Sex-biased genetic component distribution among populations: additive genetic and maternal contributions to phenotypic differences among populations of Chinook salmon


Daniel D. Heath, Great Lakes Institute for Environmental Research, University of Windsor, 401 Sunset Ave., Windsor, ON, Canada N9B 3P4. Tel.: +1 519 253 3000 (ext 3762); fax: +1 519 971 3616; e-mail:


An approach frequently used to demonstrate a genetic basis for population-level phenotypic differences is to employ common garden rearing designs, where observed differences are assumed to be attributable to primarily additive genetic effects. Here, in two common garden experiments, we employed factorial breeding designs between wild and domestic, and among wild populations of Chinook salmon (Oncorhynchus tshawytscha). We measured the contribution of additive (VA) and maternal (VM) effects to the observed population differences for 17 life history and fitness-related traits. Our results show that, in general, maternal effects contribute more to phenotypic differences among populations than additive genetic effects. These results suggest that maternal effects are important in population phenotypic differentiation and also signify that the inclusion of the maternal source of variation is critical when employing models to test population differences in salmon, such as in local adaptation studies.


Many phenotypic differences among salmon populations are hypothesized to be as a result of adaptation to the local environment, genetic drift, or as a result of nongenetic physiological acclimation to the environment (Taylor, 1991; Adkison, 1995; Fraser et al., 2011). Generally, it is very difficult to distinguish among those possibilities, and it may be that in most cases population differences are due to a combination of those processes. Yet, local adaptation is given special attention in salmon because the presence of local adaptation has serious consequences for conservation and management strategies. Specifically, if differences among populations are as a result of local adaptation, reintroduction or enhancement efforts would have limited success when the introduced fishes’ genomes do not match the local environment (Garcia de Leaniz et al., 2007). Furthermore, if an introduction is successful, and the introduced fish hybridize with the locally adapted fish, the average fitness of the population is expected to decline as a result of outbreeding depression (Gilk et al., 2004).

Demonstrating local adaptation is not straightforward (Kawecki & Ebert, 2004; Garcia de Leaniz et al., 2007; Fraser et al., 2011). In salmon, only a handful of studies have effectively shown local adaptation (e.g. Riddell et al., 1981; Unwin et al., 2007). One precondition for demonstrating local adaptation, which often is not evaluated, is to show that the divergence of fitness traits among populations has an additive genetic basis. Indeed, studies that do attempt to estimate additive genetic variation are usually flawed, in that nonadditive sources (i.e. maternal, dominance and epistatic effects) of variation are often confounded within the estimate, under the assumption that nonadditivity contributes negligibly to the variance structure (reviewed in Heath & Blouw (1998) for maternal effects, see also: Pante et al., 2002; Gallardo et al., 2010). In fact, nonadditive genetic effects may constitute a substantial component of phenotypic variation within and among salmon populations (i.e. Rye & Mao, 1998; Pante et al., 2002; Gilk et al., 2004; Pitcher & Neff, 2006; Roberge et al., 2008; Gallardo et al., 2010; Aykanat et al., 2011). Although local adaptation can develop not only via selection on additive genetic variance, but also through selection on indirect genetic effects, such as heritable genetic influence of maternal effects, such possibilities are seldom explored (Mousseau & Fox, 1998; Wolf et al., 1998; Rasanen & Kruuk 2007). Therefore, the underlying genetic architecture (relative contribution of additive and the various nonadditive genetic variance components) of trait divergence among populations should be carefully investigated when exploring potential local adaptation.

In addition, many common garden experiments designed to evaluate local adaptation employ only one generation of common rearing (e.g. Valdimarsson et al., 2000; Jonsson, 2001; Stewart et al., 2002; Jensen et al., 2008; Kavanagh et al., 2010). Common garden designs are assumed to minimize environmental effects, including maternal effects, and thus differences in mean trait values reflect primarily additive genetic effects (assuming no genotype-by-environment interactions). However, maternal effects and other nongenetic effects may persist for more than one generation of common garden rearing (Roff, 1997; Mousseau, 2000; Richards, 2006), and therefore, estimates of the additive genetic variance component for such traits would be inflated (Rye & Mao, 1998; Pante et al., 2002; Wilson et al., 2005). Furthermore, maternal effects may have heritable genetic basis and can have adaptive significance that persists over generations and therefore should be monitored and characterized in quantitative evolutionary research (Wolf et al., 1998).

Here, we perform two quantitative genetic experiments to partition phenotypic variance into additive genetic (VA) and maternal (VM) components of variance for 17 traits in Chinook salmon (Oncorhynchus tshawytscha), which were held in a common environment from fertilization. We use a modified North Carolina II factorial breeding design with which we explored population-specific maternal (dam) and paternal (sire) contributions to phenotypic differentiation among populations by analysing among-population genetic variance structure. In salmonids, maternal effects are a substantial source of nonadditive variation common in early life history traits within populations (Heath & Blouw, 1998; Heath et al., 1999); however, our results show that maternal effects can also contribute substantially to among-population differences observed in experiments, which employ only a single generation of common garden rearing. Our results highlight the essential role maternal effects may play in among-population trait divergence and that they must be taken into account when designing experiments to test for local adaptation. Furthermore, given their role in salmon population trait divergence, maternal effects should be investigated in more detail for their impact on salmon evolution and population viability through local adaptation.

Materials and methods

Data from two different breeding experiments are used in this study. Hereafter, the two experiments are referred as the YIAL and QRRC experiments after the hatchery where the rearing took place.

The YIAL experiment involved breeding wild and domestic stocks of Chinook salmon (O. tshawytscha) to characterize the genetic architecture underlying phenotypic differences between a natural and a domestic population. Mature wild Chinook salmon from the Big Qualicum River and mature domestic Chinook salmon from Yellow Island Aquaculture Limited (YIAL; Quardra Island, BC) were mated to create 104 families. Breeding design details are described in Bryden et al. (2004), but briefly, one male and one female from each of the two stocks were mated to produce four families in a 2 × 2 factorial design. The cross was replicated 26 times with different parent fish to generate 104 families. Husbandry conditions are described in Bryden et al. (2004). Due to losses during the experiment, the numbers of families used in our analyses ranged from 80 to 94 (Table 1).

Table 1.   A list of the traits measured in this study for both the YIAL (wild vs domestic) and QRRC (four wild populations) experiments. The number of families and individuals within each family are also indicated.
 TraitShort descriptionNo. of familiesMedian per family (range)
YIALFAFluctuating asymmetry index 278 days post-fertilization803 (1–3)
FL-615Fork Length at 615 days post-fertilization8210 (2–18)
W-615Weight at 615 days post-fertilization8210 (2–18)
FL-420Fork Length at 420 days post-fertilization827 (1–28)
W-420Weight at 420 days post-fertilization827 (1–28)
HctHaematocrit count after saltwater challenge, 230–234 days post-fertilization918 (5–11)
[Cl]Plasma Cl concentration (meq L−1) after saltwater challenge. 230–234 days post-fertilization918 (5–11)
Egg survivalEgg survival94200
Eyed egg survivalEyed egg survival94159 (65–195)
Fry survivalFry survival91155 (63–195)
Outbreak survivalNatural vibriosis outbreak survival 520–610 days post-fertilization82143 (61–192)
Relative fecundityRelative fecundity 3 years post-fertilization492 (1–10)
W at 3 yearsW at 3 years (Female only, no gonads)492 (1–10)
W at 4 yearsW at 4 years (Female only, no gonads)532 (1–9)
QRRCEgg survivalEgg survival80470 (284–1110)
Fry survivalFry survival frequency641
FL-210Fork length at 210 days6424 (3–41)

Fourteen traits were measured in the YIAL experiment (Table 1): six were body size traits (wet weight and fork length measures at age 420 and 615 days post-fertilization, female offspring wet weight excluding gonads at 3 and 4 years post-fertilization), two were osmotic stress response traits after saltwater challenge (haematocrit and plasma chloride ion concentration), four were survival measures (egg, eyed egg, fry and natural vibriosis outbreak survival), an adult reproductive trait (relative fecundity at age 3), and fluctuating asymmetry index (FA). Survival measures were coded with each fish represented as an independent binomial data point such that a ‘0’ was assigned for a mortality event and a ‘1’ for the survivors. FA index was calculated using eight bilaterally measured traits, as described in Bryden & Heath (2000). Haematocrit (per cent packed red blood cells × 100) and plasma chloride ion concentration (meq L−1) in response to a 24-h saltwater exposure challenge were calculated as described in Bryden et al. (2004). Conventional salmon aquaculture rearing practices were followed for all fish.

Many of the traits studied here are likely to be important components of fitness and hence good candidates for possible locally adaptive traits. Whereas survival measures are clearly direct measures of fitness, traits such as body size at age and the fluctuating asymmetry index are considered potential proxies for performance or fitness (Clarke, 1995; Garcia de Leaniz et al., 2007). Saltwater tolerance is a vital physiological process for anadromous salmon, and variation associated with it is important for survival (i.e. Kreeger, 1995; Leonard & McCormick, 2001).

The QRRC experiment involved cross-breeding four wild stocks of Chinook salmon (O. tshawytscha) and was designed to partition the genetic variance components underlying phenotypic differences among natural populations. We included this second breeding experiment to provide a comparison to the results of the YIAL experiment. Fish from Harrison River (HR), Quinsam River Hatchery (QN), Big Qualicum (BQ) and Robertson Creek (RC) were crossed to generate pure-type and reciprocal families. On 17th October 2005, eggs and milt were obtained from parental fish at the river of origin and were immediately shipped to the Quesnel River Research Center (QRRC) on ice. Eggs and milt were received on the same day from all four populations (within 24 h of collection), and fertilizations were performed on that day. Milt from one male and eggs from one female from each of four stocks were crossed in a 4 × 4 factorial design to generate 16 families. This breeding design was replicated five times with different individual fish to generate 80 families in total. Fin clip tissues from the parental fish were sampled for later microsatellite genotyping.

Fertilized eggs were incubated in vertical stack incubation trays, each family separated by dividers. At the eyed egg stage (18th November 2005), 200 eggs from 16 families from each 4 × 4 cross were pooled and reared together to minimize tank effects (total 3200 individuals per 4 × 4). Three families had < 200 eggs to survive to the eyed stage: BQxBQ, BQxCH and BQxQN crosses (dam-first notation) had 153, 65 and 182 eggs, respectively. These reduced numbers had little effect on the final rearing density because each group consisted of ≈3200 fish. In Jan 2006, each of the pooled 4 × 4 crosses had reached the first-feeding stage (mean mass = 0.41 g) and was transferred to five outdoor 6 m3 freshwater troughs (flow rate = approximately 8 L s−1). In May 2006, a subsample of fish from all five 4 × 4 crosses (troughs) was sampled and weighed, and fin clips were taken for subsequent DNA analysis for parentage assignment.

Egg survival (to eyed egg stage), fry survival and total length at 210 days data were used as the fitness-related traits for the phenotypic variance partitioning in the QRRC experiment (Table 1). Similar to survival data from YIAL, early egg survival was calculated by binning the total number of fish per family, where each egg provides an independent binomial data point; ‘0’ for a dead egg, and ‘1’ for a surviving egg. However, for fry survival, we were not able to genotype and assign all fry survival in each mixed-family group to family of origin to calculate actual family survival, and instead, we genotyped and assigned a subsample of the fish from each trough. Thus, fry survival is estimated as the occurrence of each family in each trough relative to the total number of fish sampled in that trough. Because each trough is a replicate, we have one survival estimate for each family, replicated four times. We sampled 406, 280, 437 and 339 fish from each 4 × 4 cross. (We have excluded one of the 4 × 4 crosses for fry survival because of poor resolution in the parentage assignment.) Using two or three microsatellite loci, individual fish were assigned to their parental cross (Table S1). There were a few ambiguous genotype combinations that failed to provide positive assignment of individuals to their parental crosses (Table S1). Individuals with ambiguous genotypic combinations were simply excluded from the data set (less than a total of 10 fish from two of the four troughs), and the relative occurrence data for the affected crosses were weighted accordingly.

Statistical analysis

Unless otherwise stated, statistical analyses were performed using R software (R Development Core Team, 2009). In the YIAL experiment, we tested for differences in the mean values for the various traits between the wild and domestic population pure-type crosses using t-tests and nonparametric Kruskal–Wallis test for normally distributed and for non-normally distributed traits, respectively. In the QRRC experiment, pure-type cross differences among the four wild populations were tested using anova and Tukey’s HSD post hoc multiple comparison test, and Kruskal–Wallis test with multiple comparison test for normally distributed and for non-normally distributed traits, respectively.

Phenotypic variance was partitioned using an animal model with maternal effects, which we refer to as the ‘basic model’:


Where zi is the phenotypic value, μ is the mean phenotype, ai is the breeding value (VA), and mi is the maternal effect (VM), and ei is the residual error term (Wilson et al., 2009). To evaluate the relative contribution of population-specific sire and dam effects, we included population sire and population dam as fixed effects to the basic model as follows:


Where Pdi and Psi are the fixed population of origin dam and sire effects of ith individual, respectively. We fit all model parameters using maximum likelihood (ML) with the pedigreemm package in R (Vazquez et al., 2010), which is an extension of lme4 package (Bates et al., 2009) and allows fitting of correlated random effects (i.e. pedigree structures). The pedigreemm package was slightly modified to execute in cases when the number of factors is equal to the number of observations (because each animal is a factor in the animal model) in which the between-animal correlations are defined by the additive relationship matrix (modified code can be found at: We used ML instead of restricted maximum likelihood (REML), because a change in the likelihood estimation in REML does not depend on fixed effect variance structure; hence, it cannot be used to contrast fixed effects. We then compared the basic model to ‘forward stepwise’ models using a likelihood ratio test (LRT) where the log-likelihood ratio statistic is chi-square (χ2) distributed with degrees of freedom equal to the number of factors omitted.

By examining the pattern of changes in the fit of the model with the addition of population of origin fixed effects (e.g. popSire & popDam models), we infer the quantitative genetic architecture of observed phenotypic differences among the pure-type population crosses. For example, significant improvement (< 0.05) in the fit of the model with the inclusion of population of origin fixed effects would indicate that ‘population of origin’ significantly contributed to the observed variation among individuals, and such a model improvement would be detected by an increase in the log-likelihood of the model. More specifically, if the basic model fit is improved by the inclusion of both population-dam (popDam) and population-sire (popSire) effects equally, the phenotypic differentiation among populations for that trait has an additive genetic basis (plus we would predict that the within-population additive genetic variance (VA) estimate will be lower in the full model). If, on the other hand, we were to observe a significant improvement in the popDam model, but not in the popSire model, relative to the basic model, the phenotypic differences among populations are best explained by maternal origin of population. In that case, we would predict that the within-population maternal variance component (VM) will be reduced in popDam model. Similarly, an improvement in the fit resulting from the inclusion of the population-sire factor only would indicate that nonadditive ‘paternal effects’ are the basis of phenotypic divergence.

The total phenotypic variance (VT) within the basic and full models was partitioned into additive genetic (VA) and maternal effect (VM) variance components using an approach similar to that described above, but this time using a REML approach to fit the animal model. Similarly, the significance of the random effect terms (i.e. VA and VM) was evaluated using a model reduction approach in which the model with the exclusion of random terms (one at a time) is compared with the complete model using a LRT. One assumption inherent in the analysis is that the random variance components are homogeneous across populations.

Unfortunately, the lme4 software package does not provide a direct accuracy estimate for mixed model random factors in models with pedigree structure, because REML standard error estimates for random effects can be highly asymmetric and thus may be misleading (Bates, 2010). Hence, we used another freely available software package WOMBAT to obtain standard error estimates for the variance components (Meyer, 2007).

Our design did not allow the partitioning of tank effects (i.e. common environmental effects) that may arise due to full-sib rearing. Full-sib families were reared in separate holding tanks in the YIAL experiments until 210 days post-fertilization (as eggs and larvae to 125 days post-fertilization). Similarly, the full-sib families in the QRRC experiment were reared separately as eggs for 30 days post-fertilization. Therefore, if common environment effects were present, they will be confounded with our variance component estimates, especially for the early life history traits. However, the presence of such common environment effects will have little impact on our analyses and conclusions, because our focus is not the absolute value, but rather the change in the variance components upon including population-specific effects.


In the YIAL experiment, 9 of 14 traits show a difference between pure-type crosses of wild Big Qualicum and domestic YIAL fish (Table 2). All four measures of survival showed differences among pure-type crosses. The wild population has higher early survival, but the disease outbreak data show that the domestic fish have significantly higher survival rates later in life (Table 2). Body size traits (weight and length at 420 days and 615 days post-fertilization) had significant population differences, but the differences gradually reduced with age and become nonsignificant by the age of 4 years post-fertilization (Table 2). Fluctuating asymmetry was also significantly different between the populations, with the domestic line displaying higher FA than the wild population (Table 2). Relative fecundity and saltwater challenge response traits (haematocrit and plasma chloride ion concentration) were not significantly different between the populations (Table 2).

Table 2.   Comparisons of mean pure-type cross traits in the YIAL experiment (domestic vs. wild populations). The standard error of the mean is given in parenthesis. Significant P values (< 0.05) are marked with boldface type. Comparisons made using the nonparametric Kruskal–Wallis test are marked with ‘†’.
Trait (unit)Pure wild crossPure domestic crossP value
FA2.07 (0.18)2.99 (0.32)0.019
FL-615 (cm)  31.8 (0.3)  29.3 (0.4)< 0.001
W-615 (g) 395 (13) 319 (10)< 0.001
FL-420 (cm)  22.4 (0.2)  20.2 (0.3)< 0.001
W-420 (g) 145 (3.9) 111 (4.6)< 0.001
Hct (100×% RBC content)  44.1 (0.6)  44.2 (0.5)0.573
[Cl] (meq L−1) 155 (2) 161 (2)0.069
Egg survival (%)0.84 (0.03)0.64 (0.03)< 0.001
Eyed egg survival (%)0.99 (0)0.93 (0.02)< 0.001
Fry survival (%)0.99 (0)0.97(0.01)0.021
Outbreak survival (%)0.83 (0.01)0.92 (0.01)< 0.001
Relative fecundity (%)1224 (70)1059 (57)0.34
W at 3 years (kg)1.95 (0.14)1.62 (0.06)0.056
W at 4 years (kg)1.88 (0.10)1.82 (0.07)0.514

In the YIAL experiment, when population-dam effects are included (popDam model), the model likelihood significantly increased for 11 of 14 traits, compared with five of the 14 traits where the likelihood increased due to the inclusion of population-specific sire effects (popSire model; Fig. 1, Table S2). Because a greater improvement in model fit for the popDam over the popSire model relative to the basic model is indicative of maternal effect contribution to among-population divergence, our results indicate that maternal effects likely contribute substantially to the domestic vs. wild population divergence. More specifically, although both popSire and popDam models performed better than the basic model for FL-615, W-615, FL-420 and W 420 and outbreak survival traits, in all cases, popDam model improved the likelihood more than the sire model (Fig. 1). Based on our predictions, these results suggest that divergence in those traits is driven by both additive genetic and maternal effect components. On the other hand, FA, egg survival, eyed egg survival, fry survival, W at 4 years and W at 3 years were only improved by popDam and not the popSire model, suggesting that phenotypic divergence between the two populations for those traits is due to primarily maternal effect contributions (Fig. 1, Table S2). It is important to note, however, that among those traits, W at 3 years showed only marginal significance, and W at 4 years divergence was not significantly different among populations (= 0.056 and = 0.514, respectively, Table 2).

Figure 1.

 A scatterplot of negative log-likelihood ratio estimates for population-specific sire and dam models relative to the basic model (which does not incorporate population effects) for 17 traits measured in the YIAL (14 traits) and QRRC (three traits) experiments. Equal log-likelihood ratios (slope = 0) indicate primarily additive genetic effects associated with population trait divergence, whereas higher population-dam effects relative to population-sire effects (positive slope) indicate among-population maternal effects contributing to population divergence for the trait. The shaded region denotes models that do not represent a significant improvement over the basic model (< 0.05), such that the log-likelihood values in the shaded region of the graph do not show population-dam and/or population-sire effects that are contributing to phenotypic differences among populations. Boldface type for the trait abbreviation (right-hand margin) indicates that the trait significantly differs among populations (pure-type crosses).

In the QRRC experiment, popSire and popDam models improved the basic model at FL 210 days similarly, suggesting an additive genetic basis for population divergence (Fig. 1, Table S2). For egg survival, the popDam model performed significantly better than the basic model, suggesting that only maternal effects influenced population divergence for that trait (Fig. 1, Table S2). In fry survival, both models significantly improved the likelihood over the basic model (Fig. 1, Table S2), but little phenotypic divergence was observed in the among pure-type comparison (Table 3).

Table 3.   The effect of population on the mean trait values in the QRRC experiment pure-type crosses. The standard error of the mean is given in parenthesis. Significant post hoc comparison differences are marked with different letters (< 0.05). Egg survival data distribution was non-normal and comparison made with nonparametric Kruskal–Wallis test.
Egg survival0.81(0.06)b0.98(0.01)a0.78(0.12)b0.96(0.01)a< 0.001
Fry survival7.20(0.36)7.16 (0.86)6.36 (0.21)5.28 (0.98)0.22
FL-2107.34 (0.05)a7.15 (0.15)ab6.95 (0.04)c6.98(0.22)bc< 0.002

The total variance explained by random effects (VT = VA + VM + VR) for traits measured in the QRRC and YIAL experiments was reduced by a maximum of 24% (to 0.76 VT ratio) in the full model relative to the basic model (Table 4, last column). For most traits, the decrease in variance from the basic to full model (indicative of population effects because the full model included the population effect) was driven by a decrease in the maternal effect variation (VM; Table 4, Fig. 2).

Table 4.   Variance structure and standard errors (SE) for 18 traits measured in both the YIAL and QRRC breeding experiments for the full and basic models, where the full model includes population-specific sire and dam effects as fixed factors. Significant variance components (VA and VM) are indicated in boldface type (< 0.05) and are also given as percentages of VT (the total phenotypic variation explained by random variation = VA + VM + Vresidual). ‘VT ratio’ is ratio of random variances for the full and basic models (i.e. VT-Full/ VT-Basic).
 TraitFull ModelBasic ModelVT ratio
  1. *SE estimation failed for relative fecundity VA values.

YIALFA0 (0.73)0 (19)0.11 (0.32)3 (9)3.780 (0.70)0 (18)0.31 (0.32)8 (8)3.950.96
FL-6152.01 (1.21)25 (15)0.66 (0.63)8 (8)8.051.76 (1.02)19 (11)1.81 (0.71)20 (7)9.100.89
W-6152110 (1377)20 (13)944 (746)9 (7)104912268 (1308)20 (11)1640 (789)15 (7)113090.93
FL-4201.16 (0.49)57 (22)0 (0.21)0 (10)2.041.36 (0.55)51 (19)0.52 (0.30)20 (10)2.670.76
W-420257 (131)36 (18)53 (64)7 (9)718326 (145)38 (16)166 (85)19 (9)8690.83
Hct6.08 (2.76)31 (14)0.76 (1.24)4 (6)19.485.72 (2.61)29 (13)1.01(1.20)5 (6)19.531
[Cl]19 (24)6 (8)0 (11)0 (4)302.218 (23)6 (8)0 (10)0 (4)301.41
Egg survival0.05 (0.01)29 (7)0 (0.01)0 (3)0.1760.06 (0.01)30 (8)0.01 (0.01)3 (3)0.1830.96
Eyed egg survival3 × 10−4 (3 × 10−4)1 (1)2.7 × 10−3 (6 × 10−4)7 (2)0.03793 × 10−4 (3 × 10−4)1 (1)3.4 × 10−3 (7 × 10−4)9 (2)0.03860.98
Fry survival4.4 × 10−4 (2.6 × 10−4)2 (1)5.2 × 10−4 (1.8 × 10−4)3 (1)0.01943.9 × 10−4 (2.4 × 10−4)2 (1)6.5 × 10−4 (2.0 × 10−4)3 (1)0.01951
Outbreak4.8 × 10−3 (1.9 × 10−3)4 (2)0 (7.9 × 10−4)0 (1)0.1136.5 × 10−3 (2.5 × 10−3)6 (2)1 × 10−4 (1 × 10−3)0 (1)0.1140.99
Relative fecundity*21847 (NA)43 (NA)0 (7190)0 (14)5117019309 (NA)39 (NA)0 (6201)0 (13)498991.03
W at 3 years0.05 (0.06)47 (55)0 (0.03)0 (24)0.1090.07 (0.05)55 (43)0 (0.02)0 (19)0.1240.88
W at 4 years0.31 (0.59)26 (48)0 (0.26)0 (21)1.2140.32 (0.53)26 (43)0 (0.24)0 (19)1.250.97
QRRCEgg survival0.003 (0.002)4 (2)0.019 (0.005)25 (6)0.0760.003 (0.002)3 (2)0.023 (0.006)29 (5)0.0800.95
Fry survival2.20 (1.70)74 (52)0 (0.61)0 (21)2.962.52 (1.63)74 (45)0.28 (0.62)8 (19)3.380.88
FL-2100.081 (0.047)18 (10)0 (0.017)0 (4)0.4510.201 (0.083)39 (15)0 (0.030)0 (6)0.5110.88
Figure 2.

 A visual representation of differences in the genetic components of variation (%) between the basic and full models for all 17 traits measured in both experiments. YIAL and QRRC traits are denoted by filled and empty circles, respectively. The dotted lines indicate the y = 0 axis, the x = 0 axis and the x = y 1 : 1 relationship. Positive x and y values indicate the reduction in VA and VM (respectively) associated with the inclusion of population effects into the model (i.e. full model). This reduction reflects the VA and VM contributions to population differentiation, and the x = y line denotes equal magnitude of ΔVA and ΔVM for each trait.


We found significant phenotypic differences among the study populations held in common environments for a number of traits that are expected to contribute to fitness and performance in natural populations of salmon. Curiously, we found few examples of traits showing population trait divergence and for which the sire-population (popSire model) variance explained was greater than the dam-population (popDam model). This indicates that among-population additive genetic variance contributes relatively little to population differentiation, despite sizeable population mean trait differences in some cases. This is especially notable for YIAL body size at age traits and for early survival traits. Instead, differentiation between the wild and domestic populations is driven by population-dam (maternal) effects. On the other hand, body size variation among the populations in the QRRC experiment has more of an additive genetic basis, suggesting that additive genetic effects do play a role in population differences for some traits, and for some populations, or that the nature of the genetic architecture differs between domestic–wild vs. wild–wild comparisons.

Separating genetic and environmental sources of maternal effects requires multiple generations of breeding experiments and phenotype data, which we did not have here. However, the substantial role of maternal effects in population differentiation, especially for body size at older ages (i.e. weight and length at 420 and 615 days post-fertilization), is surprising. Maternal effects are known to be influential in early life history traits in Chinook salmon but are expected to erode 1-year post-fertilization (Heath & Blouw, 1998; Heath et al., 1999). Our maternal effects variance components reflect this pattern, with no measurable within-population VM estimates for body size traits later in life (under the assumption that a lack of VM in the full model is also indicative of a lack of within population VM; Table 4). Thus, it may be that the among-population maternal effects we demonstrate (as inferred through model comparisons) are not simply maternal environment effects, but rather include more profound factors, such as indirect genetic effects of maternal influence (Wolf et al., 1998; Heath et al., 2003; Houde et al., 2011) or other genetic maternal effects (i.e. genetic imprinting, sex-linkage, mitochondrial inheritance; Perry et al., 2005). However, such effects have been rarely reported in salmon (i.e. Perry et al., 2005; Houde et al., 2011) and are poorly understood, especially when considering among-population effects, rather than the more traditional within-population maternal effects. On the other hand, although environmental maternal effects are not inherited across generations, theoretical and empirical studies indicate that such effects can have dramatic effects on short-term evolutionary responses to selection (Riska, 1989; Cheverud & Moore, 1994; Kawecki & Ebert, 2004), and we see no reason to exclude such possibilities at the among-population level.

One important implication of our study is that common garden experiments designed to test for genetic contributions to population differences that employ less than one generation of common environment rearing overestimate the additive genetic basis of differentiation among populations. This also holds true for translocation studies, where first-generation differences in the traits of interest may not reflect longer-term differences or additive genetic variance. Unfortunately, common garden experiments lasting less than two generations are common in local adaptation research (Garcia de Leaniz et al., 2007; Fraser et al., 2011), and such studies of the genetic basis for local adaptation are biased towards finding an additive genetic basis for trait differentiation among populations. Either multiple generations of common garden rearing or half-sib breeding designs (or ideally, both) should be used to show that differences do reflect the additive genetic variance necessary for local adaptation arguments. Studies such as ours that use more detailed quantitative genetic breeding designs (compared with full-sib breeding) allow the partitioning of the variance components contributing to traits that show potential for local adaptation.


We would like to thank John Heath, Ann Heath, Rick Holmes, Bill Best, Richard Rujanschi, Melinda Shaw and Sarah L. Roberts for help in the field. The experiments were conducted in the facilities of Yellow Island Aquaculture Ltd. (YIAL) and Quesnel River Research Center (QRRC) of University of Northern British Columbia (UNBC). Funding for this project was provided by YIAL, as well as the Natural Science and Engineering Research Council of Canada (to DDH). The authors have no conflict of interest associated with this work.

Data deposited at Dryad: doi: 10.5061/dryad.26mv3f6d