Population differences in response to hypoxic stress in Atlantic salmon


Correspondence: Jessica Côte, INRA, UMR 985 Ecologie et Santé des Ecosystèmes, F-35042 Rennes, France.

Tel.: +33 2 23 48 55 29; fax: +33 2 23 48 54 40; e-mail: cote.jessica@wanadoo.fr


Understanding whether populations can adapt to new environmental conditions is a major issue in conservation and evolutionary biology. Aquatic organisms are increasingly exposed to environmental changes linked with human activities in river catchments. For instance, the clogging of bottom substratum by fine sediments is observed in many rivers and usually leads to a decrease in dissolved oxygen concentrations in gravel beds. Such hypoxic stress can alter the development and even be lethal for Atlantic salmon (Salmo salar) embryos that spend their early life into gravel beds. In this study, we used a common garden experiment to compare the responses to hypoxic stress of four genetically differentiated and environmentally contrasted populations. We used factorial crossing designs to measure additive genetic variation of early life-history traits in each population. Embryos were reared under normoxic and hypoxic conditions, and we measured their survival, incubation time and length at the end of embryonic development. Under hypoxic conditions, embryos had a lower survival and hatched later than in normoxic conditions. We found different hypoxia reaction norms among populations, but almost no population effect in both treatments. We also detected significant sire × treatment interactions in most populations and a tendency for heritability values to be lower under stressful conditions. Overall, these results reveal a high degree of phenotypic plasticity in salmon populations that nevertheless differ in their adaptive potential to hypoxia given the distinct reaction norms observed between and within populations.


In the current context of global changes, numerous species face severe modifications of their environment (Vitousek et al., 1997; Balmford et al., 2003). Except migrating to unaltered habitats, there are two nonexclusive ways by which species can adapt to these changes. First, adaptive changes are possible for traits with a significant additive genetic variation (i.e. heritable traits). Such a micro-evolution has been described in a number of species even in few generations (Merila et al., 2001; Kruuk et al., 2002; Hoffmann & Sgro, 2011). Second, phenotypic plasticity can allow organisms to cope with environmental changes when a given genotype can express various phenotypes in a range of environments (Stearns, 1989). However, phenotypic plasticity can itself have a genetic basis and heritable reaction norms (i.e. genotype × environment interactions) have been described in many taxa (Haugen & Vollestad, 2000; Kawecki & Ebert, 2004; Fraser et al., 2007; Charmantier et al., 2008; Dingemanse et al., 2009).

The inference of mechanisms underlying the adaptation of populations to their current environment is central to predict their adaptive potential to environmental changes. Common garden experiments combined with quantitative genetics crossing designs can be used to disentangle the genetic and environmental bases of life-history traits and measure reaction norms. Such experiments have demonstrated that populations can be adapted to local optima via divergent selection leading to genetically based differences in life-history traits (Koskinen et al., 2002; Cano et al., 2004; Perry et al., 2005; Jensen et al., 2008). Alternatively, variations in such traits can be solely due to plastic responses or to a combination of genetic adaptation and phenotypic plasticity depending on traits and populations (Hendry et al., 1998; Haugen & Vollestad, 2000). Importantly, half sib crossing designs should be used to avoid including nongenetic maternal effects in measurements of genetic variation within and among populations (Hunt & Simmons, 2002). The evolutionary potential depends on standing genetic variance of populations, notably additive genetic variance (Va) that is directly linked with the heritability of a trait and thus to its response to selection. However, heritability varies with environmental conditions: lower values are usually observed under stressful conditions (Réale et al., 1999; Uller et al., 2002; Cano et al., 2004; Charmantier et al., 2004; Garant et al., 2004, 2005). As a result, experimental investigations of adaptive potential in wild populations should ideally be carried out under regular ‘benign’ settings but also under more realistic stressful conditions.

Aquatic organisms are highly sensitive to certain environmental factors, notably water temperature and dissolved oxygen (DO) levels. Adaptive potential of fish populations to variations in water temperature has been widely investigated to predict the consequences of climate change on fish populations' viability (Beacham & Murray, 1985, 1986; Hendry et al., 1998; Haugen & Vollestad, 2000; Jensen et al., 2008). However, few studies investigated the evolutionary potential of populations to lower levels of DO (Crispo & Chapman, 2008, 2010, 2011; Martinez et al., 2009). These studies were all carried out on African cichlids while hypoxia is also an important stress for fish from the northern hemisphere (Shields & Knight, 2011). Such hypoxic stress is increasingly frequent in river catchments where anthropogenic activities enhance sediment transfers to aquatic ecosystems (Waters, 1995). Notably, the deposition of fine sediments induces river substrate clogging (Brunke, 1999) and modifies the hydrochemistry of the hyporheic zone and the availability of DO to aquatic organisms (Boulton et al., 1998; Soulsby et al., 2001; Greig et al., 2007).

Atlantic salmon (Salmo salar) is an anadromous species that reproduces in rivers where juveniles spend 1–4 years before migrating to sea for 1–3 years depending on populations. Most S. salar populations are considered endangered due to anthropogenic pressures including recreational and professional fishing, alteration of feeding grounds at sea by global changes, habitat loss through dam construction or degradations of spawning habitats (Baglinière et al., 1990; Schindler, 2001). Salmon embryos develop into gravel riverbeds and are thus often exposed to hypoxic stress causing high mortality rates (Malcolm et al., 2003; Dumas et al., 2007). If embryos survive to hypoxia, their development is nonetheless slower, and they are smaller at hatching compared to nonstressed individuals (Ciuhandu et al., 2005; Greig et al., 2005; Finn, 2007). Despite the widespread incidence of hypoxic stress both in North-American and European salmon rivers (Reeves et al., 1993), no study investigated the variation among populations in embryo resistance to hypoxia.

In this study, a common garden experiment was designed to detect potential differences in early life-history traits measured under normoxic and hypoxic conditions among four Atlantic salmon populations. We investigated whether hypoxia reactions norms differed among populations and assessed their adaptive potential by (i) calculating heritability values under normal and stressful conditions and (ii) testing for genotype by environment interactions.

Materials and methods

Study populations

We studied four Atlantic salmon populations located in Western and Central France and separated by several hundred kilometres from each other (Fig. 1). These populations are genetically differentiated (pairwise FST: 0.04–0.11) and belong to four of the five genetic groups described in France (Perrier et al., 2011). These populations also live in river systems of various sizes with very distinct environmental characteristics. The Oir River is a 20-km-long tributary of the Sélune River that flows into the English Channel. This river drains an area of 87 km2 with a mean annual discharge of 1 m3 s−1. The Scorff River is 62-km-long and its watershed covers an area of 480 km2. It flows into the Atlantic Ocean, and has a mean annual discharge of 8 m3 s−1. The Oir and Scorff rivers have a similar geology characterized by metamorphic siliceous rocks. The Saison River is a 54-km-long tributary of Oloron Gave. It drains an area of 627 km2 with a mean annual discharge of 24.5 m3 s−1. The geology of this watershed is mostly made of sedimentary rocks. The Allier River is a tributary of Loire River that flows into the Atlantic Ocean. This river has a mean annual discharge of 147 m3 s−1 and drains an area of 14 310 km2 characterized by a particular geology made of volcanic and metamorphic rocks. Adult Atlantic salmon reach spawning areas in the Allier River after an upstream migration of more than 700 km, that is, the longest freshwater migration reported in Western Europe. This population is the most genetically differentiated from all other French populations and the predominance of multisea-winter fish combined with a precocious run-timing suggests a local adaptation to a long migration distance (Perrier et al., 2011). According to local environmental agencies water quality is good for Allier, Saison and Scorff Rivers but medium for Oir in relation with high levels of nitrate and suspended solids. No quantitative information is available on levels of DO in the rivers studied except for Oir River where low oxygen levels have been recorded in gravel beds where Atlantic salmon embryos develop (Massa et al., 1998). Overall, the genetic differentiation among these four populations combined with their strongly contrasted environmental characteristics suggests they may have evolved some degree of adaptive divergence and/or they may have different adaptive potentials to environmental variations.

Figure 1.

Location of study sites and description of crosses designs for each population and experiment. The life history of progenitors and the number of families generated in each experiment are indicated (G, grilse; MS, multi-sea-winter individual; MP, Mature parr).

Fish sampling and artificial fertilization

The study was performed during two consecutive winters (2009–2010, experiment 1; 2010–2011, experiment 2) to increase the number of families in each population and to investigate two levels of hypoxic stress. We used wild progenitors captured in each river to make full-factorial crossing designs with resulting embryos being reared under normoxic and hypoxic conditions. Due to technical constraints, the number of progenitors varies between populations and experiments and we used either small precocious males or large anadromous males (see Fig. 1 for details). Males and females were captured few weeks before the spawning period (November–December) on Oir, Scorff and Saison Rivers, using deep-netting, electrofishing or upstream migration traps. The capture of precocious males was not successful in Saison River during experiment 1, thus fish from the nearby Nive river were used (Fig. 1). Nive and Saison salmon populations belong to the same genetic cluster and are not significantly differentiated (Perrier et al., 2011). Fish from Allier (experiment 1 only) were caught in an upstream migration trap in spring and kept in Chanteuges fish farm facilities. Fertilizations took place either locally (Oir and Allier) or at INRA Rennes where fish or gametes were transferred (Scorff and Saison).

For each population four males were used but the number of females varied between 1 and 4 depending on the success of capture (Fig. 1). Fish were anaesthetized and gametes were individually collected by manual stripping. Twenty ova per female were kept and individually weighed after freeze-drying. Fertilizations were then made in a full-factorial design, every male being crossed with every female of the same population. For Scorff, Oir and Saison, in vitro fertilizations were made in Petri dishes according to the protocol of Jacob et al., 2010;. Ova were distributed in Petri dishes and fertilized with 40 μL of milt. Petri dishes were then half filled with water and after 1 h, eggs were carefully rinsed. Water used for fertilizations and rinsing the eggs was aerated and chemically standardized water reconstituted according to OECD guidelines (OECD 1992). For the Allier population fertilizations were done according to the protocol routinely used at the Chanteuge hatchery, which mainly differed by the use of plastic beakers and a standard diluent for fertilizations. Eggs from the Allier were then rinsed with the same water used for other populations. Over both experiments, a total of 76 families were analysed.

Rearing experiment and offspring measurements

Eggs were incubated at INRA Rennes in a recirculated water system. Water temperature was kept constant and automatically recorded every hour (mean ±SD = 10.06 °C ± 0.17 and 10.22 °C ± 0.18 for experiment 1 and experiment 2, respectively). Each sib group was split into four replicates of 15 eggs, which were placed separately in perforated containers (50 mm × 50 mm) that ensure water circulation. Two replicates were incubated with normal oxygen supply (normoxia), and two replicates with water at lower DO level (hypoxia). Hypoxic water was obtained by using an oxygen-depletion system adapted from Roussel (2007). During incubation, DO level was checked twice a day using an oxymeter (Hach portable LDO HQ10; Hach, Düsseldorf, Germany). The average ± SD DO concentrations in normoxic water were 10.37 ± 0.44 mg L−1 (min–max 9.03–10.88) and 9.99 ± 0.52 mg L−1 (min–max 9.13–11.27) for experiment 1 (Exp. 1) and experiment 2 (Exp. 2), respectively. For the hypoxic treatment, the average DO concentration was slightly higher in experiment 1 (4.79 ± 0.24 mg L−1; min–max 4.42–5.76) than in experiment 2 (4.15 ± 0.60 mg L−1; min–max 3.66–5.15). The DO level was intentionally lower in experiment 2 to challenge populations with a stronger hypoxic stress.

Unfertilized eggs that kept the appearance of ova were counted and removed from the containers at 313 Accumulated Degree-Days (ADD) after fertilization. Dead eggs and embryos were daily removed to prevent any infection. Three early life-history traits were measured on a total of 4560 embryos. For each sib group, survival was calculated as the number of embryos at 900 ADDs divided by the number of developing embryos at 313 ADDs to avoid any bias due to unfertilized eggs. The incubation time (hatching date) was recorded at the individual level. At 900 ADD, embryos were over-anaesthetized in a solution of benzocaïne (0.1%) and photographed using a digital camera (Nikon D5000; Nikon, Tokyo, Japan). Photographs were processed with imagej software (http://rsbweb.nih.gov/ij/download.html) to measure embryos' fork length to the nearest 0.01 mm.

Statistical analyses

Data were analysed for each experiment using linear mixed-effects models or generalized linear mixed-effects models implemented in the r software (http://cran.r-project.org) via the lme4 package (Pinheiro & Bates, 2000).The population, treatment and population ×treatment interaction were treated as fixed effects and sire, dam, container and dam × sire interaction were considered as random effects. Differences among populations were investigated for each treatment separately to account for significant populations × treatment interactions and significant differences in survival between treatments. We also investigated genotype × environment interactions in each population by testing sire × treatment interactions (in that case, treatment, sire and dam were considered as fixed effects). Average egg weight per female was included as a fixed covariate in all models except in within population models with a fixed dam effect. We used a binomial error family for models with survival as a response variable. Data on incubation time and offspring length were log-transformed and analysed with a Gaussian error family. To investigate the amount of between-years variation relative to the among-populations (within year) variation we tested for a year effect among the two experiments for the three traits measured in normoxia. We also tested for a year effect in each population and trait separately. The statistical significance of each factor was tested by comparisons of models including or not the focal variable using likelihood ratios tests (LRT) based on a χ2 distribution. We also investigated year and population effects on (log-transformed) egg size using hierarchical analyses of variance with female effect nested within years and female effect nested within populations, respectively. For the Scorff population, as only one female was used in experiment 1, the year effect was tested singly in a linear model.

We computed heritability for each population using an ‘animal model’ approach and a Bayesian inference developed by Waldman (2009). Theses analyses were carried out with the OpenBugs software (http://www.openbugs.info/w/, Lunn et al., 2009). Three parallel MCMC chains were run, and 50 000 iterations for each chain were retained after an initial burn-in of 10 000 iterations. Convergence of MCMC sampling was assessed with the Brooks-Gelman-Rubin diagnostic (Brooks & Gelman, 1998) and the omnibus χ2 discrepancy was used to verify the consistency between models and data (Gelman, 2004). Heritability was calculated using estimates of additive genetic variance (Va) and residual variance (Vr) as: h2 = Va/(Va + Vr). We also computed coefficients of genetic variation for each trait following Houle (1992).


Variations among experiments, treatments, populations and parents

For both experiments, the hypoxic treatment had a significant effect on all traits measured (Table 1). Overall, offspring reared under hypoxic conditions had a lower survival and hatched later than embryos reared under normoxic conditions (Fig. 2). Offspring size was lower under hypoxic conditions in experiment 2 but it was unexpectedly higher under such conditions in experiment 1 (Table 1 and Fig. 2). When considering each population separately, the treatment did not equally affect all traits and populations (Table 2). The length of embryos at the end of development was only significantly influenced by hypoxic stress in Oir and Allier populations during experiment 1 and in Oir population only during experiment 2 (Table 2). Significant population × treatment interactions were observed for all traits in experiment 2 but only for incubation time in experiment 1 (Table 1). Surprisingly, we observed a significant difference among populations only for survival under normoxic conditions (LRT χ2 = 6.77, P = 0.03, Table S2). This effect seems to be due to the low survival recorded for the Saison population (Fig. 1, Tukey tests: Oir-Saison P = 0.007, Saison-Scorff P = 0.006, Scorff-Oir P = 0.99). Egg size did not differ among populations for both experiments (Exp. 1: F3,209 = 0.65, P = 0.61; Exp. 2: F2,265 = 0.21, P = 0.88). Egg size only differed among the 2 years for the Scorff population in which the single female used in experiment 1 had larger eggs than females from experiment 2 (average ± SD in Exp. 1: 0.13 ± 0.003 (g); Exp. 2: 0.09 ± 0.003; F1,58 = 1493.5, P < 0.001). Considering the three traits measured in normoxia, a significant year effect was only observed for embryo length (LRT: χ2 = 24.32, P < 0.001). This effect was linked with smaller embryos in the Scorff and Saison populations in experiment 2 (LRTs, Scorff: χ2 = 12.86, P < 0.001; Saison: χ2 = 9.64, P = 0.002; Oir: χ2 = 0, P = 1; Fig. 2e,f).

Figure 2.

Hypoxia reaction norms for survival, incubation time and length of embryos in four Atlantic salmon populations (means ± SE). Results from experiment 1 (a, c, e) and experiment 2 (b, d, f) are presented for Oir (black solid line), Scorff (dotted line), Saison (dashed line) and Allier (grey solid line).

Table 1. Results of linear mixed models testing the effects of population (Pop.), treatment (Tr.) and their interaction on early Salmo salar life-history traits for both experiments. P-values in bold are significant
 Experiment 1Experiment 2
Tr.38.21 <0.001 145.631 <0.001
Pop. × Tr.4.333ns13.182 0.001
Egg size1.841ns0.011ns
Incubation time
Tr.228.701 <0.001 154.831 <0.001
Pop. × Tr.13.623 0.003 30.302 <0.001
Egg size01ns01ns
Tr.16.171 <0.001 7.251 0.007
Pop. × Tr.1.663ns30.272 <0.001
Egg size32.861 <0.001 9.591 0.002
Table 2. Linear mixed models testing treatment (Tr.), parental fixed effects and their interactions on early life-history traits of Atlantic salmon. For each population, P-values obtained from experiment 1 (Exp. 1) and experiment 2 (Exp. 2) are in bold when significant. Degrees of freedom are indicated in brackets; Dam and Dam × Sire effects could not be tested for the Scorff population in experiment 1 because only one female was used (–). The Allier population was only used in experiment 1
Exp. 1Exp. 2Exp. 1Exp. 2Exp. 1Exp. 2Exp. 1
Tr. <0.001 (1) <0.001 (1) ns(1) 0.004 (1) 0.004 (1) 0.001 (1) 0.002 (1)
Sirens(3) 0.006 (3) ns(3) 0.006 (3) 0.010 (3) ns(2) <0.001 (3)
Sire × Tr.0.080(3)ns(3)ns(3)ns(3) 0.020 (3) ns(3)ns(3)
Dam <0.001 (3) 0.001 (2) ns(3)ns(1) 0.010 (3) 0.003 (3)
Dam × Sire0.050(9)ns(6)ns(3) <0.001 (3) ns(6)0.070(9)
Incubation time
Tr. <0.001 (1) <0.001 (1) <0.001 (1) <0.001 (1) <0.001 (1) <0.001 (1) <0.001 (1)
Sire <0.001 (3) 0.060(3)ns(3)ns(3) 0.040 (3) ns(2) <0.001 (3)
Sire × Tr.ns(3)ns(3) 0.040 (3) ns(3)ns(3)ns(3) 0.030 (3)
Dam <0.001 (3) 0.001 (2) ns(1)ns(1) <0.001 (3) <0.001 (3)
Dam × Sirens(9)ns(6)ns(3)ns(3)ns(6)ns(9)
Tr. 0.003 (1) <0.001 (1) ns(1)ns(1)ns(1)ns(1) 0.020 (1)
Sirens(3)ns(3)ns(3) 0.002 (3) ns(3) 0.003 (2) <0.001 (3)
Sire × Tr.ns(3)ns(3)ns(3)ns(3) 0.009 (3) 0.020 (3) ns(3)
Dam <0.001 (3) <0.001 (2) ns(1) <0.001 (1) <0.001 (3) <0.001 (3)
Dam × Sirens(9)0.090(6)ns(3)ns(3)ns(6)ns(9)

Parental effects are presented for each population and experiment in Table 2. As expected, dam effects are often significant on embryo length and incubation time but also survival (Table 2). Dam effects on embryo length were probably partly linked with significant egg size variations (Tables 1 and S1). Significant sire (genetic) effects are observed for all traits in the Allier population and for at least one trait in other populations (Table 2). Sire × dam interactions are only significant for survival in the Saison population (Table 2). Sire × treatment interactions are observed at least once in each population except for Oir River (Table 2). Such genotypes × environment interactions are also more frequent in experiment 1 than in Experiment 2 (Table 2).

Quantitative genetic variation

Heritability varied between 0 and 0.71 depending on traits and populations (Table 3). Heritability for survival in the Scorff population in Experiment 1 could not be computed because of a lack of convergence of MCMC chains for this data set (Tables 3, S2 and S3). Null heritability values were noticed for survival and highest values were observed for length. For incubation time and length, heritability tended to be higher under normoxic conditions than hypoxic conditions, except for the Saison population. However, 95% Bayesian credible intervals were overlapping between both treatments except for the Oir population in Experiment 2. Coefficients of genetic variation followed a similar trend of lower heritability in hypoxia (Table 3). Accordingly, higher Va values were observed under normoxic conditions except for the Saison population (Table S2). Residual variation usually did not significantly differ between normoxia and hypoxia except five cases of significant increase under hypoxia in Experiment 2 (Oir and Scorff populations, Table S3).

Table 3. Heritability values for each early life-history trait in both treatments (N, normoxia; H, hypoxia) and for the two experiments. Significant P-values are indicated in bold, 95% Bayesian credible intervals are in brackets and the coefficient of genetic variation in italic (Houle, 1992)
 Experiment 1Experiment 2
SurvivalIncubation timeLengthSurvivalIncubation timeLength
  1. –, data not available.


8.7 × 10−17

[1.6 × 10−30–2.8 × 10−25]

3.3 × 10 5







4.0 × 1015

[2.3 × 10−31–3.4 × 10−22]

1.1 × 10 4








1 × 1.10−19

[7.2 × 10−32–3.0 × 10−28]

9.9 × 10 6







6.6 × 10−3

[1.7 × 10−30–5 × 10−2]


1.08 × 10−9

[1.7 × 10−31–2 × 10−9]

5.9 × 10 4









[2 × 10−3–0.99]


6.0 × 10−21

[0–7.2 × 10−20]

9.4 × 10 6


[6.8 × 10−6–0.99]







[1.8 × 10−36–0.75]



[1 × 10−3–0.63]


9.1 × 10−6

[0–4.07 × 10−8]



[2.2 × 10−6–0.75]



[4.0 × 10−29–0.02]



2.5 × 10−24

[1.3 × 10−30–6.5 × 10−25]

1.7 × 10 6

5.9 × 10−4

[4.9 × 10−30–2.7 × 10−7]





1.7 × 10−16

[1.3 × 10−32–3.6 × 10−29]

6.0 × 10 5








4.2 × 10−22

[6.6 × 10−29–5.1 × 10−23]

2.0 × 10 6







2.4 × 10−20

[5.8 × 10−31–1.6 × 10−24]

1.7 × 10 10








7.5 × 10−12

[1.05 × 10−29–1.11 × 10−22]

6.9 × 10 4








1.2 × 10−12

[6.6 × 10−30–5.7 × 10−23]

5.7 × 10 4








We detected an effect of hypoxic stress on most traits measured but surprisingly almost no differences between populations were observed either under normal or stressful conditions. However, significantly different reactions norms were observed among populations suggesting genetic differences in response to hypoxia at this level. We also detected sire × treatment interactions that indicate adaptive responses to hypoxia within certain populations. Such genotype × environment interactions may enhance the adaptive potential of populations to hypoxia. Alternatively, lower heritability values observed under stressful conditions may constrain the evolutionary potential of Atlantic salmon populations (Charmantier & Garant, 2005; Hoffmann & Sgro, 2011).

Individuals exposed to hypoxic stress had a higher mortality and delayed hatching compared to embryos reared under normal conditions. Such effects had already been observed in previous studies (Ruggerone, 2000; Ciuhandu et al., 2005; Greig et al., 2005; Finn, 2007). The size at the end of embryonic development under hypoxia was lower in only one case (Oir population in experiment 2) compared to normoxic conditions. The delay at hatching was likely compensated by a higher growth rate leading to a similar size at the end of embryonic development in both treatments (Hamor & Garside, 1976; Roussel, 2007). We also noticed that a slight change in stress intensity (from 4.8 to 4.2 mg L−1 of DO) can dramatically increase the mortality of salmon embryos by 2–3 orders of magnitude.

These effects of hypoxia on early life-history traits were mediated by the population of origin of embryos as reflected by significant population × treatment interactions. Unexpectedly, almost no population differences were noticed both under normoxic and hypoxic conditions, except for survival in normoxia in experiment 2. This last result contrasts with many studies reporting significant differences among populations of various taxa based on measurements of early life-history traits in common garden experiments (e.g. Dingle & Mousseau, 1994; Marangoni & Tejedo, 2008; Gonda et al., 2009; Reardon & Chapman, 2009; Crispo & Chapman, 2010, 2011; Rogell et al., 2011; Aubret, 2012). In salmonids, Haugen & Vollestad (2000) and Jensen et al. (2008) both detected significant embryo length variations among grayling and brown trout populations, respectively. This lack of population effect in our experiment is surprising given the genetic differentiation among the four populations (notably the Allier population vs. the others, Perrier et al., 2011), the high geographic distance between the rivers studied as well as their distinct environmental characteristics. Our results thus suggest a higher plasticity in early life-history traits among Atlantic salmon populations compared to other salmonids. Similarly, Paez et al. (2010) measured S. salar embryonic life-history traits in sub-populations from a Canadian river under controlled conditions and did not detect any significant differences. Later during the juvenile period, however, Nicieza et al. (1994a,b) found significant differences in growth rate and digestion traits for 1-year-old S. salar collected in Spanish and Scottish rivers. We also investigated whether trait variations among both experiments (in normoxia) may be important relative to variations among populations in each experiment. We only detected annual differences for embryo length. These differences were probably linked with significant egg size variations among experiments in the Scorff population and to annual variations in maternal and/or paternal effects for the Saison population (Table S1).

Even though almost no population differences were observed in each treatment, we noticed several significant population × treatment interactions notably in experiment 2 under a more severe hypoxic stress. These population-specific reaction norms may suggest a local adaptation to hypoxic stress. However, we cannot link population-specific responses to their local hypoxia conditions. This is because contrary to environmental variables such as discharge, water temperature and water chemistry, the large scale monitoring of DO levels in gravel river beds is hardly possible. Consequently our results based on a common garden experiment reveal genetic differences in response to hypoxia among populations but further investigations are required to test whether these differences reflect a pattern of local adaptation. Results from experiment 2 suggest that certain populations are more vulnerable to hypoxia. For instance, the Oir population had the most delayed hatching and the most important decrease in size under hypoxic conditions and the Scorff population displayed the largest decrease in survival under such conditions. If one refers to the recent review on reactions norms in salmonids by Hutchings (2011), our study is the first to provide reaction norms for hypoxia in salmonid species. Most population-specific reaction norms have been described for temperature or pH (e.g. Hendry et al., 1998; Haugen & Vollestad, 2000; Fraser et al., 2008; Jensen et al., 2008; Darwish & Hutchings, 2009). In Atlantic salmon, most studies focused on comparisons between wild and farmed populations (e.g. Fraser et al., 2008 or Darwish & Hutchings, 2009). It is also worth notice that some authors did not detect significantly different reaction norms among populations (Kinnison et al., 1998; Morris et al., 2011).

In addition to populations-specific reaction norms, we observed several significant sire × treatment effects that reveal genotype × environment interactions. Heritable plastic responses to hypoxia have been described for gill size in an African cichlid species (Crispo & Chapman, 2010). In salmonid species, heritable reaction norms were so far mainly reported for temperature or pH (Beacham & Murray, 1985, 1986; Hebert et al., 1998; Einum & Fleming, 1999). The fact that genotype × environment interactions were not observed in all populations and traits suggests that this component of genetic variation may contribute to differences in adaptive potential among populations. The evolutionary potential also depends on heritability of traits under standard and stressful conditions, the latter being more critical in the current context of global changes. We found a general tendency for heritability (and coefficients of genetic variation or ‘evolvability’) to decrease under hypoxic conditions, which corresponds to the pattern usually observed in wild animal populations (Réale et al., 1999; Uller et al., 2002; Garant et al., 2003, 2004; Cano et al., 2004; Charmantier et al., 2004; Charmantier & Garant, 2005). We also observed null heritability values for survival but relatively high values for incubation time and embryo size, which follows the general prediction of lower heritability for fitness traits compared to morphological traits (Mousseau & Roff, 1987). Finally maternal effects had a highly significant influence on incubation time and embryo size, in accordance with previous studies in salmonids (e.g. Einum & Fleming, 1999; Aykanat et al., 2012).

To conclude, our results demonstrate some significant differences in response to hypoxic stress among four wild Atlantic salmon populations. However, almost no population difference was observed in normal or stressful conditions contrary to other studies in salmonids. Our data show a high degree of phenotypic plasticity for early life-history traits in genetically differentiated salmon populations. This plasticity may help populations to cope with current global changes but their adaptive potential is probably variable as reflected by distinct heritable reactions norms and different responses of heritability under stressful conditions among populations. Hutchings (2011) recently reviewed the literature on reaction norms in salmonids and described this research topic as being in its infancy, while it can address numerous questions allowing a better understanding of adaptation in a changing environment. In this perspective, we hope this study has demonstrated that adaptation to hypoxia in salmonids represents a promising field of investigation.


We thank the following colleagues from INRA ‘Unité Expérimentale d'Ecologie et d'Ecotoxicologie Aquatique' for their help during experiments: D. Azam, M. Coke, P. Delaunay, A. Gallard, N. Jeannot, B. Joseph, C. Lacoste, F. Marchand, A. Quémeneur, C. Saget and J. Tremblay. We are also grateful to the following people who helped to capture Atlantic salmon or provided environmental data on the rivers studied: A. Baisez, D. Balestin, A. Bardonnet, D. Barracou, E. Bussy, R. Delanoë, P. Etchécopar, P. Gaudin, Y. Guilloux, M. Hoffman, D. Huteau, F. Lange, C. Lousto, Y. Moulia, P. Martin, Y. Moello, E. Prévost, J. Rancon and V. Vauclin. We thank two anonymous referees for their helpful comments. SLC was supported by a grant from Plan Loire Grandeur Nature (project no. 34108) to GE.