Correspondence: Jun Kitano, Ecological Genetics Laboratory, Center for Frontier Research, National Institute of Genetics, Mishima, Shizuoka 411-8540, Japan. Tel.: +81 55 981 9415; fax: +81 55 981 9416;
Although similar patterns of phenotypic diversification are often observed in phylogenetically independent lineages, differences in the magnitude and direction of phenotypic divergence have been also observed among independent lineages, even when exposed to the same ecological gradients. The stickleback family is a good model with which to explore the ecological and genetic basis of parallel and nonparallel patterns of phenotypic evolution, because there are a variety of populations and species that are locally adapted to divergent environments. Although the patterns of phenotypic divergence as well as the genetic and ecological mechanisms have been well characterized in threespine sticklebacks, Gasterosteus aculeatus, we know little about the patterns of phenotypic diversification in other stickleback lineages. In eastern Hokkaido, Japan, there are three species of ninespine sticklebacks, Pungitius tymensis and the freshwater type and the brackish-water type of the P. pungitius–P. sinensis species complex. They utilize divergent habitats along coast–stream gradients of rivers. Here, we investigated genetic, ecological and phenotypic divergence among three species of Japanese ninespine sticklebacks. Divergence in trophic morphology and salinity tolerance occurred in the direction predicted by the patterns observed in threespine sticklebacks. However, the patterns of divergence in armour plate were different from those previously found in threespine sticklebacks. Furthermore, the genetic basis of plate variation may differ from that in threespine sticklebacks. Because threespine sticklebacks are well-established model for evolutionary research, the sympatric trio of ninespine sticklebacks will be an invaluable resource for ecological and genetic studies on both common and lineage-specific patterns of phenotypic diversification.
Both parallel and nonparallel patterns of phenotypic diversification can be found between different lineages exposed to the same environmental gradients. Parallel patterns have been observed in a number of taxa, including lizards (Losos, 2009), fishes (Taylor, 1999), snails (Chiba, 2004), crustaceans (McPeek & Wellborn, 1998) and insects (Nosil et al., 2002; Stoks et al., 2005). Parallel evolution of similar traits may result from the evolutionary response of organisms to the same selective pressures (Schluter, 2000; Schluter et al., 2004). North template fishes in post-glacial lakes, such as sticklebacks, whitefishes, charr and smelt, often diverge in similar ways along the benthic–limnetic axes (Bernatchez et al., 1996; Taylor, 1999; Schluter, 2000). One of the diverged species usually exploits pelagic zooplankton, whereas the other exploits benthic invertebrates or large prey in the littoral or deeper zones (Bernatchez et al., 1996; Taylor, 1999; Schluter, 2000). In such cases, patterns of morphological divergence are similar across diverse taxa: planktivores usually have more numerous gill rakers (foraging apparatus in fishes) than benthivores (Schluter, 2000). The ecological gradients from the lower to the upper reaches of coastal rivers, involving many abiotic and biotic shifts, also provide a stage for repeated phenotypic diversification of multiple organisms (Fuller et al., 2007; McCairns & Bernatchez, 2008; Whitehead et al., 2011). Closely related species often exploit the upper and lower reaches of a river, respectively, showing phenotypic divergence in salinity tolerance, trophic ecology and armour morphology along coast–stream gradients (Hagen, 1967; Fuller et al., 2007; McCairns & Bernatchez, 2008; Schluter & Conte, 2009; Whitehead et al., 2011). Thus, investigation of parallel phenotypic evolution has been central to the studies on the roles of divergent natural selection in phenotypic diversification.
However, even when independent lineages are exposed to the same environmental gradients, the magnitude and direction of phenotypic divergence often differ between lineages. Nonparallel aspects have been found in diverse taxa (Price et al., 2000; Losos, 2009; Berner et al., 2010; Landry & Bernatchez, 2010; Romero, 2011; Rosenblum & Harmon, 2011; Kaeuffer et al., 2012). Kaeuffer et al. (2012) suggested that four potential factors might cause the differences. First, different species might use different combinations of traits to achieve the same performance. Second, simple discrete classification of ecotypes might overlook other important environmental variables. Third, sexual selection might influence evolutionary trajectories. Finally, variation in genetic architecture might bias the evolutionary response. Recently, the genetic architecture or even genes underlying adaptive traits have been increasingly elucidated. The parallel evolution of similar adaptive traits may be caused by the repeated fixation of the same alleles, different mutations at the same genes or mutations at different genes (Arendt & Reznick, 2007; Hoekstra & Coyne, 2007; Nadeau & Jiggins, 2010; Wake et al., 2011). In cases where the same alleles are used, populations may respond in similar ways to the same selective pressures, resulting in the parallel patterns of phenotypic diversification (Gompel & Prud'homme, 2009; Stern & Orgogozo, 2009; Conte et al., 2012). In contrast, when different genes are used, different patterns may appear, because different mutations will not always have the same phenotypic effects. Therefore, in addition to the investigation of the parallel and nonparallel aspects of phenotypic diversification, it is crucial to investigate whether the same alleles or the same genes are used when species of independent lineages respond to the same selective pressures.
The stickleback family is a good model with which to explore the ecological and genetic basis of repeated phenotypic evolution (Cresko et al., 2006; Kingsley & Peichel, 2007; Schluter & Conte, 2009; Jones et al., 2012b). For example, the repeated reduction in armour plates, pelvic spines, gill pigmentation and thyroid hormone levels has been observed in threespine stickleback (Gasterosteus aculeatus) populations that repeatedly colonized freshwater environments during the post-glacial dispersal (Wootton, 1976; Bell & Foster, 1994; Miller et al., 2007; Kitano et al., 2010). Repeated reduction in armour plates in freshwater populations is caused by the repeated fixation of the same allele of the Eda gene, which is a major gene controlling plate development (Colosimo et al., 2005; Schluter & Conte, 2009). In contrast, reduction in pelvic spine is caused by independent mutations at the same gene, Pitx1 (Shapiro et al., 2004; Chan et al., 2010). The majority of previous studies on the stickleback family have been confined to intraspecific variation within G. aculeatus (Wootton, 1976; Bell & Foster, 1994; McKinnon & Rundle, 2002). Because we already have extensive knowledge about the genetic and ecological mechanisms underlying phenotypic divergence and reproductive isolation within threespine sticklebacks, the study of other stickleback lineages would provide great opportunities to investigate the same and different patterns of phenotypic diversification between phylogenetically independent lineages.
The present study focused on three sympatric species of ninespine stickleback (genus Pungitius) in eastern Hokkaido, Japan (Fig. 1). One species is P. tymensis, which is endemic to Hokkaido Island and Sakhalin Island (Wootton, 1976; Takahashi & Goto, 2003). Pungitius tymensis might have diverged from P. pungitius around 5.7 million years ago, because of some geological events during the glacial cycles (Aldenhoven et al., 2010). The other two species belong to the P. pungitius–P. sinensis species complex. Pungitius pungitius has an almost continuous circumpolar distribution in coastal regions of the Northern Hemisphere (Wootton, 1976). The taxonomy of P. sinensis has been controversial. Pungitius sinensis was once distinguished from P. pungitius by its complete row of lateral armour plates and was thought to be distributed around the Sea of Japan and the Sea of Okhotsk and in some coastal areas in China (Wootton, 1976). However, recent studies have revealed that these two species cannot be distinguished by observation of plate morph, because plate morph variation does not correspond to allozyme variation (Takata et al., 1987). Therefore, the P. pungitius–P. sinensis species complex is now classified into the ‘freshwater type’ and the ‘brackish-water type’ based on habitat use and the allozyme differences (Takata et al., 1987; Takahashi & Goto, 2003). Because crosses in both directions between the freshwater type and the brackish-water type exhibit hybrid male sterility (Takahashi et al., 2005), these two cryptic species might have diverged a very long time ago during the glacial cycles (Takahashi & Goto, 2003). Although a previous study showed that these three species might have different spawning sites along a single stream (Tsuruta et al., 2008), the genetic, morphological, trophic and physiological differentiations among the sympatric species have not yet been investigated.
This study aimed to elucidate the habitat isolation and phenotypic divergence among the three Pungitius species and to compare patterns of divergence with those found in the sympatric pairs of Gasterosteus lineage and other fishes. First, by using microsatellite markers, we conducted genetic analysis of the ninespine sticklebacks caught in eastern Hokkaido, where these three species are often sympatric or parapatric. Although previous allozyme data demonstrated that these three species are genetically differentiated (Takata et al., 1987), mitochondrial DNA data indicated the presence of gene flow (Takahashi et al., 2001). Therefore, it is important to know the frequencies of hybrids in sympatric and parapatric sites. Furthermore, because they are morphologically similar and difficult to distinguish on the basis of external morphology alone (Data S1), genetic identification is essential for the subsequent investigation of ecological and phenotypic divergence among these three cryptic species. Second, we investigated morphological, ecological and physiological divergence among them. Because these three species were collected at different sites along coast–stream gradients (see below), we predicted that the differences in trophic morphology, armour morphology and salinity tolerance might occur in the direction predicted by the patterns found in threespine sticklebacks: fishes in the lower reaches would be expected to be planktivores, more armoured and more tolerant of higher salinity than the fishes in the upper reaches. Differential adaptations to divergent habitat use might also contribute to ecological speciation, because ecological selection against migrants and hybrids with intermediate phenotypes can contribute to reproductive isolation between ecotypes (Nosil et al., 2005; Nosil, 2012). Finally, we investigated whether variation in armour plate morphology within the ninespine stickleback species complex is controlled by the Eda gene, which controls the armour plate variation in threespine sticklebacks (Colosimo et al., 2005) in order to examine whether the same allele is used for plate morph variation in the phylogenetically independent lineages.
Materials and methods
For genetic, morphological and physiological analyses, three species of ninespine stickleback were collected using minnow traps and hand nets from the Oboro, Biwase and Shiomi River Systems of eastern Hokkaido, Japan, in May 2011 (Fig. 1b). Because we detected divergence in trophic morphology (see below), we conducted additional sampling in December 2011 from an estuarine pond, a midstream site and an upstream site of the Shiomi River System (Fig. 2c) to analyse stomach contents. For genetic analysis of the Eda locus, we conducted additional sampling in May 2012 from the estuary of the Biwase River and the upstream of the Oboro River and November 2012 from the downstream of the Oboro River, because the population sizes of these populations seem relatively larger and they showed variations in the plate number. Details of each sampling site are available in the Table 1. Because it is very difficult to distinguish among these three species by visual observation (Data S1), all of the fishes were first analysed blindly and genotyped later to identify the species (see below).
Genomic DNA was isolated from the right pectoral fins of ethanol-fixed specimens with the Qiagen DNeasy Blood and Tissue Kit (Qiagen, Valencia, CA, USA). For microsatellite analysis, we first tested for 12 microsatellite markers that are located on 12 different linkage groups (Table S1) (Shapiro et al., 2009). The forward primers were labelled with either HEX, NED or FAM. The 5′-end of the reverse primers was tailed with GTTTCTT to increase the accuracy of the fragment analysis (Ballard et al., 2002), as described previously (Adachi et al., 2012). Three differently coloured primer sets were combined and the microsatellite loci were amplified from the genomic DNA by using a KAPA2G Fast Multiplex PCR Kit (KAPA Biosystems, Woburn, MA, USA). The amplified fragments were analysed in the genotyping centre of BEX Co. Ltd (Tokyo, Japan). Two of the primer sets (Pun238 and Pun255) yielded no PCR products in any species; therefore, these markers were excluded from further analysis. We then tested whether the other markers were in Hardy–Weinberg equilibrium by using an exact test with Genepop (Raymond & Rousset, 1995a,b). Two markers deviated from Hardy–Weinberg equilibrium (Pun44 in the freshwater type collected from the downstream of the Oboro River System and in the brackish-water type collected from the Shiomi River System and Pun261 in the brackish-water type collected from the Shiomi River System). Analysis using the Micro-Checker program (van Oosterhout et al., 2004) revealed that these markers might have null alleles. Therefore, these two markers were excluded from further analysis and the remaining eight markers (Pun212, Pun134, Pun19, Pun68, Pun117, Stn433, Pun171 and Pun78) were used for further population genetic analyses (Table S1).
The data were first analysed using the STRUCTURE software (Pritchard et al., 2000) with Markov chain Monte Carlo simulations to identify groupings that minimize Hardy–Weinberg and linkage disequilibrium within cluster groups, as described previously (Kitano et al., 2008, 2009; Adachi et al., 2012). We did not include the information on the sampling sites. The cluster number (K) value with the highest log-likelihood L(K) and the highest ad hoc statistics ∆K, which is based on the rate of change in L(K) between successive K values (Evanno et al., 2005), was calculated. Parameters were estimated after 200 000 iterations with burn-in of 25 000 iterations. Simulations were run 20 times for clusters from K =1 through K =10. Because the STRUCTURE analysis within each river also gave qualitatively the same results, we only present the data obtained from all populations together. Next, we conducted principal coordinate (PCo) analysis, as described previously (Kitano et al., 2008). Briefly, the pairwise genetic distance between individual fish was first calculated based on the proportion of shared alleles (Jin & Chakraborty, 1993) with the use of Populations 1.2.31 (http://bioinformatics.org/~tryphon/populations/#ancre_bibliographie). Then, the distance matrix was subjected to principal coordinate analysis by using the R statistical package (R Development Core Team, 2011).
We identified two hybrids among 149 fish with the STRUCTURE analysis (Fig. 2a); these two individuals were excluded from the investigation of genetic differentiation between populations. Pairwise FST between populations were calculated using the Arlequin software version 3.5 (Excoffier et al., 2005). The Jost's estimator of actual differentiation Dest (Jost, 2008) and Nei's genetic distance (Nei's D) (Nei & Kumar, 2000) were calculated using the SMOGD software (Crawford, 2010) and the Populations 1.2.31 software, respectively. Jost's Dest is considered as a better estimator for highly polymorphic markers, such as microsatellite, than FST and Nei's D (Jost, 2008). Genetic variance explained by geographical location and species was calculated using the analysis of molecular variance (amova) of the allele frequencies with the Arlequin (Excoffier et al., 2005). Gene diversity and allelic richness were calculated using the FSTAT software (Goudet, 1995). The gene diversity and allelic richness were compared between populations using the Mann–Whitney U-test.
For genotyping at the Eda locus, a major locus responsible for the repeated reduction in the armour plate in the threespine stickleback G. aculeatus (Colosimo et al., 2005), three microsatellite markers (Stn364, Stn380 and Stn381) at the Eda locus on the linkage group 4 were used for genotyping the ninespine sticklebacks. Because Stn381 amplified no fragments in any ninespine stickleback species, only the Stn364 and Stn380 markers, which are both located in introns of the Eda gene (Colosimo et al., 2005), were used for further analysis. In threespine sticklebacks, these regions are significantly associated with plate morph not only in the laboratory crosses between differently armoured morphs (Colosimo et al., 2004, 2005), but also in natural populations (Colosimo et al., 2005; Cano et al., 2006; Kitano et al., 2008). In addition, signatures of natural selection were repeatedly observed at this chromosomal region in natural populations of threespine stickleback (Cano et al., 2006; Raeymaekers et al., 2007; Hohenlohe et al., 2010; Jones et al., 2012a,b).
Stomach content analysis
To investigate trophic divergence among the three species of ninespine stickleback, stomach contents of the fishes collected at different sites along the Shiomi River were analysed (Fig. 2c). Prey items were counted and weighed to the nearest 0.001 mg. Prey items were sorted under a dissecting microscope into 10 categories: terrestrial insects, Mysidacea, Copepoda, Amphipoda, Isopoda, Chironomus larvae, Plecoptera larvae, other benthos, plants and the rest (including unidentified items).
To characterize food items of the three species, the percentage index of relative importance (%IRI) was calculated from total weight and occurrence rate as described previously (Pinkas et al., 1971; Kume et al., 2010).
where %Fi is the frequency of occurrence of the item i (%), %Wi is the relative weight of the item i (%), and %Ni is the relative number of items i (%).
For observation of lateral plates and gill rakers, the samples were stained with alizarin red to visualize the bony structures, as described previously (Kitano et al., 2007, 2008). The number of gill rakers on the first right gill arch and the number of left lateral plates were counted under a dissecting microscope. Photographs of the first gill arch were taken with a CCD camera connected to the dissecting microscope, and gill raker length (GRL) was measured from the longest gill raker on the first arch by using the ImageJ software 1.38 (Abramoff et al., 2004).
For phenotyping the plate morph, the plate number was counted from the left side of the alizarin red–stained fishes (Fig. 5a). All fish analysed were larger than 32 mm in standard length, by which all plates ossify in threespine sticklebacks (Bell et al., 2004). In addition, the fishes were classified into completely plated morphs with a continuous row of lateral plate along the body side and partially plated morphs with a gap in the plate row (Fig. 5a). Fish with a continuous row with one gap were classified into completely plated morphs, as previously conducted in threespine sticklebacks (Hagen & Gilbertson, 1972). All the ninespine sticklebacks had caudal plates (i.e. keels), so none of the fishes was classified into low-plated morphs.
Salinity tolerance analysis
In order to investigate seawater tolerance, we measured the plasma Na concentration in fishes after transfer to seawater. Fishes with adequate osmoregulatory abilities are able to maintain plasma Na concentrations at relatively low levels even after transfer to seawater, whereas fishes vulnerable to seawater will have substantially increased plasma Na concentrations when exposed to seawater (Foote et al., 1992). Seawater was made with Instant Ocean (Instant Ocean, Aquarium Systems, Mentor, OH, USA). The fishes were first acclimated to 10% seawater for 1 month before the experiments and then transferred to either freshwater, 10% seawater (control) or 100% seawater. The fishes were sampled 24 h after the transfer except for P. tymensis exposed to seawater, which were sampled after 3 h because most of the P. tymensis fish started to die within 3–6 h of placement in 100% seawater. After euthanasia with tricaine methanesulfonate, blood was collected from the severed caudal vein with heparinized capillary tubes and centrifuged at 3000 x g for 10 min, as described previously (Kitano et al., 2010, 2011). Then, the plasma was frozen at −80 °C until use. Just prior to measurement of plasma Na concentrations, the plasma was thawed on ice and 1 μl was used for the assay. Na concentrations were determined using an atomic absorption spectrophotometer (Z-5300 Hitachi, Tokyo, Japan), which calculates the Na concentrations in the sample from the light intensity at a particular wavelength emitted from a flame. To evaluate the significant differences in plasma Na concentrations among three species, analysis of variance (anova) was performed, followed by Tukey–Kramer post hoc tests.
Reproductive isolation along coast–stream gradients
STRUCTURE analysis revealed three genetically distinct clusters in eastern Hokkaido with both Evanno's ∆K and L(K) indicating the most likely cluster number as three (K =3) (Fig. 2a). Pungitius tymensis and the freshwater type were sympatric in the upstream site of the Oboro River System, whereas all three forms were sympatric at the midstream site of the Shiomi River (see below and Fig. 2c). In other sampling sites, three species were parapatric. Among 149 sampled fish, we found two hybrids: one hybrid between the freshwater type and the brackish-water type in the Oboro River and another hybrid between the freshwater type and the brackish-water type in the Shiomi River (Fig. 2a,b). Although there was significant genetic differentiation among the sampling sites within the species, amova revealed that only 3.7% of the total genetic variance was explained by geographical location (Table S3). In contrast, 43.3% of the total genetic variance was explained by interspecies variation (Table S3). The FST values, Jost's Dest and Nei's D also supported the results of STRUCTURE and amova. Genetic differentiation among the three species within the same watersheds was substantial, whereas the genetic differentiation among populations within the species was low (Tables 2 and S4).
Table 2. Pairwise Dest (upper) and FST (lower) between populations. FST values significantly larger than 0 (P <0.05) are indicated by asterisks
Pungitius tymensis (OboU)
Pungitius tymensis (Shi)
Pungitius tymensis (OboU)
Pungitius tymensis (Shi)
To further investigate the habitat isolation among the three species, we investigated the proportions of each species for fishes collected at three sites along the Shiomi River (Fig. 2c). All specimens at the upstream site were P. tymensis, whereas all specimens at the estuarine pond were the brackish-water type. At the midstream site, all three species, including the freshwater type, were caught. Thus, even within this small watershed, there was substantial habitat isolation among the three species.
Pungitius tymensis exhibited the lowest gene diversity and the lowest allelic richness (Tables S5 and S6). In both the Oboro River and the Shiomi River, the gene diversity of P. tymensis was significantly lower than that of the freshwater type (Oboro River, U =11.5, Z = P =0.031; Shiomi River, U =7.00, Z =2.69, P =0.007) and the brackish-water type (Shiomi River, U =0, Z =3.40, P <0.001). Allelic richness was also lower in P. tymensis than in the brackish-water type in the Shiomi River (U =7.0, Z =2.58, P =0.010).
Trophic and eco-physiological divergence along the ecological gradients
The ecological transition from the lower to the upper reaches of a coastal river involves differences in food items and salinity with more plankton and higher salinity in the lower reaches than in the upper reaches. First, to investigate the dietary differences, the stomach contents of the three species were compared (Fig. 3a, Table S7). The brackish-water type, which inhabited the estuarine pond, mainly preyed on zooplankton such as Mysidacea. In contrast, P. tymensis, which inhabited the upstream site, predominantly preyed on benthos such as the larvae of Chironomus and Plecoptera, terrestrial insects and zooplankton (e.g. Copepoda). At the midstream site, where the three species occurred sympatrically, benthos, such as Isopoda, were preyed on.
Planktivores generally have longer and greater number of gill rakers than benthivores (McPhail, 1994). Pungitius tymensis had the shortest and the lowest number of gill rakers (mean ± SE of gill raker number = 8.56 ± 0.17) (Fig. 3b,c; Table S2). In contrast, the brackish-water type had the longest and the largest number of gill rakers (mean ± SE of gill raker number = 12.98 ± 0.12; Table S2). The gill raker traits of the freshwater type were intermediate between those of the brackish-water type and P. tymensis (mean ± SE of gill raker number = 11.41 ± 0.10; Table S2) (anova of gill raker number, effect of species, F2, 208 = 283.68, P <0.01; Tukey–Kramer post hoc test, P <0.01 for all pairs). Thus, gill raker morphology diverged in the direction predicted by the patterns found for the post-glacial teleost species pairs.
Divergence in salinity tolerance also occurred in the expected direction (Fig. 4). When the fishes were exposed to 100% seawater, the plasma Na levels increased to more than 250 mmol L−1 in P. tymensis, with all P. tymensis starting to die within 3–6 h of exposure (Fig. 4). The plasma Na levels of P. tymensis exposed to 100% seawater were significantly higher than those observed in the brackish-water type and the freshwater type exposed to 100% seawater (anova, F2,9 = 283.77, P <0.01; Tukey–Kramer post hoc test, P <0.01 for all pairs). In the seawater challenge experiment, the plasma Na levels of the freshwater type increased more than those of the brackish-water type (anova, Tukey–Kramer post hoc test, P <0.01), although none of the freshwater type died even after the 24 h of exposure to 100% seawater. In the other river systems, similar divergence in the seawater tolerance was observed among the three species (Fig. 4).
Nonparallel patterns of armour plate variation
Armour plate number is another important ecological trait of G. aculeatus (Bell & Foster, 1994). Sticklebacks in the lower reaches are generally more plated than those inhabiting the upper reaches (Hagen, 1967; Bell & Foster, 1994; Colosimo et al., 2005). All populations of P. tymensis and the brackish-water type were partially plated, although P. tymensis had fewer lateral plates (range: 5–12) than the brackish-water type (range: 15–26). In contrast, we found both partially plated and completely plated morphs in the freshwater type. There was substantial variation in the frequency of plate morphs among populations within the freshwater type (range of plate number: 6–35) (Fig. 5b). Most of the freshwater type in the Shiomi River System (Shi) were partially plated, whereas most of the freshwater type at the upstream site of the Oboro River System (OboU) were completely plated morphs. Thus, the patterns of plate morph variation in the freshwater type could not be simply explained by the coast–stream gradients.
Because there is variation in the plate morph among populations within the freshwater type, we genotyped fishes of two polymorphic populations of the freshwater type (OboU and OboD) at the Eda locus. The allele distribution of Stn380 at the Eda locus was virtually identical between partially plated and completely plated morphs of the freshwater type (Fig. 5c) (Fisher's exact test: P =0.464 for OboD and P =0.853 for OboU). Even when these two populations were pooled, we found no association between the genotype at Stn380 and the plate morph (Fisher's exact test: P =0.463). Another microsatellite marker linked to Eda (Stn364) also did not show any significant association with the plate morph (Fisher's exact test: P =0.967 for OboD and P =0.073 for OboU). In brackish-water type, bimodal peaks in plate number were observed with one peak at 18 plates and another at 21 plates (Fig. 5b). Therefore, in the brackish-water type, we investigated differences in allele distribution at Stn364 and Stn380 between the fish with 18 plates and the fish with 21 plates. Allele frequencies were not significantly different between differently plated fish of the BiwaT population (Fisher's exact test: P =1.000 for Stn364 and P =0.495 for Stn380) and the Shi population of the freshwater type (Fisher's exact test: P =0.497 for Stn364 and P =0.566 for Stn380). Thus, the Eda genotype is not associated with plate morph variation in ninespine sticklebacks.
The patterns of phenotypic diversification in ninespine sticklebacks have both parallel and nonparallel aspects compared with those in threespine sticklebacks. Parallel patterns were found in the divergence of foraging morphology and salinity tolerance (Fig. 6). Diversification of foraging morphology is one of the most widely observed features in sympatric pairs of fishes, including threespine sticklebacks, whitefishes, charr and smelt, and may contribute to ecological speciation (Bernatchez et al., 1996; Taylor, 1999; Schluter, 2000). As predicted by previous studies on these other fish species, the brackish-water type exploiting zooplankton has more gill rakers than the other two species exploiting benthic invertebrates. Divergence in salinity tolerance has been also observed for several aquatic organisms along the coast–stream gradients (Fuller et al., 2007; McCairns & Bernatchez, 2008; Whitehead et al., 2011). If species inhabiting the upper reaches (i.e. streams) cannot survive in high salinity in the estuary, migrants from the upper reaches to the lower reaches are most likely to be selected against, thus preventing the hybridization with the species inhabiting the lower reaches. Pungitius tymensis cannot survive in seawater, and thus, it would be selected against if it migrated to the habitats of the brackish-water type. Our genetic studies confirmed that no hybrids between the brackish-water type and P. tymensis were found. Thus, divergent adaptation might at least partially contribute to the maintenance of reproductive isolation among these three species distributed along the coast–stream gradients.
In contrast, the patterns of armour plate evolution in ninespine sticklebacks differed from those in threespine sticklebacks. In threespine sticklebacks, fishes in the lower reaches are generally more armoured than fishes in the upper reaches (Hagen, 1967; Bell & Foster, 1994; Colosimo et al., 2005). Similarly, the brackish-water type of ninespine stickleback, inhabiting the lower reaches, has more plates than the P. tymensis, inhabiting the upper reaches. However, the plate phenotype of the freshwater type was so variable that the freshwater type often has more plates than the brackish-water type (Fig. 6). In addition, plate morph variation in ninespine sticklebacks may have a different genetic basis from threespine sticklebacks. The probability of sharing the same genes for parallel evolution appears to be negatively correlated with the genetic distance between lineages. In other words, genetically distant lineages use different genes for parallel phenotypic alterations than closely related lineages (Conte et al., 2012). Consistent with this hypothesis, the same allele of the same gene was used within threespine sticklebacks, whereas distantly related ninespine sticklebacks, which diverged from the threespine sticklebacks around 8.5 million years ago (Aldenhoven et al., 2010), likely used different genes for plate evolution.
The nonparallel patterns of armour plate evolution may be caused by several factors. Kaeuffer et al. (2012) suggested that four potential factors might cause the nonparallel patterns between different lineages: functional, ecological, sexual and genetic factors. First, threespine and ninespine sticklebacks may use different combinations of traits for antipredation. Although the lateral plates of threespine sticklebacks can prevent body puncturing when attached by toothed predators and increase the post-capture survival rate (Reimchen, 2000), the lateral plates of ninespine sticklebacks are much shorter in height than those of threespine sticklebacks and do not cover the entire body even in the completely plated freshwater type (Fig. 5a), so it is doubtful that they can protect the body trunk from puncture. Antipredation involves multiple traits in sticklebacks, including armour morphology (Reimchen, 1994), crypsis (Greenwood et al., 2011), freezing behaviour (Huntingford et al., 1994) and escape response (Taylor & McPhail, 1986; Bergstrom, 2002). Therefore, it is not surprising that threespine and ninespine sticklebacks use different combinations of traits for antipredation. Second, we might overlook some important environmental variables. The Eda gene may have pleiotropic functions, and genes important for osmoregulation, parasite resistance and other morphological traits are linked to the Eda gene (Colosimo et al., 2005; Barrett et al., 2009; Le Rouzic et al., 2011), suggesting that predation is not the only selective pressure shaping the plate morph variation in threespine sticklebacks. In addition, environmental calcium levels might constrain the plate evolution (Giles, 1983). Furthermore, interesting correlations were found between the alignment of mechanoreceptive lateral line and the plating patterns in threespine sticklebacks (Wark & Peichel, 2010), suggesting that selection on the lateral line system may also influence the armour plate phenotypes. Therefore, the patterns of plate morph variation in threespine sticklebacks come from multifarious selective pressures, multiple constraints and hitchhiking effects. The future studies need to quantify the predation pressures and other environmental variables potentially associated with plate phenotype of ninespine sticklebacks. Although sexual selection is the third potential contributing factor, we think that sexual selection plays little roles in producing the plate morph variation in ninespine sticklebacks. No previous studies have demonstrated the presence of sexual selection on stickleback plate morph. In addition, we found no sexual dimorphism in plate number in any ninespine sticklebacks analysed here (A. Ishikawa and J. Kitano, unpublished data). Fourth, genetic factors might contribute to the different patterns. Different mutations on different genes might have different phenotypic effects. Furthermore, hitchhiking effects of different neighbouring genes will also produce different patterns of phenotypic divergence. Further quantitative trait locus (QTL) mapping of plate morphology, salinity tolerance and other functional traits is possible by using genomic tools available for sticklebacks (Peichel et al., 2001; Colosimo et al., 2004, 2005; Shapiro et al., 2009) and is required to directly compare the genetic architecture of multiple adaptive traits between threespine and ninespine sticklebacks.
Finally, cryptic phenotypic differences between species pose particular challenges for conservation (Bickford et al., 2007). Our microsatellite and morphological data revealed that genetically differentiated freshwater and brackish-water types could not be easily distinguished on the basis of external morphology alone. Thus, further genetic studies on ninespine stickleback populations are needed to assess the population size of each population of each species. Several stickleback species or populations are now endangered because of human disturbances (Mori, 1997, 2003; Foster et al., 2003; Patankar et al., 2006; Taylor et al., 2006; Behm et al., 2010). One ninespine stickleback species, P. kaibarae, has already become extinct in Japan because of urban development (Mori, 1997). Our present genetic study revealed that P. tymensis populations have lower genetic diversity than other species, suggesting that particular attention should be given to the conservation of P. tymensis. The present study is only a snapshot of sympatric habitats. Because environmental disturbances can erode the isolating barriers among sympatric species and lead to the hybridization (Taylor et al., 2006; Seehausen et al., 2008; Vonlanthen et al., 2012), long-term genetic studies are required for maintenance of cryptic species. Our study clearly demonstrates that the preservation of ecologically diverse environments in coastal lakes and rivers is essential for conserving cryptic species, because existing ecological gradients can maintain diverse species by providing diverse habitats.
This research is supported by JST PRESTO program and Grant-in-Aid for Scientific Research on Innovative Areas (23113007 and 23113001) from the Ministry of Education, Science, Sports and Culture, the Cooperative Program of the University of Tokyo AORI (No. 103, 2011) to JK, NIG Cooperative Research Program (2011-A69 and 2012-A63) to SM, NIG Cooperative Research Program (2011-A86 and 2012-A61) to M. Kusakabe and Akkeshi Town Grants-in-Aid for Scientific Research in the Lake Akkeshi/Bekanbeushi Wetland (No. 7, 2011) to M. Kume. AI is a Fellow of the Japan Society of Promotion of Science. All experimental procedures were reviewed and approved by the Institutional Animal Care and Use Committee of the National Institute of Genetics (#23-15). We thank Kohta Yoshida and three anonymous reviewers for constructive comments on the manuscript, Mark Ravinet for helpful discussion and Rumi Suzuki, Yasuko Ogino, Yoshio Takei, Susumu Hyodo and Sanae Ogasawara for technical help. We have no conflict of interest.