Local adaptation to salinity in the three-spined stickleback?


  • J. DeFaveri,

    Corresponding author
    1. Ecological Genetics Research Unit, Department of Biosciences, University of Helsinki, Helsinki, Finland
    • Correspondence: Jacquelin DeFaveri, Ecological Genetics Research Unit, Department of Biosciences, University of Helsinki, PO Box 65, Helsinki FI-00014, Finland. Tel.: +358 9 191 57710; fax: +358 9 191 57694;

      e-mail: jacquelin.defaveri@helsinki.fi

    Search for more papers by this author
  • J. Merilä

    1. Ecological Genetics Research Unit, Department of Biosciences, University of Helsinki, Helsinki, Finland
    Search for more papers by this author


Different lines of evidence suggest that the occurrence and extent of local adaptation in high gene flow marine environments – even in mobile and long-lived vertebrates with complex life cycles – may be more widespread than earlier thought. We conducted a common garden experiment to test for local adaptation to salinity in Baltic Sea sticklebacks (Gasterosteus aculeatus). Fish from three different native salinity regimes (high, mid and low) were subjected to three salinity treatments (high, mid and low) in a full-factorial experimental design. Irrespective of their origin, fish subjected to low (and mid) salinity treatments exhibited higher juvenile survival, grew to largest sizes and were in better condition than fish subjected to the high salinity treatment. However, a significant interaction between native and treatment salinities – resulting mainly from the poor performance of fish native to low salinity in the high salinity treatment – provided clear cut evidence for adaptation to local variation in salinity. Additional support for this inference was provided by the fact that the results concur with an earlier demonstration of significant differentiation in a number of genes with osmoregulatory functions across the same populations and that the population-specific responses to salinity treatments exceeded that to be expected by random genetic drift.


The role of ecological and environmental heterogeneity in initiating population divergence has received much attention in recent years (e.g. Schluter, 2000; Nosil, 2012). When environmental differences exist among local habitats, the associated divergent selection pressures can act on traits that promote a fitness advantage in local conditions, resulting in locally adapted populations (Kawecki & Ebert, 2004). As such, many studies have been aimed at assessing the presence and magnitude of local adaptation by comparing populations that have evolved under different environmental conditions (reviews in: Savolainen et al., 2007; Leimu & Fischer, 2008; Hereford, 2009; Fraser et al., 2011).

In the strictest sense, local adaptation is defined as not only native genotypes having the highest fitness in their local environment as compared to foreign genotypes originating from different environments (i.e. ‘local vs. foreign’ criterion), but also vice versa (i.e. ‘home vs. away’ criteria; Kawecki & Ebert, 2004). As such, trade-offs are likely to arise as a result of alleles having opposite fitness effects in different habitats. Therefore, reduced fitness in a different environment is also an important element of local adaptation. However, it has been suggested that this ‘home vs. away’ criterion alone does not provide as much support for local adaptation as the ‘local vs. foreign’ criterion because it may only be reflecting inherent differences in habitat characteristics – hence, environmental properties – rather than true genetic products of evolution (Kawecki & Ebert, 2004; Fraser et al., 2011). A third contrast, sympatric vs. allopatric, has recently been introduced by Blanquart et al. (2013) in attempt to reconcile these issues. These authors suggest that local adaptation should be viewed as a property of a metapopulation rather than a single population and should therefore be defined as the difference between a population's fitness in sympatry (home sites) and allopatry (away sites; Blanquart et al., 2013).

Regardless of the criteria, genotype × environment interactions are necessary for local adaptation. A classic approach towards directly demonstrating these involves comparing fitness-related traits between populations originating from different environments, when reared in the same environmental conditions (Kawecki & Ebert, 2004). Reciprocal transplant experiments carried out in the field provide a powerful way to conclusively demonstrate genetically based fitness differences among populations in their original habitats. However, practical limitations usually restrict the feasibility of these types of field experiments. Moreover, it is often difficult to identify the exact selective agents driving local adaptation in the field, where many environmental factors vary concordantly. Reproduction of standardized habitats in the laboratory under ‘common garden’ conditions offers an alternative approach to field transplants. Although common garden experiments provide the benefit of being able to isolate particular variables that differ between habitats and directly test their role in adaptive divergence between populations, they also come with the caveat of potentially overlooking other factors that might act as key selective pressures in natural environments (Kawecki & Ebert, 2004). In addition, certain genotypes may be best adapted to the laboratory conditions, confounding the results (Kawecki & Ebert, 2004; Winkler & Van Buskirk, 2012).

Interpretations of local adaptation can be further challenged because forces other than selection (e.g. drift or migration) can also lead to genetic differentiation between populations, reproducing genotype × environment interactions similar to local adaptation (Kawecki & Ebert, 2004). However, population replication can aid in ruling out these forces as the cause of the differentiation, because different populations from similar habitats should show similar responses to given environmental factors if the differentiation is driven by those environmental factors (e.g. Langerhans & DeWitt, 2004). Similarly, populations can be reared in multiple ‘away’ habitats, allowing for more than one ‘home vs. away’ comparison. This provides the opportunity to decompose the genotype × environment interaction, isolating the interactions that are not related to selection (Thrall et al., 2002; Kawecki & Ebert, 2004; Moloney et al., 2009). Also, comparative studies of quantitative trait and molecular marker differentiation can be applied to common garden data and used to test whether the observed population divergence exceeds that expected by genetic drift (e.g. Leinonen et al., 2013).

When divergent selection pressures vary spatially as a result of environmental heterogeneity in physically continuous systems, the outcome of local adaptation largely depends on the balance between selection and gene flow. Indeed, the concept of locally adapted populations existing within a physically connected, high gene flow environment has a long history (e.g. Endler, 1973), but has gained renewed attention in recent years (e.g. Savolainen et al., 2007; Nosil et al., 2009; Blanquart et al., 2012; Via, 2012). The homogenizing effects of gene flow have traditionally been thought to counter and overwhelm the diversifying effects of selection. Yet, examples of pronounced genetic divergence continue to be revealed in high gene flow species such as wind pollinated trees (Savolainen et al., 2007), marine fishes (e.g. Conover, 1998; Conover et al., 2006; Hauser & Carvalho, 2008; Nielsen et al., 2009) and invertebrates (Grosberg & Cunningham, 2001; Sanford & Kelly, 2011), indicating that divergence in adaptive traits and their underlying alleles can accumulate among populations even in the absence of barriers to gene flow. However, much of the evidence for adaptation from genomic approaches remains indirect, as the adaptive significance of these genomic patterns remains undetermined without demonstration of their effects on fitness (Storz & Wheat, 2010; Barrett & Hoekstra, 2011). Particularly in the case with marine fishes, common garden experiments remain a rare approach towards elucidating the genetic basis of quantitative trait differentiation and their interaction with the environment (but see: Conover & Present, 1990; Schultz et al., 1996; Marcil et al., 2006; Hutchings et al., 2007; McCairns & Bernatchez, 2012; Hice et al., 2012). Moreover, most of the studies that have used a common garden approach to address local adaptation in marine fishes have compared only few demes inhabiting the extremes of the environmental gradient (but see: Hice et al., 2012). As such, the spatial scale of adaptive divergence cannot be determined due to the coarse sampling.

Indirect evidence for local adaptation in three-spined sticklebacks (Gasterosteus aculeatus) along the environmental gradient of the Baltic Sea was recently provided by a population genetic study (DeFaveri et al., 2013). Several candidate gene-associated microsatellite markers, whose putative functions were related to freshwater adaptation (Shimada et al., 2011), were identified as outliers, and their allelic variation displayed highly significant correlations with local variation in salinity (DeFaveri et al., 2013). Evidence for adaptive differentiation among Baltic Sea three-spined stickleback populations has also been obtained from a study uncovering phenotypic and the underlying genotypic differentiation in lateral plate numbers (DeFaveri & Merilä, 2013). However, in this particular case, patterns of adaptive differentiation were independent of the steep salinity gradient across the Baltic Sea. Hence, local adaptation to salinity in Baltic Sea sticklebacks remains to be demonstrated.

In this study, we conducted a common garden experiment to test for local adaptation to salinity in Baltic Sea sticklebacks. This was carried out by rearing sticklebacks from six different populations representing three different salinity regimes (viz. high, mid and low; representative of the gradient across the Baltic Sea and connecting Danish Straits) under the same three experimental salinity treatments using a full-factorial experimental design. For each population, we measured survival, body size and condition index as proxies for fitness in their native and reciprocal salinity conditions, from hatching to the age of 8 months. We also investigated whether the results of this study concur with those from an earlier genome scan study, which found evidence for adaptive differentiation in genomic regions associated with osmoregulatory functions in the same populations used here (DeFaveri et al., 2013). Using differentiation in neutral marker genes as a yardstick, we also evaluated whether observed differences in population-specific responses to salinity treatments exceeded those to be expected from random genetic drift alone.

Materials and methods


Broodstock were collected from six locations along the coast of the Baltic Sea and Kattegat Straits (Table 1; Fig. 1) during the breeding season (May–June) of 2011. The aim of the sampling scheme was to collect two replicate samples from each of three salinity regimes (viz. high, mid and low salinity, respectively), broadly representative of the salinity gradient across the Baltic Sea and Danish Straits. Average annual salinity at the two Danish strait locations is above 20‰ (Table 1; Fig. 1), deemed ‘high salinity region’ from here on; less than 10‰ at the two Central Baltic locations (Table 1; Fig. 1), deemed ‘mid salinity region’ from here on; and less than 5‰ at the Gulf of Finland and Bay of Bothnia populations (Table 1; Fig. 1), deemed ‘low salinity region’ from here on. After capture, live fish were transported to the aquaculture facility at the University of Helsinki in 50-L tanks with constant aeration. Upon arrival, they were transferred to 140 L tanks and held at their native salinities. To maintain the fish in breeding condition, a 24-h light cycle was employed, and fish were fed chironomid larvae twice daily, ad libitum.

Table 1. Sampling information of populations used in the study. Code names in bold text represent populations that were used for the long-term experiment (240 days post-fertilization). Ten full-sib families were produced from each population except KAL, from which five full-sib families were produced.
LocationCodeCoordinatesAverage annual salinity (ppt)aNative salinity classificationN.SN.L
  1. N.S, Number of individuals used at the beginning of the experiment (until 56 days post-fertilization). N.L, Number of individuals used in the long-term experiment (from 56 to 240 days post-fertilization). Number of individuals from each family can be found in Table S4.

  2. a

    From DeFaveri et al. (2013).

FiskebäckskilFIS58°14′05″N, 11°24′06″E23.52High1199
Mariager fjord MAR 56°38′58″N, 09°57′05″E26.82High1416472
KalmarKAL56°39′49″N, 16°21′25″E6.82Mid543
Tvärminne TVA 59°50′20″N, 23°12′15″E5.57Mid1140392
Kaskinen KAS 62°23′02″N, 21°13′30″E3.15Low1620431
HaminaHAM60°33′55″N, 27°12′01″E2.98Low1399
Figure 1.

A map showing the location of the study populations. High salinity populations in red, mid salinity populations in green and low salinity populations in blue. Open symbols denote populations that were reared until 56 days post-fertilization, closed symbols denote populations reared until 240 days post-fertilization. Average annual surface salinities (‰) are indicated with dashed lines.


Ten females and ten males were randomly selected from each population and used to generate ten full-sib families in vitro. Briefly, testes were dissected from each male following decapitation, and macerated in separate sterile Petri dishes with a sterile scalpel blade. Several drops of sterile 5‰ water were added to the Petri dish to make a sperm solution. Eggs were then gently squeezed from a female into the sperm solution, and a clean fine-bristled paint brush was used to gently mix the eggs with the sperm solution. Clutches from each pair were divided into three batches (within-family replicates), all of which were held in Petri dishes with 5‰ water until hatching (approximately 8 days post-fertilization). This salinity was chosen because the coastal areas of all sampling sites did not exceed 5‰. This is common for anadromous sticklebacks, which migrate to low salinity areas for breeding (Münzing, 1963), and hence, eggs in the wild would be naturally fertilized in low salinity/freshwater. Embryos were incubated at 18 °C with an 18 h light: 6 h dark cycle. Water in the Petri dishes was changed twice daily. Any unfertilized eggs/undeveloped embryos were removed and recorded.

Common garden experiment

Experimental conditions were established such that one of the within-family replicates continued to be reared in a salinity representative of their native sampling location, and other within-family replicates were reared in the two alternative salinities. Briefly, once the yolk sacs had been resorbed (approximately 3 days post-hatch), one of the groups from each family remained at 5‰ (‘low salinity’ treatment), the second group was slowly acclimated to a salinity of 10‰ (‘mid salinity’ treatment) at a rate of 5‰ per day and third group was slowly acclimated to a salinity of 20‰ (‘high salinity’ treatment) at a rate of 5‰ per day (following McCairns & Bernatchez, 2010, 2012). Fry were fed live Artemia nauplii, ad libitum, twice daily. Mortality was monitored and recorded daily. Experimental salinities were achieved by adding Instant Ocean synthetic sea salt mix (Aquarium Systems Inc., Mentor, OH, USA) to filtered water, and checked regularly with a hand-held refractometer.

Once the final experimental salinity was reached, hatchlings from each group were further subdivided into three groups (to represent within-family replication) and randomly distributed into either the front or back half of a 1.4-L tank (divided using a 400-micron screen). The back of each tank was also fitted with a 400-micron screen. Tanks were held in Allentown Zebrafish Rack Systems (referred to as ‘rack’ from here on; Aquaneering Inc., San Diego, CA, USA). A total of three racks were used, with each rack having closed recirculating water at 5‰, 10‰ or 20‰. Water temperature remained constant at 18 °C. Each rack was equipped with multilevel filtering systems, including physical, biological and UV filters. Though the rack system provided physical and chemical filtration, water quality was further maintained by manually pipetting any settled debris. Water quality (alkalinity, pH, nitrate, nitrite, hardness and ammonia) was tested weekly with 6-way test strips (Lifeguard Aquatics, Cerritos, CA, USA). Water circulation was ceased for 1 h post-feed to avoid nauplii being passed through the filter system before larvae had enough time to consume them. Larval mortality was monitored and recorded weekly. At 56 days post-fertilization, fish within each replicate were photographed together for standard length measurements.

Due to space constraints, only one population from each of the three representative salinities was maintained for long-term monitoring under the same experimental conditions as above. For these populations, the three within-family replicates were pooled and a random sample of 15–20 fish from each pool was transferred to 10-L tanks in the rack system. Fish were fed chironomid larvae twice daily, ad libitum. At 120, 180 and 240 days post-fertilization, fish within each family were photographed together for standard length measurement. All photographs were taken from a standard angle using a Nikon D60 digital camera, with a scale placed in every photograph for size reference. Digital measurements were made using the program tpsDig 2.10 (Rohlf, 2006).

After 240 days post-fertilization, three fish from each family were randomly selected for health assessment, using Fulton's condition factor (K; Ricker, 1975). Total body weight was recorded, and the relationship between body weight (WB) and length (L) was calculated as:

display math(1)

The underlying assumption of K is that fish in good condition will have higher body weights relative to their length (hence higher K values) than those in poor condition. We note that even if some part of variation in K could reflect among population differences in body form known to occur among European three-spined stickleback populations (Leinonen et al., 2006; Leinonen et al. 2011), differences in respect to treatment effects should still be informative about actual ‘condition’.

Data analyses

Survival was monitored and analysed separately on days 12, 13 and 14 post-fertilization, because these were days when experimental salinity began to be manipulated. Generalized linear mixed models (GLMMs) were conducted with the R package ‘lme4’ fitting survival as a binary response variable. Population was nested in native salinity, and family was nested in population; both were treated as random effects. Native salinity, treatment salinity and their interaction were fixed factors, whereas the density was a covariate. The same analyses were performed on the survival data for each of the following 6 weeks (i.e. until day 56 post-fertilization), with the difference that replicate effect (nested within family and population) was added as a random factor, and tank position (front/back) was included as a fixed factor. We did not use repeated-measures design as the faiths of fish were monitored at the tank, rather than individual, level.

The body length data were also analysed with GLMMs, treating body length at the given age as a normally distributed dependent variable. Native salinity and treatment salinity, as well as their interaction, were fitted as fixed effects. Population (nested within native salinity), family (nested within native salinity and population) and replicate (nested within native salinity, population and family) were fitted as random effects. Because only three of the six populations were grown after 56 days post-fertilization, the analyses for sizes at ages 120, 180 and 240 post-fertilization did not include the population term. Also, because within-family replicates were pooled and there were no partitions within tanks, replicate and tank position were not included in these later stage analyses. In all analyses, density at the time of measurements was included as a covariate. Because there were no differences in maternal size among all populations (F5,49 = 1.435, P = 0.22), it is likely that maternal effects associated with maternal size would not impact our results. As such, this information was not included into the analyses. Condition index was analysed only at day 240 post-fertilization (same model as for the size at this age), which was the only time point when the fish were weighed. All size and condition index analyses were conducted with JMP Pro 10 (ver. 10.0.2; SAS Institute Inc., Cary, NC, USA) package.

To test whether the among-population divergence in survival probability at 56 days post-fertilization, and size at 240 days post-fertilization in each salinity treatment was greater than would be expected if they had evolved neutrally, the QST − FST approach of Whitlock & Guillaume (2009) was followed. Briefly, the expected among-population variance component (σ2BE) of a neutrally evolving trait was obtained using FST and the within-population variance component (σ2W) from the family effect in the mortality and size data. The FST was calculated using data from DeFaveri et al. (2013), which included genotypes at 20 microsatellite loci, selected based on their distance from annotated genes in the stickleback genome and hence presumably neutral, from individuals collected at the same six locations as in the current study. The σ2W from the size data was calculated separately for each salinity treatment using a linear model, which included population and family (nested within population) as random effects, and density as a covariate. This within-population variance component and the expected among-population variance component were used to calculate the expected QST of a neutral trait as (Whitlock & Guillaume, 2009):

display math(2)

The among-population variance component (σ2B) for body size and survival probability in each treatment was extracted from the linear models explained above to calculate the observed QST, following eqn (1) (σ2B in place of σ2BE). The null hypothesis of a neutrally evolving trait, QST − FST, was tested by simulating QST − FST 1000 times and observing if it fell outside of the distribution of the neutrally evolving trait. These analyses were performed in R, using the codes from Lind et al. (2011).

Because the earlier candidate gene-based marker data from DeFaveri et al. (2013) suggest adaptive salinity-related differentiation among Baltic Sea stickleback populations, we tested for concurrent patterns of quantitative genetic (this study) and population genetic (DeFaveri et al., 2013) differentiation across the same six populations used in both studies. Specifically, we aimed to confirm that the divergence in the genes studied by DeFaveri et al. (2013) – whose putative roles are functionally relevant for freshwater adaptation (Shimada et al., 2011) – is related to differences in local salinity conditions among the populations included in this study. Following the classification of native salinity for each population (high, mid or low; Table 1), we calculated global and pairwise FST-estimates among populations from similar (i.e. sympatric) and different (i.e. allopatric) salinities, using FSTAT 2.9.3 software (Goudet, 2001). This was carried out with the candidate gene-based markers (referred to as ‘genic’ henceforth) identified as outliers in DeFaveri et al. (2013), plus an additional seven candidate genes with putative osmoregulatory and growth roles; n = 14 loci (Table S1). These estimates were compared with those calculated from 20 neutral (i.e. ‘nongenic’) markers (see DeFaveri et al., 2013 for details). The expectation was to observe (i) increasing levels of divergence in the genic markers among populations separated by increasing salinity differences and (ii) that this divergence exceeds that in the nongenic neutral markers in the same comparisons. For pairwise estimates, correlations between genetic and environmental distances were tested for with partial Mantel tests to control for geographic distance. This was performed separately for genic and nongenic markers, using the vegan package (Oksanen et al., 2007) in R.



Visual inspection of the patterns of survival in the different experimental groups indicated that survival probabilities were highest in the low salinity treatment (regardless of native salinity) and lowest in the high salinity treatment, especially for fish originating from the low salinity localities (Fig. 2). The GLMM confirmed these interpretations: the effect of treatment was significant on the second day of salinity manipulation (day 13 post-fertilization) and remained significant at the rest of the time points (Table 2). The treatment × salinity interaction was significant on the final day of salinity manipulation (day 14 post-fertilization) and remained significant at the rest of the time points (Table 2). Survival probabilities at all time points were also negatively affected by increasing density (Table 2), whereas the main effect of native salinity was nonsignificant until the last two time points (days 49 and 56 post-fertilization; Table 2).

Table 2. Results of the generalized linear mixed models testing for variation in survival probability in different time points (Day; days post-fertilization) of the experiments.
Native salinity (N)Treatment salinity (T)N*TDensityTank position
Estimate z Estimate z Estimate z Estimate z Estimate z
  1. *P < 0.05, **P < 0.01, ***P < 0.001.

12− 0.016− 1.600.0060.610− 0.04− 0.021− 7.14***
13− 0.008− 0.840.0252.71**− 0.001− 1.54− 0.021− 7.16***
14− 0.006− 0.640.0303.24**− 0.001− 1.91*− 0.020− 7.13***
21− 0.011− 1.160.0313.46**− 0.001− 1.89*− 0.021− 7.51***0.0580.99
28− 0.011− 1.190.0374.33**− 0.002− 2.53*− 0.021− 7.70***0.0480.87
35− 0.013− 1.510.0414.78***− 0.002− 2.84**− 0.019− 7.46***0< 0.01
420.014− 1.570.0445.31***− 0.002− 3.23**0.019− 7.65***− 0.028− 0.53
490.056− 1.73*0.0441.64**− 0.003− 3.14**− 0.148− 4.05***− 0.315− 1.95
560.019*− 2.28*0.0526.52***− 0.0023.94***− 0.019− 8.18***− 0.125− 2.48*
Figure 2.

Cumulative proportion of surviving individuals in different experimental groups as function of time (days post-fertilization). High salinity populations in red, mid salinity populations in green and low salinity populations in blue. Bold line denotes high salinity treatment, thin line mid salinity treatment, dashed line low salinity treatment. Salinity treatments were employed at day 11 post-fertilization.

Body size

At all four time points, the main effect of native salinity was not significant, whereas the effect of treatment salinity was always significant (Table 3). In general, increasing salinity tended to reduce the size of individuals (Fig. 3). However, apart from this main effect of treatment salinity, there was also a significant interaction between treatment and native salinities in all ages (Table 3). Inspection of Fig. 3 shows that these interactions derive mainly from two effects: individuals originating from the high salinity population were always largest in the high salinity treatment, whereas those originating from the low salinity population were always smallest (Fig. 3). At 120, 180 and 240 days post-fertilization, this difference was significant. In the other two treatments, the populations responded to salinity roughly in a similar fashion, and there is little indication of ‘fitness’ trade-offs in these treatment conditions (Fig. 3). Apart from these effects, the density had a negative and significant effect on size in all treatments (Table 3), as did the tank position (fish were smaller in the front side of the tanks) after 56 days post-fertilization (Table 3). The random effect of family was always significant and accounted for 4–23% of variance depending on the measurement occasion (results not shown). The random effects of population were never significant (results not shown), indicating that variation due to population replicates within habitats was an unimportant source of variance in the data. Also the variance component due to within-family replication was always nonsignificant (results not shown).

Table 3. Results from generalized linear mixed models explaining variation in body length at four different time points.
SourceDays post-fertilization
n.d.f.d.d.f. F n.d.f.d.d.f. F n.d.f.d.d.f. F n.d.f.d.d.f. F
  1. n.d.f., numerator degrees of freedom; d.d.f., denominator degrees of freedom.

  2. **P < 0.01, ***P < 0.001.

Native salinity (N)23.100.29229.720.66229.392.74229.582.09
Treatment salinity (T)2210550.57***2113134.49***2105447.05***2111557.76***
Tank position170.7450.32***
Figure 3.

Mean (± SE) body length of fish from different native salinities in different treatment salinities at different time points. (a) 56 days, (b) 120 days, (c) 180 days and (d) 240 days post-fertilization. The plotted values are mean square estimates from generalized linear mixed models reported in Table 3. Black vertical bars represent significant differences according to ‘local vs. foreign’ criteria of local adaptation; blue, green and red horizontal lines represent significant differences according to the ‘home vs. away’ criteria for the native low, mid and high salinity populations, respectively. Asterisks (*) refer to significant (P < 0.05) differences according to Tukey's post hoc tests.

Condition index

The model for condition index explained 8.7% of variation in the data, and the only significant terms in the model were treatment salinity (F2,220.2 = 6.79, P = 0.001) and the negative effect of density (b = − 0.0092, SE = 0.0037; F1,17.54 = 6.12, P = 0.02). Fish from low salinity treatment had higher condition indices (x = 1.112 ± 0.012) than those from mid (x = 1.064 ± 0.011) or high (x = 1.045 ± 0.013) treatments. The means for mid and high salinity treatments were not significantly different (Tukey's HSD tests; P > 0.05). Native salinity (F2,223.7 = 1.82, P = 0.16) as well as the interaction between treatment and native salinity (F2,256 = 0.68, P = 0.60) were not significant. The random effect of family accounted 0.02% of variance.

Comparison to molecular data

Comparison of population differentiation in neutral nongenic markers and QST-estimates within each salinity treatment revealed that, for mortality (at day 56 post-fertilization), divergence within the low salinity treatment was low (QST = 0.04), and not significantly different from what would be expected from drift alone (FST = 0.01; Fig. 4a). Divergence in the mid and high salinity treatments was higher than that for a neutrally evolving trait (QST = 0.09 and 0.47, respectively; Fig. 4a). At day 240 post-fertilization, divergence in size within the low (QST = 0.03) and mid (QST = 0.01) salinity treatments was low, and not significantly different than what would be expected under pure drift alone (Fig. 4b). However, in the high salinity treatment QST was higher (QST = 0.11) and fell outside the tail of the distribution of a neutrally evolving trait (Fig. 4b). Hence, the size divergence in the high salinity treatment exceeded neutral expectation.

Figure 4.

Simulated distributions of QST − FST for neutrally evolving traits in the case of (a) mortality at 56 days post-fertilization and (b) size at 240 days post-fertilization. The observed values of QST − FST in the low, mid and high salinity treatment are indicated with blue, green and red arrows, respectively.

Further evidence for adaptive differentiation among populations originating from high and low salinity localities was provided by comparison of the genic and nongenic marker data from DeFaveri et al. (2013), in the six populations used in both studies. There was no differentiation among populations originating from the same salinity locations (i.e. sympatric comparisons), neither in genic nor in nongenic markers (Fig. 5; see DeFaveri et al., 2013 for details). However, in allopatric comparisons, the degree of population differentiation in the genic markers reflected the differences in salinity at the sampling locations (Fig. 5; Table S2). This pattern was not observed in the neutral nongenic markers. For example, differentiation between the mid and high salinity populations was similar to that between the low and high salinity populations in the nongenic markers, even though there is a larger environmental difference between the latter pair (Fig. 5; Table S2). On the other hand, differentiation in the genic markers increased with increasing differences in salinity (Fig. 5; Table S2). Accordingly, a significant correlation was observed between genetic and environmental distances after controlling for geographic distance (partial R = 0.69, P = 0.01). This was also true for the nongenic markers, albeit to a lesser degree (partial R = 0.51, P = 0.02).

Figure 5.

Levels of genetic differentiation as reflected in global FST-estimates (± 95% confidence intervals) in genic (filled circles) and nongenic (open circles) microsatellite markers between six stickleback populations originating from high (H), mid (M) or low (L) salinity areas. Differences in salinity between sampling sites are indicated with grey squares. Marker data and average annual surface salinities at each sampling location were obtained from DeFaveri et al. (2013).


The most salient finding of this study was the reduced survival probability and body size of sticklebacks originating from low salinity regions of the Baltic Sea when reared under high salinity conditions, fulfilling the ‘home vs. away’ criterion of local adaptation. The ‘local vs. foreign’ criterion (cf. Kawecki & Ebert, 2004) was fulfilled by the populations originating from high salinity regions, as they had better survival and growth than the low and mid salinity populations in the high salinity treatment. Divergence in both mortality and size at the end of the experiment in the high salinity treatment was greater than would be expected by drift alone, providing further support for salinity driven adaptive divergence. Nevertheless, all populations had the best survival and growth in the low salinity treatment. Data from an earlier population genetic study were also used to demonstrate that genetic divergence among the allopatric populations increased with the extent of environmental differences, as measured with candidate gene-based – but not when measured with neutral nongenic – markers. In the following, these findings are discussed in light of local adaptation of sticklebacks to different salinity regimes.

Local adaptation

According to the conventional definition proposed by Kawecki & Ebert (2004), a strict demonstration of local adaptation requires evidence for a genetic fitness trade-off between native and foreign environments. However, many studies of local adaptation indicate that the lack of trade-off may not be sufficient criteria to rule out local adaptation (e.g. Gomes-Mestre & Tejedo, 2003). In fact, a meta-analysis of local adaptation in plants revealed that fitness trade-offs were reported in less than 50% of the cases studied (Leimu & Fischer, 2008). Recently, Blanquart et al. (2013) demonstrated that if local adaptation is quantified on the metapopulation – rather than population – level, the average degree of local adaptation can be similar whether the fitness of one population is higher in ‘away’ habitats (in our case, the high salinity population in low salinity treatment) or if the populations exhibit classical fitness trade-offs. This was also highlighted by Conover & Schultz (1995) in the context of countergradient variation studies, which show that there may be covariance – instead of, or in addition to, interaction – between the genotypic and environmental contribution to a phenotype.

The conventional criteria for local adaptation might not always match the expectations stemming from knowledge of the biology of a given species. For example, most stickleback populations residing in highly saline marine environments migrate to low salinity regions for breeding (Münzing, 1963; Guderly, 1994). Since reproduction, fertilization and early development therefore occur in low salinity conditions, high survival and growth rates could be expected in low salinity during early life stages – regardless of the salinity in their native habitat. Hence, fitness trade-offs across environments should not be expected. This was in fact observed in our data: all populations had the highest survival rates and condition factor in low as compared to the other salinity treatments. Similar results have also been reported by Marchinko & Schluter (2007), who found that marine sticklebacks had the highest hatching success in freshwater – compared not only to their hatching success in sea water, but also to that of the freshwater populations in freshwater. Likewise, McCairns & Bernatchez (2010) reared marine and freshwater sticklebacks in their native and reciprocal salinities and also found similar juvenile survival rates and individual fitness between demes in the freshwater treatment. While a study by Campeau et al. (1984) conversely demonstrated high mortality of anadromous larvae in low salinity, it is important to note that in their experimental design, eggs were hatched in high salinity (20‰) and subsequently transferred to low salinity. However, Belanger et al. (1987) found that if anadromous eggs were hatched in low salinity – as was also carried out in our experiment – larval survival and growth rates were higher in low than in high salinity, concluding that the salinity experienced during development influences subsequent salinity tolerance. Hence, the strong performance of the high salinity populations in low salinity may also be related to the fact that they were hatched in low salinity. Ultimately, the expectation that marine sticklebacks should perform well in freshwater matches their biology, but may end up conflicting with the expectations of trade-offs for local adaptation.

A reduction in fitness in freshwater sticklebacks transferred to sea water was anticipated: after observing reduced osmoregulatory efficiency of freshwater sticklebacks in seawater, Gutz (1970) suggested that adaptation to low salinity environments compromises their osmoregulatory capacity in seawater. Our data support this conjecture, as survival rates and size of the native low salinity populations were significantly reduced in the high salinity treatment. Similarly, a significant reduction in fertilization and hatching success among freshwater sticklebacks reared in seawater was observed by Marchinko & Schluter (2007). McCairns & Bernatchez (2010) also noted a 22% reduction in relative survival of the freshwater deme in sea water, indicating that plasticity in salt tolerance has been lost among freshwater populations. Hence, the reduction in survival and size of low salinity populations in ‘away’ conditions satisfies a component of local adaptation and also matches the expectations of stickleback biology. Moreover, the divergence in size within the high salinity treatment exceeded neutral expectations, reinforcing the interpretation that selection was a likely driver of the smaller size of the low salinity fish. Because small size means smaller reproductive output (Wootton, 1984; Baker et al., 2008), a reduction in size may translate to lowered fitness through the production of smaller clutches (i.e. fewer individuals). Finally, as size-assortative mating has previously been demonstrated in sticklebacks (Borland, 1986; Nagel & Schluter, 1998; Conte & Schluter, 2013), it would be of interest to study whether salinity-related size divergence contributes to the reduction in gene flow between high and low salinity environments through the evolution of reproductive barriers.

Salinity, genetics and Baltic Sea adaptation

Reduced gene flow between populations in high and low salinity regions of the Baltic Sea has been noted in several species of fish, including herring (Jørgensen et al., 2005; Gaggiotti et al., 2009; Limborg et al., 2012), turbot (Nielsen et al., 2004), sprat (Limborg et al., 2009), flounder (Hemmer-Hansen et al., 2007) and three-spined sticklebacks (DeFaveri et al., 2013). However, the conclusions that these populations are adapted to their local conditions are primarily based on observed shifts in allele frequencies in neutral or candidate gene-based markers, which are correlated with changes in salinity. Yet, the adaptive significance of these genetic patterns has not been experimentally confirmed. On the other hand, the few studies that have applied experimental approaches to directly test for local adaptation with respect to spatial variation in salinity have not provided validation for their inferences with molecular genetic approaches. Moreover, they have only focussed on the ‘home vs. away’ aspect by subjecting single populations to different salinity treatments (e.g. Kuhlman & Quantz, 1980; Nissling & Westin, 1991a,b; Westin & Nissling, 1991; Nissling et al., 1994; Karås & Klingsheim, 1997; Jørgensen et al., 2010). As such, the results of this study represent an important contribution towards providing clear evidence for local adaptation in the Baltic Sea through the integration of experimental and genetic data. Specifically, we observed marked allele frequency shifts in candidate genes with functions related to osmoregulation and growth (Shimada et al., 2011) that corresponded with differences in environmental salinity among our study populations. Not only did the observed shifts in these genic markers exceed neutral expectations established by differentiation in nongenic markers, but they also aligned with the observed divergence in mortality rates and mean body size among different populations when grown in high salinity treatment. This combined evidence for salinity-associated differentiation in both population and quantitative genetic data strongly suggests an adaptive basis for the observed differentiation.

Maternal and plate morph effects on growth

There is growing evidence that an individuals' growth and development can be influenced by maternal and other cross-generational effects (e.g. Rossiter, 1996; Salinas & Munch, 2012). In this study, all the inference was based on responses of first generation laboratory fish whose parents originated from the wild. Hence, it is possible that rather than reflecting strictly genetic effects, part of the observed treatment responses have been influenced by maternal or transgenerational effects, and their interactions with treatments. Although these possibilities cannot be ruled out with the data in hand, the fact that patterns of treatment responses among populations aligned with patterns among population differentiation in markers associated with genes with functional roles in osmoregulation suggests that the observed differentiation was indeed genetically based. Nevertheless, a multigenerational study utilizing second or third generation laboratory fish would be needed to dissect the genetic, epigenetic and environmental influences from each other.

Furthermore, several stickleback studies have suggested that the number of lateral plates affects growth differently depending on the salinity in which they are raised. However it seems as though the interaction between salinity and growth rate may not be as clearly related to plate morph as previously thought. For example, although some studies have suggested that fully plated marine sticklebacks experience slower growth rates and reach smaller sizes than low-plated sticklebacks in freshwater (e.g. Marchinko & Schluter, 2007; Barrett et al., 2009), others have reported higher growth rates of a fully plated anadromous population in freshwater (e.g. Spence et al., 2012). Some have found different growth trajectories between marine and freshwater demes, even though both were fully plated (e.g. McCairns & Bernatchez, 2010) whereas others have not found any differences between different plate morphs (e.g. McGuigan et al., 2011). Hence, although we did not include plate morph data into our analysis, it is unlikely that the effects of plate morph were confounding our results. Specifically, one of the low salinity populations was fully plated (HAM), whereas the other was polymorphic for all the three plate morphs (KAS; Table S3). The two mid salinity populations (TVA and KAL) were also both polymorphic with similar proportions of each plate morph as KAS. Yet, at the end of the experiment, there was no difference in size between the mid and high salinity population (FIS), whereas the low salinity population was significantly smaller. Hence, despite having similar frequencies of each plate morph, the low salinity population was comparatively smaller than the mid salinity population in the high salinity treatment. Therefore, variation in plate morph frequencies cannot easily explain the results of the present study.

Phenotypic and marker-based inferences of selection

It is worth taking note that the there was no evidence for local adaptation in the mid-salinity populations. This stands in contrast to the evidence from marker-based (DeFaveri et al., 2013) and quantitative trait (DeFaveri & Merilä, 2013) analyses, which suggest for more fine-grained scale of adaptive differentiation in sticklebacks within the Baltic Sea. Similar results are available from stickleback studies made in St. Lawrence estuary: whereas population genetic and gene expression analyses suggested physiologically based local adaptation (McCairns & Bernatchez, 2008, 2010), phenotypic analyses did not support this (McCairns & Bernatchez, 2012). Likewise, in recent study of sea urchin responses to experimental ocean acidification, Pespeni et al. (2013) found clear patterns of genome-wide selection in hundreds of loci with putative functions in mediating adaptation to acidification. At the same time, larval development and morphology showed little response to acidification (Pespeni et al., 2013). Hence, it appears that although local adaptation to environmental stressors such as salinity and acidification can be built up at genomic level, this differentiation is not necessarily always expressed at phenotypic level. There are number of possible explanations for these discordant results with phenotypic and genotypic data sets. These include for instance the experimental conditions under which phenotypic inference is made, as well as buffering of growth and development through compensatory plastic responses. Hence, the fact that mid-salinity populations showed little evidence for local adaptation at phenotypic level does not conflict with the fact that genomic data suggest local adaptation also at this scale (cf. Fig. 5). Nevertheless, these considerations emphasize the need for future studies looking into the actual physiological differentiation in responses to salinity treatments, including studies of gill and kidney structure – key traits known to be associated with salinity tolerance in sticklebacks (Guderly, 1994).


The results of this study substantiate earlier inferences – born from genome scan (DeFaveri et al., 2013) and phenotypic (DeFaveri & Merilä, 2013) studies – that three-spined sticklebacks in the high gene flow Baltic Sea environment are genetically structured, and this structuring appears to be, at least partly, driven by spatial heterogeneity in salinity. The results also provide strong support for the contention that sticklebacks from low salinity regions might have a reduced capacity to deal with high salinity conditions, whereas the capacity to survive and grow well in low salinity is retained in the ancestral marine populations. In general, the combined quantitative and molecular genetic data presented in this study makes an important contribution to the otherwise thin and mostly circumstantial evidence for local adaptation in the Baltic Sea organisms.


We thank Sara Neggazi, Niina Nurmi, Mehedi Hasan, Peter Kullberg, Ismo Rautiainen and Varpu Pärssinen for their assistance with field work and fish husbandry. Special thanks to Mackenzie Sgro for helping with the fish measurements and to Rymy for support in manuscript finalization phases. Fish were collected in accordance with national fishery regulations, with permissions issued from the counties requiring such (Kalmar Län: 623-1902-11P; Västra Gotlands Län: 623-8412-2011; Denmark: 2011-01696). This research was supported by grants for Academy of Finland (#:s 200940, 108601 and 118673 to JM). The research leading to these results has also received funding from the European Community's Seventh Framework Programme (FP/2007-2013) under grant agreement no 217246 made with BONUS, the joint Baltic Sea research and development programme.