The question of how diverging populations become separate species by restraining gene flow is a central issue in evolutionary biology. Assortative mating might emerge early during adaptive divergence, but the role of other types of reproductive barriers such as migration modification have recently received increased attention. We demonstrate that two recently diverged ecotypes of a freshwater isopod (Asellus aquaticus) have rapidly developed premating isolation, and this isolation barrier has emerged independently and in parallel in two south Swedish lakes. This is consistent with ecological speciation theory, which predicts that reproductive isolation arises as a byproduct of ecological divergence. We also find that in one of these lakes, habitat choice acts as the main barrier to gene flow. These observations and experimental results suggest that migration modification might be as important as assortative mating in the early stages of ecological speciation. Simulations suggest that the joint action of these two isolating barriers is likely to greatly facilitate adaptive divergence, compared to if each barrier was acting alone.
Rapid emergence of reproductive isolation has been predicted by theoretical models, which suggest that assortative mating can evolve rapidly and under a broad range of selective conditions (Yukilevich and True 2006). In contrast, when divergent selection is strong, migration modification might be more efficient in restraining gene flow and causing speciation than assortative mating (Yukilevich and True 2006). With the exception of one previous empirical study on salmonids (Hendry et al. 2000) which found that reproductive isolation could emerge as early as after only 13 generations, little is known about the temporal order and the rate of emergence of different isolation mechanisms during speciation (Nosil and Crespi 2006b; Rundell and Price 2009). Moreover, total reproductive isolation might also become weakened due to antagonistic interactions between assortative mating and other types of barriers to gene flow such as habitat choice, and these antagonisms might slow down the process of speciation (Yukilevich and True 2006; Hendry et al. 2007).
Here, we have estimated the strength and importance of assortative mating and migration modification during adaptive divergence between two ecotypes of the aquatic isopod Asellus aquaticus. This aquatic isopod is common in many lakes and ponds in southern Sweden. In two lakes (Lake Krankesjön and Lake Tåkern) independent oligotrophication events have taken place during the last two decades (Hargeby et al. 2004, 2007). These ecological shifts resulted in the emergence of submerged vegetation (mainly a stonewort, Chara tomentosa) which formed a new habitat in the limnetic zone of both lakes. The new stonewort habitat was rapidly colonized by isopods from neighboring reed belts (Phragmites australis) along the shores of both lakes (Hargeby et al. 2004). In less than 50 generations, isopods diverged phenotypically between these different habitats, resulting in the emergence of two distinct ecotypes (Eroukhmanoff et al. 2009a, b). Molecular analyses (mtDNA and AFLP-markers) indicate that the novel stonewort ecotype has evolved independently in the two lakes (Eroukhmanoff et al. 2009a). Pigmentation and body size have an additive genetic basis, both within (Harbeby et al. 2004) and between populations (Eroukhmanoff et al. 2009b). We also have indirect evidence (FST–QST analyses, Eroukhmanoff et al. 2009b) for a strong role for divergent selection, and at least pigmentation traits are under divergent selective pressures in the different ecotypes.
Adaptive divergence in this system is likely to be a result of predator-mediated natural selection, caused by qualitative and quantitative differences in predator faunas between the reed and the stonewort habitats (Hargeby et al. 2004; Eroukhmanoff and Svensson 2009). Because this diversification process is relatively recent and took place over a few decades, this system provides a unique opportunity to study the emergence and temporal order of different reproductive isolation mechanisms that might reduce ongoing gene flow in the early stages of population divergence and speciation. We have investigated the strength of assortative mating and habitat isolation between ecotypes and estimated their relative contribution to total reproductive isolation. We have also performed numerical simulations using previously estimated quantitative genetic parameters from these populations to estimate the relative importance of these two isolating barriers when operating either in isolation, or jointly. Our results and conclusions in this study should hopefully stimulate future research on other reproductive barriers in addition to assortative mating, such as migration modification, a factor that might have been overlooked in speciation research (Yukilevich and True 2006).
STUDY ORGANISM AND STUDY POPULATIONS
Asellus aquaticus is a freshwater isopod that is widespread in lakes, ponds, and slow-flowing rivers in Eurasia (Smock and Harlowe 1983). Populations of A. aquaticus occupy various habitats in lakes, and mainly occur in reed stands (P. australis) (Smock and Harlowe 1983). Two shallow Swedish lakes have (starting in 1987 in Lake Krankesjön and in 2000 for Lake Tåkern) experienced dramatic ecological shifts from a phytoplankton dominant state toward a macrophyte-dominant state (Hargeby et al. 2007). Following these large-scale environmental shifts, stonewort (C. tomentosa) colonized the old sediment areas, forming a massive area of submerged vegetation in the limnetic zone (Hargeby et al. 2007). Following the establishment of these extensive stonewort stands, isopods subsequently colonized this novel habitat in both Lake Tåkern and Lake Krankesjön, where they can be found at very high densities (Karlsson et al. 2010).
In the new stonewort habitat isopods became brighter and smaller in size, compared to darker and larger isopods in the source populations in the reed habitat (Eroukhmanoff et al. 2009a, b). Variation in body size and pigmentation brightness is largely heritable, with significant additive genetic variation both within and between populations (Hargeby et al. 2004; Eroukhmanoff et al. 2009b).
Local adaptation in isopod pigmentation is likely to have resulted from the action of divergent selection pressures, caused by different visual backgrounds and different predator faunas in the two different habitats (Hargeby et al. 2005). Several fish species are efficient predators on aquatic invertebrates (Wellborn et al. 1996) and A. aquaticus represents a common food source (Rask and Hiisivuori 1985). Predation from fish is likely to be more intense in the stonewort than in the reed habitat (Eroukhmanoff and Svensson 2009), due to much higher densities of perch (Perca fluviatilis) in the stonewort (Wagner and Hansson 1998). In contrast, in the original source habitat (reed), invertebrate predators relying on tactile cues (i.e., dragonfly and damselfly larvae) are the main threat toward the isopods (Eroukhmanoff and Svensson 2009). Recent molecular data suggest that this ecological diversification has occurred independently in these two lakes, suggesting that this system is a case of rapid contemporary parallel evolution (Eroukhmanoff et al. 2009a).
SAMPLING AND PHENOTYPIC ANALYSIS
Isopods were captured with a net on their original substrate and at multiple locations within their source habitats, in both Lake Tåkern and Lake Krankesjön during two reproductive seasons (February–June) in 2005 and 2006. We only used individuals captured as pairs in precopula, where the male holds the female until molt and receptive to mating. We did this to ensure that both males and females used in the experiments had reached sexual maturity. All individuals were photographed live in a Petri dish with water under natural light conditions. Pictures were analyzed with our own software (more information is available in a previously published study (Eroukhmanoff et al. 2009a). We measured pigmentation brightness (V) over the entire body (with values ranging from 0 [completely dark] to 1 [extremely lightly pigmented individuals]). For the frequency distribution of pigmentation brightness (V), a total of 805 individuals were measured (Fig. 1). We calculated the phenotypic variance from all individuals from both ecotypes of Lake Krankesjön for further use in the simulations described below.
To investigate if assortative mating was present and to quantify the degree of sexual isolation between lakes or ecotypes, we performed no-choice experiments (Jennions and Petrie 1997). We randomly paired one sexually active male and one sexually active female from two given populations and observed them in a Petri dish filled with water. From these trials, we were able to estimate the average propensity to form a precopula. We used the same threshold time as in a previous study (520 s, Eroukhmanoff et al. 2009a) to determine if individuals would have mated or not under natural conditions.
Couples were attributed to either the value 0 (did not mate) or 1 (mated). We conducted these mating experiments and tested all possible mating combinations between the two ecotypes from the two different lakes (four different crosses involving individuals of the same lake and ecotype (KR-KR, KS-KS TR-TR, TS-TS), two heterotypic crosses between lakes (TR-KS, KR-TS), two heterotypic crosses within lakes (KR-KS, TR-TS), and two homotypic crosses between lakes (TR-KR, TS-KS) (Abbreviations above: KR: Krankesjön Reed, KS: Krankesjön Stonewort, TR: Tåkern Reed, TS: Tåkern Stonewort). In total we performed a total of 589 such experimental mating trials (involving a total of 1178 individuals). These mating trials were distributed across 16 different pair combinations, and involved male and female ecotype and lake in the different categories.
MIGRATION MODIFICATION EXPERIMENTS
To investigate whether habitat isolation was present in this system, we conducted additional experiments. A total of 300 individuals from each ecotype from Lake Krankesjön were captured in the field and transported to our laboratory. Isopods were acclimated for a period of two days. They were fed on their original substrate sampled at the study sites during this period. Animals were thereafter randomly divided into 50 individuals in each replicate, and placed in an aquarium (30 cm × 70 cm) containing the substrate from their original habitat (stonewort shoots or decaying reed leaves) in one end, and the substrate of the other habitat on the other end, separated by a distance of 40 cm, which formed a “neutral” zone where no substrate of any kind was present. Isopods were then either placed in what we called “experimental habitat,” which could either be their own source habitat or a different habitat than from which they originated. After 24 h, we counted the number of isopods within each substrate, to estimate the proportion of individuals that moved between substrates.
It is possible that a longer duration of the experiment might have enabled some sort of behavioral accommodation to an unknown substrate through repeated samplings and successive dispersal events by the individuals, and that habitat fidelity would decline over time. However, this is unlikely to bear any strong significance in natural conditions, as both habitats are usually separated by at least 10 m of water and it is quite unlikely that isopods would migrate forth and back several times during their life under natural conditions, due to the fact that these small, short-lived and slow-moving animals are likely to suffer from high energetic expenses and high predation risk in the open water. We used three different replicates of each possible combination, for a total of 12 replicates. Because of logistical difficulties, we were not able to perform this migration modification experiment also in Lake Tåkern, and hence only results from Lake Krankesjön are reported here.
We used a fully factorial generalized linear model (GLZ) (Type III) to investigate which factors influenced mating probability. In this model, we assumed that mating probability (the dependent variable) followed a binomial distribution, and we included female's and male's ecotype and lake as fixed factors and all their possible interactions (both two- and three-way, as independent variables (Table 1). To correct for overdispersion, we rescaled the deviance parameter when it was needed. To assess to what extent lake-specific factors influenced the emergence of assortative mating, we categorized the trials as being either within lake/between lakes or within ecotype/between ecotypes, following the procedure of previous studies on ecological speciation (Rundle et al. 2000; Boughman et al. 2005). We estimated the average mating probability for each type of combination. Each of the four categories involved four different average mating probabilities from four different mating combinations. We then compared these four categories using a two-tailed t-test.
Table 1. Generalized linear model (GLZ) of how mating probability is affected by female and male ecotype and lake, as well as all their possible interactions.
Male lake female lake
Male lake × male ecotype
Female lake × male ecotype
Male lake × female ecotype
Female lake × female ecotype
Male ecotype × female ecotype
Male lake × female lake × male ecotype
Male lake × female lake × female ecotype
Male lake × male ecotype × female ecotype
Female lake × male ecotype × female ecotype
Male lake × female lake × male ecotype × female ecotype
In the migration experiments, we analyzed variation in the probability of migration using a fully factorial general linear model (GLM) including experimental habitat and original ecotype and their interaction (Table 2). In this analysis, we took into account the average migration probability of the 50 individuals per replicate for each treatment category. Thus, we used the average value per replicate, pooled across all individuals, to ensure statistical independence and avoid pseudoreplication.
Table 2. Generalized linear model (GLZ) of how migration probability is affected by ecotype and experimental ecotype as well as their interaction.
Ecotype × experimental ecotype
We partitioned the relative contribution of assortative mating and migration modification, where habitat isolation was quantified as 1 – (% of individuals which chose the foreign habitat in all trials) and which equals 0.5 when habitat choice is random and sexual isolation as 1 – (heterotypic mating frequencies/homotypic mating frequencies) which equals 0 when mate choice is random following a procedure described by Ramsey et al. (2003). We used a previously developed spreadsheet to calculate total isolation and the absolute contributions to the total isolation by any number of isolating barriers (Ramsey et al. 2003; available at http://www.plantbiology.msu.edu/schemske.shtml). Finally, we used the software JMATING (Antonio Carvajal-Rodríguez and Rolán-Alvarez 2006) to estimate Ipsi, the index of sexual isolation between ecotypes.
THE EFFECT OF MIGRATION MODIFICATION AND ASSORTATIVE MATING ON ADAPTIVE DIVERGENCE
As we have shown in a previous study (Eroukhmanoff et al. 2009b) phenotypic divergence between ecotypes is likely to be adaptive and the phenotypic changes showed evidence of high evolutionary rates, especially the pigmentation traits. The two different barriers to gene flow that we studied might have substantially enhanced divergence, but their relative contribution to total isolation needs to be quantified and the total extent to which they amplify phenotypic divergence when working in isolation, as well as jointly. To quantify their relative importance of these two barriers to phenotypic barriers, we used a previously developed theoretical framework for how migration–selection balance influences population divergence in quantitative traits (Hendry et al. 2001; Bolnick et al. 2009).
Hendry et al. (2001) showed that, when migration precedes selection within a generation, the equilibrium adaptive difference in a quantitative trait between populations in the process of speciation, or in our case ecotypes, can be quantified as a function of D, the optimal trait difference, m, the sum of the migration rates in each direction, G the genetic variance, P the phenotypic variance, and ω2 the variance of the fitness function (inversely related to the strength of stabilizing selection).The following equation describes the ratio of adaptive divergence in a phenotypic trait (as it is the case for pigmentation brightness (Eroukhmanoff et al. 2009b) with versus without a given isolating barrier
where D and DI are the optimal trait differences with and without one or several isolating barriers, VG and VP are the genetic and phenotypic variance for one quantitative trait (here, pigmentation brightness, a trait that is know to be under divergent selection [Eroukhmanoff et al. 2009b]), m is the cumulated random migration rate from one habitat to another in both directions and as defined in Hendry et al. (2001), ω2 is the variance of stabilizing selection experienced in each habitat and I is the strength of the isolating barrier. Thus, calculating this ratio between DI and D enables us to investigate the impact of isolation on adaptive divergence without knowing the optimal trait differences between populations and the exact nature of the fitness function for this trait. From this equation, we can derive a second one taking in account the two types of isolating barriers jointly:
where IMM and IAM are the proportion of individuals that stay in their native habitats (a measure of habitat fidelity implying migration modification, MM) or only mate with individuals from their own habitats (assortative mating, AM). In all of our simulations, the model parameters (additive genetic and phenotypic variances for the traits) were taken from our previously published studies on both ecotypes of Lake Krankesjön (Eroukhmanoff et al. 2009a, b). We successively used IMM and IAM alone in equation (1) to estimate their relative strengths when operating alone, as well as jointly (eq. 2). The goal of these simulations was to estimate total effect of reproductive isolation on adaptive phenotypic divergence. These two parameters were taken directly from the raw mating and migration data, as can be seen in Figs. 2A and 3B. Because these values tended to slightly differ between the Reed and Stonewort habitats, as did phenotypic and genetic variances (Eroukhmanoff et al. 2009a, b), we chose to present the results of our simulations for the reed and stonewort quantitative genetic parameters independently, instead of averaging all parameters between ecotypes. We chose a moderate variance (ω2= 5 times the phenotypic variance for pigmentation brightness) for the fitness function in each habitat. As a caveat, we note that this system is quite young and adaptive divergence might not yet have reached its equilibrium. Thus these simulations will therefore provide a conservative estimate of the extent to which adaptive divergence might be enhanced (Hendry et al. 2001; Bolnick et al. 2009). We chose to vary m, the random migration rate, to assess the role of the isolating barriers on divergence at low, moderate, or high dispersal rates. However, in our case, dispersal is likely to be quite low given both the physical and biological barriers between the two ecotypes (several meters of open water with no shelter, low food availability and predators, Eroukhmanoff and Svensson 2009).
We first conducted no-choice mating experiments that involved all possible mating combinations between the two ecotypes from the two lakes. We investigated how probability of mating was affected by a male and female ecotype and lake of origin, as well as all possible interactions between these factors (Table 1 and Fig. 2A). We found strong evidence for a significant female ecotype × male ecotype interaction (GLZ: χ2= 9.02, P < 0.01). This provides the first significant evidence for assortative mating between the different ecotypes in both lakes and reveals that mating was preferentially within, rather than between ecotypes. The coefficient of total sexual isolation Ipsi (−1 = disassortative mating; 0 = random mating, 1 = assortative mating, Carvajal-Rodríguez and Rolán-Alvarez 2006) was equal to 0.173 (SD = 0.067). After bootstrapping the data (n= 10,000 runs), it was found to be significantly different from 0 (P= 0.0092). This suggests moderate but significant assortative mating.
To further investigate the role of ecological diversification in assortative mating (Rundle et al. 2000; Boughman et al. 2005), we compared the average mating probability for all individuals crossed in each type of category (Fig. 2B). The three types of comparisons were performed using the average probability of copulation of each type of trial. We used criteria that have been outlined before in a previous study on ecological speciation (Rundle et al. 2000). First, we compared categories of individuals from similar ecotypes from the same or different lakes. They had the same probability of mating (t6= 0.352, P= 0.74). This indicates that there is no overall isolation between the two lakes. Isopods belonging to the same ecotype from different lakes also had a higher probability of mating than isopods belonging to different ecotypes from different lakes (t6= 2.84, P= 0.029). This shows that assortative mating operates with respect to ecotype across lakes, with little or no role of long-distance geographic isolation (Fig. 1). Isopods from different lakes will thus only discriminate against isopods from a different ecotype, but not against their own ecotype (Fig. 2B). Finally, we compared crosses involving individuals from different ecotypes, but from the same or different lakes. There was no significant difference between these two types of crosses in the probability of mating, which suggests that premating isolation has emerged in a similar fashion across lakes (t6= 1.53, P= 0.17).
Next, we investigated if there was any evidence for migration modification in the two ecotypes. We experimentally quantified the strength of habitat choice using isopods from Lake Krankesjön. We found that isopods from each of the two ecotypes clearly preferred their own habitats during all trials (Fig. 3). A fully factorial GLM identified a significant interaction between experimental habitat and source ecotype (F1,8= 64.00, P<0.001). There was no significant main effect of habitat (F1,8= 2.46, P= 0.15) and no intrinsic main ecotype effect in the tendency to disperse (F1,8= 1.00, P= 0.34). This demonstrates that when placed in their original source habitat, individuals do not migrate as often as when placed in a foreign experimental habitat (Table 2 and Fig. 3).
REPRODUCTIVE ISOLATION AND ADAPTIVE DIVERGENCE
Because both assortative mating and habitat choice seem to operate in this isopod system, we quantified their relative contribution to total isolation between ecotypes (Ramsey et al. 2003). Total isolation between ecotypes amounted to 0.78 (0 = no isolation, 1 = complete isolation). We found that the relative contribution of assortative mating to total isolation was relatively weak (11.9%) compared to habitat choice (88.1%). Our numerical simulations revealed that even with low random dispersal and moderate selective pressures, the joint emergence of both these barriers is likely to have a strong positive effect on adaptive divergence (Fig. 4). We explored the effects of different values of ω2 (from 1 (strong stabilizing selection within each habitat) to 15 times (weak selection) the phenotypic variance) to investigate the robustness of our findings and conclusions. Our results remained qualitatively the same across a wide range of parameter values, suggesting that our general conclusions are robust. Again, migration modification was likely to have a stronger effect on adaptive divergence than assortative mating. In combination, these two mechanisms appear to have enhanced divergence three- to fivefold compared to if they would have operated in isolation, especially if one considers the upper limits of each curve (Fig. 4).
Colonization of novel environments might lead to ecological speciation as a byproduct of adaptation to divergent selection (Rundel et al. 2000; Nosil and Crespi 2006a; Nosil et al. 2000). Although the importance of ecology in speciation is acknowledged by many, relatively little is known about the evolutionary rate by which reproductive isolation might evolve (Hendry et al. 2000, 2007). Some recent studies indicate that reproductive isolation can evolve over a few dozen generations (Hendry et al. 2007). In contrast, the more traditional view is that speciation might take hundreds of thousands of generations. In this isopod species, rapid adaptive divergence in pigmentation seems to have been facilitated by sorting of pre-existing variation in the ancestral ecotype (Fig. 1; Eroukhmanoff et al. 2009a; F. Eroukhmanoff and E. I. Svensson, unpubl. ms). The relatively rapid evolution that has taken place in this system might indicate that some degree of reproductive isolation might have reduced the constraining effects of gene flow between ecotypes. Here, we have shown that different ecotypes of an aquatic isopod mate assortatively, probably as an indirect consequence and correlated response of selection for local adaptation to different predation regimes. Assortative mating has emerged rapidly in this system, in as short time as 50 generations or less (Fig. 2A).
Some additional analyses suggest that it is local adaptation rather than geographic isolation that has indirectly resulted in premating isolation. The first two comparisons of mating propensity reveal that assortative mating is mostly based on ecotype, and not on lake of origin (Fig. 2B). Moreover, comparison III (Fig. 2B) also suggests a limited role for historical contingency on the emergence of sexual isolation (Rundle et al. 2000; Langerhans and De Witt 2004; Langerhans et al. 2006). These findings suggest parallel emergence of premating isolation in the two lakes and it is consistent with our previous findings of strong parallel divergence in morphological and behavioral traits in this system (Eroukhmanoff et al. 2009a; Eroukhmanoff and Svensson 2009). Moreover, in a previous study (Eroukhmanoff et al. 2009a) a haplotype network (mtDNA) revealed that the most likely scenario for ecotype divergence is in situ independent emergence of the stonewort ecotype in each of the lakes, which is also likely to apply to the independent emergence of assortative mating in both lakes.
Our experiments from Lake Krankesjön strongly suggest that migration modification contributes to maintain reproductive isolation of the two ecotypes (Fig. 3). Habitat fidelity is strong and due to the design we used, density-dependence is unlikely to have played a strong role. Indeed, in the beginning of each trial individuals were all placed on one substrate, thus automatically favoring migration to the other substrate present if migration was density dependent. However, no such effect was detected (Table 2). The results in this study suggest that migration modification has a stronger role than assortative mating in restraining gene flow between the two ecotypes (Fig. 3), which is also an inherent property of philopatry as it intervenes as an early reproductive barrier.
Our simulations showed that for pigmentation brightness phenotypic divergence is maximized by the joint emergence of both assortative mating and migration modification (Fig. 4). Migration modification is an alternative isolating mechanism that might counteract gene flow, before assortative mating has emerged (Yukilevich and True 2006). Ecologically divergent selection can of course be solely responsible for the evolution of assortative mating and migration modification, but has so far mainly be studied in the context of reinforcement both theoretically (Yukilevich and True 2006) and empirically (Nosil and Yukilevich 2008). The intensity of indirect selection against migrants or hybrids is likely to play a role too in the adaptive divergence process described here (Nosil and Yukilevich 2008). Unfortunately, we do not have any data on hybrid fitness to assess the role of hybridization in this system.
The findings in this study are largely consistent with a scenario of ongoing ecological speciation strengthened through migration modification, although complete reproductive isolation has not yet been achieved. The fact that this process has been relatively fast, taking only a few dozen generations (Hargeby et al. 2004), suggests that under contrasting ecological conditions and under sufficiently strong divergent selection, local adaptation might be accompanied by reproductive isolation, even in the early stages of ecological speciation. The total level of premating reproductive isolation (including both sexual and habitat isolation) is very high in this system (0.78, although the index of sexual isolation is relatively weak, 0.173), especially, when keeping in mind that the phenotypic divergence between ecotypes is relatively recent. A survey of the pattern of isolation between species of the genus Drosophila (Coyne and Orr 1997) revealed that most premating isolation indices estimated between pairs of sympatric species were close to 0.8. Hence, the system we have studied here has comparable levels of premating isolation as several sympatric species-pairs of Drosophila, which have been diverging for several hundreds of thousands of years, if one takes in account the effects of migration modification (Coyne and Orr 1997).
In conclusion, we have demonstrated that both assortative mating and habitat choice operate in this isopod system. Our experimental results and simulations suggest that both these isolating barriers are likely to efficiently and jointly restrain gene flow between the ecologically divergent populations. These barriers to gene flow are likely to be especially important in the early stages of divergence and speciation. In this isopod system, migration modification turned out to be more important in contributing to total isolation, and seems to play a more pronounced role in promoting adaptive divergence, at least in Lake Krankesjön. Our study adds to the increasing evidence that assortative mating can emerge extremely rapidly (Hendry et al. 2000) but our results also suggest an additional and important role for migration modification.
Associate Editor: C. C. Nice
We thank S. Guéchot, N. Nowshiravani-Arnberg, and K. Karlsson for their help with field-work and P. Edelaar, A. Hendry, A. Qvarnström, J. Kotiaho, K. Rengefors and “The Svensson Lab” for their comments on earlier drafts of the manuscript. This study was financially supported by the Swedish Research Council to FE and ES.