Guðbjörg Ásta Ólafsdóttir, Institute of Biology, University of Iceland, Sturlugata 7, Reykjavik IS101, Iceland. Tel.: +354 525 5230; fax: +354 525 4281; e-mail: firstname.lastname@example.org
The geographical context of divergence and local adaptation of lacustrine fish is controversial. Despite recent theoretical support for sympatric and parapatric divergence, empirical studies providing unequivocal support for this remain scant. An important component of such a case would be where multiple lakes have different morphs and a range of markers, both mitochondrial and nuclear, show monophyly within lakes. Here we describe such a situation for threespine sticklebacks in three lakes in Iceland. By analysing the variation at nuclear and mitochondrial markers in several freshwater and marine populations as well as three pairs of intra-lacustrine morphs we infer their phylogenetic relationships and colonization pattern. There were high levels of microsatellite variation in all populations and no evidence was found for either repeated colonization of marine fish or colonization from distinct glacial refugia. Intra-lacustrine threespine stickleback morphs in all three lakes show significant genetic divergence probably indicating restricted gene flow.
Following the retreat of the Pleistocene glaciers, 10 000–16 000 years ago, numerous freshwater systems were formed in the northern regions of North America and Eurasia. These were colonized by fish either by sea or from glacial refugia, greatly limiting the number of colonizing species. Moreover, changes in sea levels following the glacial retreat, as well as newly formed obstructions on river–lake passageways, could result in geographical isolation of one or few fish species within individual lakes or water systems. These recently colonized lakes would therefore have many island-like features, allowing for independent evolution within each system (Robinson & Schluter, 2000). The various unoccupied habitats within lakes coupled with intraspecific competition can give rise to rapid specialization and resource-based polymorphism. More than one morph is commonly found within lakes (Snorrason et al., 1994; Skúlason & Smith, 1995; Robinson & Schluter, 2000). Most often two morphs coexist adapted to the different foraging habitats of the benthic and pelagic zones. Examples are found in whitefish (Coregonus spp), sunfish (Lepomis spp.), Salmonids (Salmo, Salvelinus spp.) and sticklebacks (Gasterosteus). Often a remarkable amount of phenotypic variation exists between morphs with varying levels of genetic divergence found between these ecologically and potentially reproductively distinct forms or species.
Phylogeographical approaches might be most useful in situations where the recent geological history of an area is relatively well known. This will increase the likelihood of identifying the actual colonization and divergence pattern. Iceland offers an excellent opportunity to study sympatric divergence in lake morphs. Lakes are extremely recent (less than 10 000 years old) with a well-understood geological history and records of fluctuations in sea levels. During the Weichselian glaciation Iceland was almost completely glaciated, excluding only small areas. The land rose rapidly following the final retreat of the Weichselian inland ice 10 000–11 000 years ago, with subsequent fluctuations in water levels unlikely to result in repeated colonizations of the lakes included in the present study (Ingólfsson & Norddahl, 2001).
Threespine stickleback populations are often characterized by rapid phenotypic and genetic divergence following colonization of freshwater by marine sticklebacks (McKinnon & Rundle, 2002). Sympatric morphs that are divergent in morphological traits, behaviour and are sometimes genetically distinct have been found. The best documented of these are the benthic–limnetic species pair in postglacial lakes in British Colombia, Canada, but several other examples exist including lake-stream morphs, sympatric anadromous and freshwater populations and different colour morphs (reviewed in McKinnon & Rundle, 2002). Recently, sympatric and parapatric stickleback morphs have been found in Icelandic neovolcanic lakes (Kristjánsson et al., 2002). The morphs inhabit different habitats, a complex structured lava substrate or a typical benthic mud/vegetation habitat. They differ considerably in phenotype. The lava morph has shorter spines, is deeper bodied, has larger heads and eyes and smaller fins (Kristjánsson et al., 2002).
This study uses phylogeographical approaches to address questions concerning the origin of Icelandic intra-lacustrine stickleback morphs. (1) Do the intra-lacustrine morphs represent genetically distinct groups? (2) Is there evidence for colonization from multiple glacial refugia? (3) Are the intra-lacustrine morphs monophyletic or are they likely to originate from repeated marine invasions? Here, results from genetic analysis using both nuclear and mitochondrial markers are presented. The combined evidence of phylogenetic trees based on both nuclear and mitochondrial DNA and the well-documented geological history of the region provides convincing evidence for the genetic divergence of intra-lacustrine morphs following rapid colonization by a single marine ancestor.
Sample sites and collection of samples
We chose sample sites from different watersheds and marine locations distributed across Iceland, varying in altitudes and distance from the sea. Moreover, sticklebacks were collected from three lakes situated in an area of recent geological activity: Lake Thingvallavatn, Lake Myvatn and Lake Hredavatn. Different levels of phenotypic divergence have previously been demonstrated between fish caught at lava and soft substrate habitats in these three lakes (Kristjánsson et al., 2002). In addition to samples from the three proposed intra-lacustrine morph pairs, we sampled seven other freshwater populations and four marine populations (Fig. 1 and Table S1). Fish were collected using unbaited benthic minnow traps except in the lava habitat in Lake Thingvallavatn where electrofishing was used. Sampling was made in the summers of 1999–2002. All fish were anaesthetized, fin clipped and preserved in 10% formaldehyde. Fin clippings were preserved in 96% ethanol for genetic analysis. A total of 458 fish are used in this study, with sample size per site ranging from 11 to 62.
Total genomic DNA was isolated from fin clips using a standard proteinase K lysis and organic solvent purification method as described in Sambrook et al. (1989). Briefly 1–2 g of tissue was digested overnight with proteinase K at 55 °C. DNA was then extracted by treatment with phenol, phenol/chloroform/isoamyl alcohol (25 : 24 : 1) and chloroform/isoamyl alcohol (24 : 1). Adding one-tenth volume 3 m sodium acetate and two volumes of absolute ethanol achieved precipitation. The DNA was then washed in 70% ethanol before being resuspended in 50 μL of sterile distilled water.
Samples were scored for seven microsatellite loci: Gac2111, Gac4174, Gac3133, Gac7188, Gac7033, Gac1125 and Gac1097 (Largiadér et al., 1999). The loci Gac2111, Gac7033, Gac4174 and Gac1125 are adjacent to quantitative trait loci effecting morphological variation in threespine sticklebacks in other populations, although with a distance of up to 3 cM (Peichel et al., 2000; Colosimo et al., 2004). At least two of these may be affected by selection in populations of Icelandic threespine stickleback (Ólafsdóttir et al., 2007a,b). This is not expected to make them less informative for detecting monophyly of intra-lacustrine morphs (Avise, 2004). However, microsatellite analysis was repeated after excluding the loci possibly affected by selection. As this did not give notably different results for population relationships, the results presented in this article are based on all seven markers.
Conditions described in Largiadér et al. (1999) were followed for amplification of markers except for Gac7033 where the annealing temperature was raised by 1 °C. PCR products were mixed with formamide loading dye and heated to 95 °C before being loaded onto a 6% polyacrylamide gel. A 10 bp ladder was loaded in every tenth well and samples of known allele size were used as additional standards. The gels were run at 50 W for 3–4 h and the products then visualized with silver staining.
The entire mtDNA control region was sequenced for a random subset of fish from each population in total 42 individuals, sample size range from 1 to 10 per population. We used the primer pair L-Thr (5′-AGCTCAGCGYCAGAGCGCCGGTCTTGTAA-3′) and H-12S (5′TAAAGTCAGGACCAAGCCTT-3′), modified by Takahashi & Goto (2001) to amplify the total mtDNA control region. The following thermal cycles were used: 94 °C for 30 s; 94 °C for 1 min, 58 °C for 1 min, 72 °C for 1 min (30 cycles); 72 °C for 8 min, which amplified a 900–1030 bp fragment from total genomic DNA. The PCR product was cleaned and isolated using Qiagen PCR-cleanup kits (Qiagen, Dorking, UK). DNA sequencing was performed by The Sequencing Service (http://www.dnaseq.co.uk) using DYEnamic ET terminator chemistry (Amersham Pharmacia Biotech, Piscataway, NJ, USA) on Applied Biosystems (Foster City, CA, USA) automated DNA sequencers. Owing to their high similarities, sequences were easily aligned by eye and compared to published threespine stickleback control region sequences. Variable insertions/deletions were observed at the 5′ hypervariable end. These consisted of 11 bp, ATAGGCGCCAA, repeated 4–16 times causing a maximal size difference of 130 bp between sequences.
Genetic variation and population divergence
Population genetic variability, the number of alleles per locus, expected and observed heterozygosity (HO, HE) was estimated using genetix vs. 4.02 (Belkhir, 2000). The same program was used to test for Hardy–Weinberg (HW) equilibrium and linkage disequilibrium. All multiple comparisons were Bonferroni-corrected to maintain a Type 1 probability error at α = 0.05 (Rice, 1989).
Population divergence based on microsatellite loci was examined by computing FST values (Weir & Cockerham, 1984) in genetix 4.0 (Belkhir, 2000) and RST values with the program rstcalc (Goodman, 1997). Population subdivision was also measured using FST estimates from a Kimura two-parameter distance matrix based on mtDNA control region sequence data calculated with arlequin version 2.0 (Schneider et al., 2000). The significance of these estimates was determined by a 1000 step, 1000 iteration, Markov chain Monte Carlo method (1000 dememorization steps). Considering the very short divergence time of these populations (∼10 000 years) genetic distance estimates that are based on drift rather than mutation are probably most appropriate (Takezaki & Nei, 1996; Goldstein & Pollock, 1997; Gaggiotti et al., 1999).
To further examine population divergence within lakes we used the program structure version 2.1 (Pritchard et al., 2000) to identify the number of genetically distinct clusters (k) that minimized HW and linkage disequilibrium. For each value of k, five iterations were run, with a burn-in period of 100 000, followed by 200 000 iterations for values of k = 1–5. The simulation was made with a model assuming admixture and correlated allele frequencies. The ad hoc statistic ΔK was used to identify the most likely number of clusters within each lake (Evanno et al., 2005).
Several of the freshwater sites sampled are at altitudes of between 400 and 600 m a.s.l. This could limit the number of colonizing individuals and generate founder effects. Reductions in effective population size (Ne) are accompanied by reductions in the number of alleles and expected heterozygosity. The number of alleles is expected to be reduced more quickly, especially those which occur at a low frequency (Cornuet & Luikart, 1996; Luikart et al., 1998). A population that has experienced a recent bottleneck is therefore expected to show a greater HE than predicted based on the observed alleles. To assess the impact of this type of demographic change we used the heterozygosity excess test of Cornuet & Luikart (1996). Heterozygosity excess was estimated and tested for significance (mode shift test) using the computer program bottleneck (Piry et al., 1999).
Phylogenetic relationships based on microsatellites were assessed using the modules seqboot, contml and consense in phylip 3.5 (Felsenstein, 1993) to construct a maximum likelihood tree (1000 bootstraps) based on microsatellite allele frequencies. To facilitate comparison with the mtDNA haplotype tree, the tree was rooted by one of the marine stickleback groups (5 Myrar M). Marine sticklebacks are generally thought to represent the genetic diversity found in a hypothetical ancestral population as freshwater populations of threespine stickleback have resulted from recent colonization of marine stickleback (Hagen & McPhail, 1970). This view has been supported by patterns of variation of mtDNA, allozymes and microsatellites in several recent northern freshwater and marine populations (Withler & McPhail, 1985; Taylor & McPhail, 1999, 2000).
The program Modeltest 3.7 (Posada & Crandall, 1998) was used to choose among 56 substitution models of sequence evolution that best fit the data. The proportion of invariant sites (I) was 0.6319 and the gamma distribution shape parameter was 0.8706. The most likely model was F81 + I + G. The estimates from Modeltest were subsequently used to generate a maximum likelihood phylogenetic tree using paup* (Swofford, 1998). The robustness of the branching pattern of the trees was tested by 1000 bootstrap replicates.
Genealogical relationships among mtDNA sequences were also examined by constructing a haplotype network with the parsimony method of Templeton et al. (1992). This method estimates the minimum number of substitutions to connect any two haplotypes and can provide high resolution for inferring relationships among genes with low levels of divergence. It also estimates haplotype outgroup probabilities, allowing identification of the most ancient haplotypes in the sample. The analysis was performed using the software TCS version 1.06 (Clement et al., 2000).
Analysis of molecular variance (amova) was used to test hypotheses of colonization patterns and the origin of intra-lake pairs. The levels and significance of genetic variation based on different groupings of the populations reflect the structure of genetic variability. The arrangement that maximizes variation between groups was presumed to represent the true geographical structure. Several hypotheses were tested. (1) North–South division of colonization (groups based on N–S watersheds). (2) Earlier highland invasion (groups based on sampling site altitude). (3) Double invasions of lakes (groups based on different morphs). amova for both microsatellite and mtDNA data was calculated in arlequin version 2.0 (Schneider et al., 2000).
Microsatellite genetic variation was high across all populations and loci. Allelic diversity ranged from 18 to 58 alleles per locus and mean heterozygosity from 0.39 to 0.87 (Table S1). Marine stickleback had the highest microsatellite variation, with values corresponding to those reported in a review of 12 species of marine fishes (DeWoody & Avise, 2000). Reduced genetic variation in populations might be expected during colonization of highland lakes. However, microsatellite variation as measured by HE was not related to either altitude or distance from the sea. We did not find evidence of significant bottleneck effects during colonization. Tests for heterozygosity excess yielded no significant values.
A small percentage of the alleles found were exclusive to the freshwater populations (8 of a total of 220 alleles, or 3.6%). These private alleles were found in the populations in Lake Myvatn, Lake Thingvallavatn and Lake Hredavatn, which are the largest freshwater samples. When comparing the freshwater populations only, the intra-lacustrine morphs in Lake Thingvallavatn were found to share six private alleles; five private alleles were observed in Lake Myvatn and the morphs in Lake Hredavatn shared three private alleles.
The HW equilibrium was rejected in 10 of 139 possible tests, but only three remained significant after sequential Bonferroni adjustments. These tests did not show any pattern among loci or populations. Significant linkage was observed in 12 of 255 pair-wise tests. Although this may be accounted for by multiple comparisons, 4 of these 12 instances were found within the soft substrate population in Lake Thingvallavatn, three each in the freshwater populations in Hraunsfjordur and two in the Hnifa population. The observed linkage disequilibrium might reflect real genetic structure within these populations, e.g. sampling at these sites might have included individuals which originated in genetically distinct but geographically close populations.
A total of 28 mitochondrial haplotypes were identified in the sample of 42 individuals. The distribution of haplotypes reflected geographical distribution and haplotypes shared among localities were rare. In all, 33 of 1059 sites were variable, not considering the variable number of repeats observed at the hypervariable 5′ end. Nucleotide diversity for the combined sample was 0.046 ± 0.012 and haplotype diversity was 0.9751 ± 0.0076.
STRUCTURE identified two clusters as the most likely population structure within each of the three lakes, Lake Thingvallavatn, Lake Myvatn and Lake Hredavatn. The estimated likelihood was consistent between runs for each k (1–5) and all runs at k = 2 produced identical clustering solutions. In all lakes the clusters correspond well to predefined morphs, with the majority of individuals from the same morph (sample site) clustering together. Few individuals were identified as recent hybrids (Fig. 2). The ln probability of data plateaus at k = 2–4 in Lake Thingvallavatn and Lake Hredavatn and at k = 2 and 3 in Lake Myvatn; however, the ad hoc statistic ΔK (Evanno et al., 2005) found k = 2 to be the most likely number of groups in all three lakes. Genetic structuring was not observed at other sites.
FST values based on microsatellite loci range from 0.042 to 0.46 (average 0.21), and all pair-wise tests were significant (P < 0.01). Fixation index values between allopatric populations range from 0.089 to 0.46, and are generally very low between freshwater and the closest marine population (FST range = 0.089–0.211), as might be expected given the recent colonization. Intra-lacustrine morphs differ significantly in all cases, suggesting that the intra-lacustrine morphs represent at least partially isolated gene pools. The level of divergence for sympatric pairs is generally low compared to allopatric populations, the FST values lie between 0.042 and 0.082, indicating low to moderate divergence. Divergence estimates based on microsatellite stepwise mutation models (RST) give similar results (Table 1). FST values based on mtDNA haplotypes were significant for intra-lacustrine morphs in two lakes, Lake Thingvallavatn and Lake Myvatn (Table 1).
Table 1. Genetic population divergence between intralacustrine morphs in the three lakes. FST and RST values based on microsatellite allele frequencies and FST values based on a Kimura two-parameter distance matrix of mtDNA control region sequences.
All values are highly significant except for Lake Hredavatn where only a single mtDNA haplotype was observed.
*P < 0.0001.
All intra-lacustrine morphs form a monophyletic group based on microsatellite allele frequencies. The maximum likelihood tree based on microsatellite allele frequencies separates Lake Thingvallavatn, Hraunsfjordur and the marine groups from the other freshwater sites (Fig. 3). Although intra-lacustrine morphs remain monophyletic based on the mitochondrial phylogeography, there are slight discrepancies between the trees (Figs 3 and 4). Based on mitochondrial variation Lake Thingvallavatn is not grouped so closely with the marine fish. Neither tree shows any patterns that can be related to separate origins of, for example, highland–lowland or north–south threespine stickleback populations.
The haplotype network (Fig. 5) is based on the first 750 bp of the 3′ end of the control region because complex insertions/deletions at the 5′ hypervariable end resulted in very low resolution and multiple networks when using the complete sequences. The haplotype network might therefore be expected to depict the relationships of fish more conservatively than the mtDNA tree that includes all information at the hypervariable end. The haplotypes cluster according to geographical location and haplotypes are not commonly shared between sites. There is no clear distinction between sympatric morphs; these cluster together in all cases and in Lake Hredavatn they share a single haplotype.
amova was based both on microsatellite allele frequencies and on Kimura two-parameter distance estimates calculated from sequence data. Several hypotheses were tested. (1) North–South division of colonization (groups based on N–S watersheds). (2) Earlier highland invasion (groups based on sampling site altitude). (3) Double invasions of lakes (groups based on different morphs). (4) Single invasions of each lake (each group containing only one or two sites, sites where grouped when separated by less than 20 km). The between- group variance was maximized when grouping the populations into individual lakes, both for microsatellites (FST = 0.16) and mtDNA (FST = 0.29). The between-group variance was lowest when sites where grouped by morphs, i.e. lava group and soft substrate group, and was not significant for the microsatellite estimate. There was no support for the multiple colonization hypotheses (Table 2).
Table 2. Results from analysis of molecular variance. The arrangement that maximizes variation between groups was presumed to represent the true geographical structure.
Groups based on
Variation among groups (%)
Variation among populations within groups (%)
Variation within populations (%)
*Indicates the only nonsignificant value (at P < 0.05).
We find that previously described morphs in three Icelandic lakes are genetically distinct. Low but highly significant genetic structuring was detected in all three lakes using microsatellite markers and in all but Lake Hredavatn based on mtDNA control region haplotypes. Moreover, the results from this study consistently point towards the independent evolution of the different morphs within the lakes studied. All these lakes have complex and distinct habitats with the potential for significant ecological specialization. The morphs utilize habitats that differ extensively in several ecological factors including food availability, habitat stability and structural complexity (Kristjánsson et al., 2002; Ólafsdóttir et al., 2007a). No significant population structure was detected in the other freshwater lakes, but the possibility of sympatric morphs in other Icelandic lakes cannot be excluded, nor the possibility that sticklebacks have specialized to habitats other than the lava substrate, although we have focused on neovolcanic lakes in this study.
Multilocus individual assignment of intra-lacustrine morphs correctly classifies the majority of individuals. In Lake Thingvallavatn and Lake Hredavatn several individuals caught in the lava habitat were classified as belonging to the mud morph and vice versa, indicating that dispersal between habitats is not uncommon in these lakes. However, few individuals were identified as recently admixed in these two lakes, whereas admixture is more noticeable in Lake Myvatn (Fig. 2). Parallel evolution with ecological divergence has been suggested as an explanation for the existence of the sympatric morphs that are found within several species of northern freshwaters (Pigeon et al., 1997; Thompson et al., 1997; Gíslason et al., 1999; Taylor & McPhail, 2000). The parallel evolution of the Icelandic lava and mud stickleback morphs suggests that ecological selection is a factor in the divergence of the morphs. The levels of genetic divergence between sympatric morphs broadly correspond to the extent of morphological differences found within the lakes: the highest morphological divergence was found in Lake Thingvallavatn but the lowest in Lake Hredavatn (Kristjánsson et al., 2002).
Demographic effects, such as a low number of founders and duration of small population size during colonization, are likely to leave a pattern that can be detected with genetic markers. However, inferring population relationships from genetic data in very recently established populations is difficult due to limited genetic differentiation and because populations may not be at equilibrium conditions. The lack of private alleles in the freshwater stickleback populations compared to the marine groups suggests that post-colonization mutations have not been important in structuring the observed genetic variability in Icelandic threespine stickleback. This means that most of the genetic structure observed at present results from reduced population size during colonization and post-isolation genetic drift and selection (Ólafsdóttir et al., 2007a). High levels of genetic variability are observed for all the stickleback populations, especially in the marine threespine stickleback populations. Variability was slightly reduced in the freshwater populations, as expected given the marine ancestry hypothesis; this is also compatible with the loss of variation during re-colonization (Nei et al., 1975; Hewitt, 1996). Some of the sampled populations are found at high altitudes, possibly limiting the number of colonizing individuals. However, none of the populations showed signs of significant bottleneck effects.
It is generally believed that during the last Pleistocene cold stage Iceland was completely glaciated. Several studies on the postglacial colonization of Scandinavia and North America find that colonization from multiple glacial refugia has been instrumental in forming the diverse fish morphs in recently deglaciated areas (Bernatchez & Dodson, 1990a;Nilsson et al., 2001). The current species of freshwater fish are often descended from two refugia either with a narrow contact zone, complete admixture, or they co-occur without interbreeding. We did not find any evidence for colonization from multiple glacial refugia. Estimates of genetic distance were low between all populations both for mtDNA and microsatellites, and the size distribution of microsatellite alleles was relatively uniform, as would be expected given recent divergence from a common ancestor. The observed pattern seems to represent recent rather than pre-glacial divergence, the island seems to have been colonized rapidly by a single population of marine sticklebacks as demonstrated by the generally high molecular variation and low level of population divergence.
Despite several convincing examples, sympatric divergence remains difficult to prove (Schliewen et al., 1994; Gíslason et al., 1999; Coyne & Orr, 2004; Barluenga et al., 2006a,b). First, it can be difficult to distinguish between within-lake divergence and a double invasion with subsequent gene flow between morphs. A similar genetic pattern would be observed if a second colonization had occurred recently enough for mixing of the two forms to be incomplete. For example, double invasions of freshwater habitats by marine stickleback is now suggested to be a fundamental part of the evolution of the Canadian limnetic–benthic stickleback pairs, as supported by the narrow distribution of these species (Taylor & McPhail, 2000). Moreover, it is almost impossible to reject historical range expansions and contractions including an allopatric period (Losos & Glor, 2003; Coyne & Orr, 2004). In this study both microsatellite and mitochondrial genetic variation generally support the monophyly of intra-lacustrine morphs, although a single individual from Lake Myvatn clusters with samples from Hraunsfjordur on the mitochondrial tree. The Hraunsfjordur fish are of a very recent marine origin, probably with ongoing gene flow with marine stickleback (Ólafsdóttir et al., 2007b). The position of the single Lake Myvatn individual could therefore possibly point towards a more recent marine invasion in Lake Myvatn, although this is not consistent for either morph within the lake. There is no other indication of either morph in any of the three lakes being more closely related to the marine fish, as would be expected given the double marine invasion hypothesis (Svärdson, 1961; Taylor & McPhail, 2000). Three individual haplotypes of the mud morph in Lake Thingvallavatn are distinct from the other Lake Thingvallavatn samples, thus Lake Thingvallavatn may have a more complex history. These three samples do not cluster with any of the other samples despite the sampling including a marine population from the same watershed (3 Stokkseyri M) as would be expected if the Lake Thingvallavatn mud morph represented a more recent marine invasion. The other two Lake Thingvallavatn samples mud haplotypes cluster with the Lake Thingvallavatn lava samples. To what extent this pattern might represent repeated invasions is difficult to say without extensive small-scale sampling within the water system.
In agreement with the results from the genetic analysis, the geological history of Iceland does not imply that multiple marine invasions of these lakes are likely. The lakes that we sampled are widely distributed across Iceland, not sharing any obvious postglacial geological history other than occurring on neovolcanic zones. Although the colonization history of each lake can never be fully known, the extremely recent formation of the lakes (< 10 000 years), coupled with geological data showing rapidly retreating glacial ice without subsequent fluctuations in water levels (Ingólfsson & Norddahl, 2001), make an allopatric phase in the evolution of the morphs highly unlikely. In any case, any such allopatric phase must have been of extremely short duration and followed by a much longer phase in sympatry or parapatry within the current structure of the lakes.
The scarcity of convincing examples of sympatric speciation suggests that the process is either rare or underestimated because of the difficulty in excluding alternate processes of divergence. The report of Gíslason et al. (1999) of monophyly of sympatric artic char morphs in four Icelandic lakes has provided one of the most convincing natural examples of sympatric divergence (Coyne & Orr, 2004). Here we propose a similar scenario in Icelandic stickleback. This could imply that either ecological conditions in Iceland facilitate intra-lacustrine divergence, or that the occurrence of sympatric speciation has been underestimated and the young age and geological conditions in Iceland make it easier to detect. Conditions in Iceland certainly provide an extreme example of factors suggested to facilitate sympatric divergence, e.g. high intraspecific competition due to the limited number of colonizing species and variable habitats in areas of recent geological activity.
Based on our results and the known geological history of Iceland, intra-lacustrine divergence is the most likely explanation of the occurrence of threespine stickleback morphs in Icelandic lakes. However, this does not exclude more small-scale effects contributing to divergence within lakes. For example, in the African Great Lakes complex habitats result in virtually isolated populations of cichlids despite geographical proximity, and several studies emphasize microallopatric processes in intra-lacustrine population divergence (Arnegard et al., 1999; Markert et al., 1999; Turner, 1999; Danley et al., 2000). The lakes studied here are relatively small with no apparent physical barrier between habitats. Nevertheless, the complexity and patchiness of substrate structure within the Icelandic neovolcanic lakes could potentially contribute to a process similar to that seen in the African cichlids. Moreover, there are processes other than a physical barrier that could potentially contribute to limiting gene flow between morphs in different habitats during the initial divergence, including mortality of migrating individuals (Hendry, 2004). Results from a more fine-scale analysis of the stickleback in Lake Thingvallavatn indicate that high predation pressures may act to reduce the movement of individuals between habitats (Ólafsdóttir et al., 2007a). Predation is not known to limit movement between habitats in the other lakes. The Lake Thingvallavatn morphs are more genetically and morphologically distinct than at the other sites, so reduced movement may facilitate divergence after ecological differentiation.
When studying the early stages of population divergence, finding low genetic divergence is often taken to reflect ongoing gene flow. The level of divergence between the sympatric stickleback morphs in Iceland is generally low to moderate, distance estimates ranging between 0.042 and 0.082 for nuclear markers and from 0.000 to 0.223 based on mtDNA sequences (Table 1). These values are consistent with the short time since colonization. The recent divergence is further emphasized by fixation indices between each lake population and the closest marine population: these range between 0.089 and 0.211. Despite the recent divergence, FST and RST values of sympatric morphs differ significantly from zero in all lakes suggesting that the intra-lacustrine morphs represent at least partially isolated gene pools. Furthermore, strong assortative mating has been found in one of the lakes, Lake Thingvallavatn (Ólafsdóttir et al., 2006), confirming that some sexual isolation has evolved between the sympatric morphs. Although results from mating trials are only available from this single lake, it shows that the low genetic divergence of the intra-lacustrine morphs does not necessarily indicate a lack of sexual isolation.
We conclude that there is compelling evidence for a recent intra-lacustrine divergence of lava and mud threespine stickleback morphs in Icelandic neovolcanic lakes. We found no evidence that colonization from multiple glacial refugia or multiple marine invasions explain the existence of these morphs. In Lake Thingvallavatn, the results from this phylogeographic approach are further supported by a more small-scale multidisciplinary approach (Ólafsdóttir et al., 2006, 2007a), indicating the potential for assortative mating and predation to contribute to divergence, following ecological specialization.
We thank Bjarni Kr. Kristjánsson, Theódór Kristjánsson, Árni Einarsson and the people at Lake Myvatn Research Station for help with collecting fish, Tino Macias Garcia and Jeff Graves for helpful comments on an earlier version of this manuscript, and Hreggvidur Norddahl for helpful comments on sea level fluctuations. 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.