predation pressure has both direct and indirect effects on divergence between the morphs. Highest densities of sticklebacks in Lake Thingvallavatn are found in habitats that provide ample cover from predators. The physical barrier of a predation belt, where there is little available cover, might greatly reduce the dispersal of sticklebacks, reducing gene flow and thereby aiding divergent selection between the groups. The Mantel analysis provides strong support for the role of a predation barrier in limiting neutral gene flow. Recently, the role of direct natural selection against migrants has received considerable interest in studies of adaptive divergence. For example, in walking stick insects, predation on less cryptic migrants in intermediate host habitats directly limits movement of individuals between hosts and thereby contributes to the divergence of the species. This process has a stronger effect on premating isolation than does sexual selection (Nosil, 2004). Although unlikely to cause complete isolation, mortality of migrants can result in rapid divergence of populations between habitats (Hendry, 2004). However, a recent study finds no evidence for reduction in gene flow between guppies separated by high predation habitats, although predation may limit short-term dispersal between the sites (Crispo et al., 2006). This highlights the potentially contemporary role of predation in population divergence and speciation, as several other factors are likely to contribute.

Our results suggest that predation pressure and perhaps a level of habitat fidelity contribute to the maintenance of locally adapted stickleback morphs. However, the movement of individuals across the predation barrier is likely to depend on various ecological conditions, including population densities of both stickleback and predators. For these morphs to remain distinct (or to complete speciation) a mechanism contributing to the sexual isolation of the morphs must evolve. Experiments on mating preferences of the Lake Thingvallavatn sticklebacks show that there is assortative mating between the lava and nitella stickleback morphs (Ólafsdóttir et al., 2006a). The divergence of lava and nitella morphs may represent the early stages of speciation, facilitated by habitat isolation caused by predation.

Selection and parallel evolution

In cases of potential adaptive divergence neutral molecular markers give only limited understanding of the evolutionary processes. Identifying genetic variation effected by divergent selection in natural systems is needed to aid understanding evolution of genetic architecture. Molecular markers can allow us detect different levels of variation and divergence on different regions on the genome, potentially identifying regions subject to selection (Storz, 2005; Vasemägi & Primmer, 2005).

In the zone described here, there are differences in cline shape both between markers and morphological traits. Based on variation at the most common allele clines were unrealistically wide, except for locus Gac3133 (Table 4). This is most likely caused by the low level of divergence between morphs. Based on diagnostic alleles, loci Gac7033 and Gac3133 have a notably narrower cline. Allelic frequencies at locus Gac3133 are highly diverged between the lava and nitella morph and it is identified as an outlying locus with tests of ln RV and ln HV. The steep cline of locus Gac3133 and the position of its centre, very close to the lava habitat, imply selection pressures acting on this area of the genome (although we do not know if this locus is linked to phenotypic traits). The other locus identified in this study as potentially linked to selected variation, Gac2111 has been linked to adaptive variation in Canadian sticklebacks and, although it has not been directly linked to variation in phenotypic traits in this study, the Lake Thingvallavatn morphs differ in the same traits, i.e. spine length and gillraker number. Given the low divergence of the morphs, the lack of diagnostic alleles and low sample sites at intermediate sites, care should be taken when interpretating the results from the cline analysis. However, there are two noteworthy results. (1) The centre positioning at the intermediate habitat between the lava and nitella is well supported and (2) there are large differences in cline widths among markers, especially when considering that the allele variation used at locus Gac3133 also represents the most common allele. In concordance with these results, the Mantel analysis suggest that different levels of selection are acting on different loci. Variation at loci putatively linked to phenotypic variation is predominantly explained by ecological factors. These results may therefore be an exciting example of parallel evolution of the same genomic regions underlying adaptation and divergence of stickleback populations. Parallel genetic mechanisms involved in adaptive divergence of threespine stickleback populations, both within North America and between opposite sides of the Pacific Ocean, have been confirmed in genetic mapping and inheritance studies (Colosimo et al., 2004; Cresko et al., 2004; Schluter et al., 2004). Although there are clear parallelism between the Lake Thingvallavatn stickleback morphs and other studied systems of stickleback divergence (e.g. Canadian limnetic–benthic morphs) both in morphological and in genetic divergence, it is important to point out that there are also some notable differences (Table 5). In many aspects the lava morph resembles the Canadian benthic morph (deeper bodied, reduced armour structure) and the same is true of the nitella and limnetic morph (slender, smaller head, longer spines). However, in the Canadian stickleback pairs the limnetic form has longer and more numerous gillrakers than the benthic morph. Gillrakers are important in the feeding ecology of fish and the observed differences in the Canadian species pairs are thought to represent differences in trophic resource use (McPhail, 1994). The stickleback morphs in Lake Thingvallavatn do not have discrete diet preferences, largely feeding on the same food items, although the lava fish are perhaps a more specialist feeder (Kristjánsson et al., 2002). There is no divergence in lateral plate number among the lava and nitella morph and correspondingly the loci linked to variation in this trait do not appear highly diverged between the morphs. The lack of excessive divergence at loci proposed to be linked to QTL for traits that have diverged between the Lake Thingvallavatn morphs, for example the lack of cline pattern at locus Gac2111, can potentially be explained by the overall low level of variation at these loci. Therefore it is not necessarily indicative of the lack of selection, indeed selection may have caused reduced variation at these loci in the colonizing fish. Low variation, indicative of selection, has been detected at loci Gac2111 and Gac7033 in Icelandic populations of marine stickleback (Ólafsdóttir et al., 2006a,b). However, it is also possible that the microsatellite markers are not closely linked to QTL to detect a selective signal. Cano et al. (2006) find that locus Gac4174 shows a similar divergence pattern to neutral microsatellite markers in plate morphs of European stickleback despite this locus being very strongly associated with plate number in crosses of North American and Japanese plate morphs (Colosimo et al., 2004). We also detect signs of selection at loci with no known linkage to phenotypic traits. This is not unexpected as the morphs have diverged at several traits other than those with identified QTL.

Table 5.   Summary of test results. ‘Yes’ indicates test results compatible with selection on linked loci.
  1. *Gac3133 was not used individually in the Mantel analysis, as there were no prior predictions of linkage at this locus.

Potentially linked to analysesSpine lengthGillraker length, spine lengthPlate sizePlate morph, plate sizeNo known linkage
Mantel testYesYesNoNoNA*
ln RV − ln RHYesNoNoNoYes
Cline shapeNoYesNoNoYes

Adaptive divergence

Little is known about the genetic details of most cases of adaptive divergence. In some models divergence is facilitated by few ‘speciation genes’ while in others the associations built up across many selected loci contribute to the process of speciation (Barton & Turelli, 1989). In many threespine stickleback populations it has been shown that a single ecologically important character, body size, mainly accounts for initial assortative mating and consequential population divergence (McKinnon et al., 2004). In the Lake Thingvallavatn system the lava and nitella morphs have diverged at several phenotypic traits (Kristjánsson et al., 2002; Doucette et al., 2004) including assortative mating (Ólafsdóttir et al., 2006a). We also observed elevated linkage disequilibrium and a clearly identifiable centre of the divergence zone. In zones of secondary contact linkage disequilibrium will be at a maximum at the centre of the zone and the rate of breakdown will represent the strength of selection acting on the zone. Conversely, when studying clinal patterns formed by primary parapatric divergence, divergent selection is more likely to be acting on only few areas of the genome and the populations may not have accumulated sufficient neutral differences to cause strong linkage disequilibrium across the cline (Barton & Gale, 1993). The pattern of linkage disequilibrium observed here suggests that the lava and nitella morphs have already diverged at several loci and that sufficient time has passed to create associations between loci across the genome. This is in accordance with phenotypic measurements that show the morphs have diverged in several traits. Although it is possible that only a few loci contributed to the initial divergence of the lava and nitella morphs, given the very recent time scale our results indicate that several phenotypic characters with independent underlying loci are involved with this parapatric divergence from the outset. A fundamental problem in studies of the genetics of speciation is distinguishing between genes acting in speciation and genetic differences accumulated after the initial divergence. Our results show that recent systems like the Lake Thingvallavatn stickleback morphs can potentially give important information on the genetics of species divergence.


We would like to thank Theódór Kristjánsson, Bjarni Kr. Kristjánsson, Hrefna Berglind Ingólfsdóttir and Anna Kristín Gunnarsdóttir for help with fieldwork, Freyr Ævarsson for help with morphological measurements. This work was supported by grants from The Icelandic Research Council to G.Á.Ó. and S.S.S. and The Research Fund of the University of Iceland to S.S.S.