Bruno Ernande, Adaptive Dynamics Network, International Institute for Applied Systems Analysis (IIASA), Schlossplatz 1, A-2361 Laxenburg, Austria. Tel.: ++43 2236 807242; fax: ++43 2236 807466; e-mail: firstname.lastname@example.org
We investigated the quantitative genetics of plasticity in resource allocation between survival, growth and reproductive effort in Crassostrea gigas when food abundance varies spatially. Resource allocation shifted from survival to growth and reproductive effort as food abundance increased. An optimality model suggests that this plastic shift may be adaptive. Reproductive effort plasticity and mean survival were highly heritable, whereas for growth, both mean and plasticity had low heritability. The genetic correlations between reproductive effort and both survival and growth were negative in poor treatments, suggesting trade-offs, but positive in rich ones. These sign reversals may reflect genetic variability in resource acquisition, which would only be expressed when food is abundant. Finally, we found positive genetic correlations between reproductive effort plasticity and both growth and survival means. The latter may reflect adaptation of C. gigas to differential sensitivity of fitness to survival, such that genetic variability in survival mean might support genetic variability in reproductive effort plasticity.
Phenotypic plasticity could be an adaptive response to spatially heterogeneous and/or temporally varying environments (Levins, 1963; Bradshaw, 1965; Levins, 1968). A single plastic genotype could produce several phenotypes, each of them being specifically adapted to a different environment. The systematic profile of phenotypes produced by a single genotype along a given range of environmental conditions is then called a reaction norm (Schmalhausen, 1949). As for any biological feature, the evolution of phenotypic plasticity requires that (1) it is adaptive, (2) it is genetically based, and (3) it exhibits sufficient genetic variability in the population considered. The adaptive nature of phenotypic plasticity is assessed by its consequences on fitness or by the plasticity of fitness components themselves. Genetic control and genetic variability are both evaluated by heritability, the proportion of phenotypic variation in the population that originates from genetic variability. Genetic variability for phenotypic plasticity is estimated by the genotype-by-environment interaction (Becker, 1964; Via, 1984; Scheiner & Lyman, 1989), which quantifies the variation of the plastic response to environmental changes among genotypes. Together with fitness, the genotype-by-environment interaction assesses whether some genotypes are ‘specialized’ in certain environments where they have better fitness than others, whereas some genotypes are ‘generalist’ and have high average fitness across environments.
The second issue relates to the environmental effect on genetic correlations between traits. Genetic correlations constrain the evolution of living beings (Partridge & Sibly, 1991; Stearns, 1992). However, they can vary in amplitude across environments (Service & Rose 1985; Newman, 1994; Simons & Roff 1996) and, in rare cases, vary in their sign, implying different constraints in different environments (Gebhardt & Stearns, 1988; Newman, 1988a, b; Leroi et al., 1994a, b). Moreover, sign reversal of genetic correlations may reveal different degrees of genetic specialization in different environments. Genetic correlations giving correlated changes in traits balancing fitness indicate trade-offs and should allow the coexistence of different genotypes with equivalent fitness. A sign reversal of the same genetic correlations in other environments would then suggest some potential for genetic specialization in these environments. In this case, correlated changes in traits affect fitness in the same direction, thus enabling some genotypes to have higher fitness than others.
The third issue originates from the fact that organisms may be integrated in the two dimensions of phenotypic traits: their mean and their plasticity (Scheiner et al., 1991; Schlichting & Pigliucci, 1998). Genetic correlations can be observed between plasticity of different traits or between mean and plasticity, just like genetic correlations between traits in a single environment (Scheiner et al., 1991; Newman, 1994) These could constrain the evolution of phenotypic plasticity. Such correlations might be related to the costs of plasticity supposed to originate from the resource expenses for maintaining and using the physiological machinery needed for plasticity (van Tienderen, 1991; DeWitt et al., 1998).
The above considerations highlight the importance of a multi-trait approach to phenotypic plasticity. Of particular interest when considering correlated traits are traits competing for a limited resource and the strategy of resource allocation between these traits when the amount of the resource varies. Specifically, life-history traits are thought to result from the allocation of some limited resource between three main compartments: maintenance (which affects survival), growth (which affects age/size at maturity and future fecundity) and reproduction (Calow & Sibly, 1990; Partridge & Sibly, 1991; Stearns, 1992; Perrin & Sibly, 1993). Therefore, these compartments should be related by physiological trade-offs, so that any plastic increase of resource allocation in one trait should be correlated with a decrease in the others. In addition, these trade-offs may be reflected through negative genetic correlations between traits within environments. In contrast, positive genetic correlations would reveal potential for genetic specialization despite the physiological trade-offs.
In this article, we present a study on the plasticity of survival, growth and reproductive effort in the Pacific oyster, Crassostrea gigas (Thunberg, 1795). A first quantitative genetics experiment, under controlled conditions, was designed for investigating plasticity in response to food abundance variability in space. However, natural environment may vary in many other characteristics. In order to test results obtained under controlled conditions, we thus conducted a second quantitative genetics experiment in the field. We specifically focus on (i) how the three traits co-varied plastically in response to variability in food abundance by observing bivariate reaction norms, (ii) whether this plastic response was genetically variable by estimating heritability, (iii) whether different degrees of specialization across environments existed by studying the stability of genetic correlations, and (iv) whether any correlative structure existed in the two dimensions of the traits by estimating genetic correlations between trait mean and their plasticity. Finally, we try to evaluate whether the general plastic response is adaptative using a simple modelling approach and interpret our other results in light of life-history theory.
Materials and methods
The studied species, Crassostrea gigas (Thunberg, 1795)
Crassostrea gigas is a marine suspension-feeding bivalve that reproduces between June and September on the European Atlantic coast. Typically, oysters of the Crassostrea genus have protandric alternative sexuality (Coe, 1936; Galtsoff, 1964; Guo et al., 1998), but some individuals can be simultaneous hermaphrodites (Guo et al., 1998). Crassostrea gigas matures between 1 and 3 years old. Adults lay oocytes and sperm in water where fertilization occurs. Egg development leads to a planktotrophic pelagic larval stage that lasts for about 20 days. When larvae reach the pediveliger stage, they settle, typically on rocks or oyster shells in the tidal zone, and then metamorphose into sessile juveniles. Wide larval dispersal leads to heterogeneity in the biotic (food abundance, density of conspecifics, predators…) and abiotic (immersion time, temperature, salinity…) characteristics of the sessile habitat. Phenotypic plasticity is therefore very likely to play an important role in the life history of this species.
Families of oysters were produced in the hatchery at the Laboratoire de Génétique et Pathologie, IFREMER (Ronce-les-Bains, Charente-Maritime, France) according to a nested half-sib mating design. Parent oysters were randomly chosen from the population of the Baie de Marennes-Oléron (Charente-Maritime, France). Each of six sires (males) was mated with four different dams (females) from a total of 24, producing 24 full-sib families nested within six half-sib families. Further details on fertilization and early-life rearing are described in Ernande et al. (2003). One year after fertilization, two experiments were conducted using these families, one under controlled conditions and one in the field. Because of differential survival at early stages (Ernande et al., 2003), only 19 of the 24 full-sib families had enough oysters for the experiments. In order to balance the experimental design, 15 full-sib families nested within five half-sib families were randomly chosen.
Experiment 1: controlled conditions
Two groups of 120 oysters with equal mean individual weights were randomly drawn from each full-sib family and were reared in two different conditions during 6 months (April–October 1999) at the Laboratoire Conchylicole des Pays de Loire, IFREMER (Bouin, Vendée, France). Each of two similar concrete tanks with controlled inflow of seawater was equipped with eight series of six superposed trays distributed along its length, each tray being divided into two niches. The two groups of each full-sib family were randomly assigned to the tanks and then to niches within the tanks. The two treatment conditions differed in the food ration provided to oysters. In one tank, Skeletonema costatum produced in subterranean saltwater (Baud & Bacher, 1990) was added to seawater in order to feed oysters ad libitum (3.32 ± 0.67 × 109 cells oyster−1 day−1), whereas in the other tank, oysters were only fed natural food present in the inflow seawater. We refer to these two treatments as the ‘rich’ and ‘poor’ conditions, respectively. The water column was homogenized by air bubbling and a potential food gradient was avoided by distributing (and collecting) seawater on each side of the tanks through seven inflow (and outflow) cocks regularly distributed along their length (Baud et al., 1997).
Temperature and salinity were measured daily, and inflows of seawater and S. costatum were checked. Oxygen tension was monitored every 3 days, inflow and outflow nutritional fluxes twice a week (concentrations of chlorophyll a and pheo-pigments as indicators of phytoplankton abundance) and turbidity once a week. Paired t-tests (Scherrer, 1984, see Table 1) confirmed a higher availability of phytoplankton in the rich conditions.
Table 1. Hydrological parameters in experiment 1. Concentrations of chlorophyll a and pheo-pigments were significantly higher in rich conditions (Paired t-tests, Scherrer, 1984, Table 1) confirming a higher availability of phytoplankton. In addition, oxygen tension was significantly lower in rich conditions, because of the respiration of Skeletonema costatum. However, the difference was so small (1.1% in average) that it should have not affected oysters.
Poor conditions, value (SE)
Rich conditions, value (SE)
*P < 0.01.
Water inflow (m3 h−1)
Turbidity (g L−1)
Salinity (g L−1)
Oxygen saturation (%)
Chlorophyll a (μg L−1)
Pheo-pigments (μg l−1)
Every 2 weeks, a census of live individuals was taken in each group and 30 individually labelled oysters per group (plastic tag number glued on the shell) were weighed to 0.01 g precision. On the same occasion, groups were randomly assigned to new niches in the tanks in order to avoid any position effect. Thirty individuals per group were randomly drawn out just before spawning (June 1999) for fecundity estimation.
Experiment 2: field conditions
The design was roughly the same as for experiment 1. Differences were as follows. Groups contained 200 individuals instead of 120. The two treatments consisted of a salt marsh pond, with continuous inflow of inshore seawater, and the intertidal zone. Both were located at the CREMA, CNRS-IFREMER (L'Houmeau, Charente-Maritime, France). Salt marsh ponds are characterized by absence of tides and an intense production of phytoplankton, whereas the intertidal zone is characterized by the alternation of tides and a low availability of phytoplankton. Groups were reared in separate plastic net sacks placed on oyster-growing tables during 5 months (May–October 1999). Every 2 weeks, a census of live oysters was taken in each group and 30 randomly chosen oysters per group were weighed to 0.01 g precision. On the same occasion, plastic net sacks were randomly assigned new positions on oyster-growing tables in order to avoid any position effect. Thirty individuals per group were randomly sampled before spawning (July 1999) for fecundity estimation.
Estimation of life-history traits and hypothesis testing
For each experiment, survival probability was estimated as the proportion of individuals alive at the end of the experiment. To correct for the individuals sampled for fecundity estimation, the number of oysters alive at the end of the experiment was divided by the proportion of nonsampled individuals.
Survival probability was analysed as binomial data using logistic regression (McCullagh & Nelder, 1989; PROC GENMOD, SAS/STAT® Software, SAS Institute Inc. 1995). For each experiment, a first logistic regression tested for treatment, sire and dam nested within sire (dam/sire) class effects and for all interactions. Genetic effects were then tested within each treatment using nested logistic regressions with sire and dam/sire class effects.
Growth – experiment 1
Growth was computed for each labelled oyster as the weight increment from the first to the last day of the experiment. However, the dataset was strongly reduced because of high mortality during the experiment (see survival in Table 2), such that we completed it using a reconstruction procedure described in the Appendix. This reconstruction procedure was meant to obtain growth data as consistent as possible with survival data. Survival data comprised all individuals, even those that died during the experiment, whereas growth data before reconstruction only included individuals alive at the end of the experiment. This discrepancy could result in some within-family phenotypic correlations between growth and survival, because of potential size-dependent mortality, and could confound further genetic analyses. We therefore decided to reconstruct growth data using oysters that had survived until spawning, in order to reduce this potential confounding effect, whereas keeping growth data consistent with reproductive effort estimates. After reconstruction, the growth dataset consisted of 293 and 388 individual growth increments in the rich and poor conditions, respectively.
Table 2. Statistical analyses of survival, growth and reproductive effort in experiments 1 and 2.
*P < 0.05, **P < 0.01, ***P < 0.0001.
Treat., treatment. For survival, as data were neither under- nor over-dispersed, hypothesis testing relied on χ2 statistics evaluated as likelihood ratios between nested logistic models. For growth and reproductive effort, hypothesis testing relied on F statistics evaluated as ratios of mean squares in anovas or ancovas. As no cell in the design was empty, mean squares were computed using type III sum of squares. Hypothesis testing for growth in experiment 2 was performed using ancova analysis of weight1/2 regressed against time. In this case, the different interactions time × x tested for the effect of x on the slope of the regression and thus on growth. Here, we only present the statistical tests for these interactions and, for simplicity, refer to them as x instead of time × x.
Growth being size-dependent in bivalve molluscs (Bayne & Newell, 1983), we computed size-independent growth data as the residuals of ln(growth) regressed against weightinitial, using the common regression slope in an ancova (PROC GLM, SAS, Cary, NC, USA) where the regression slopes happened to be homogeneous among treatments, sires and dams/sire.
Size-independent growth was first analysed using an anova (PROC GLM, SAS) with sire and dam/sire as random class effects, treatment as a fixed effect and all interactions. Nested anovas with sire and dam/sire random class effects were then performed to test for genetic effects within each treatment. Data were normally distributed (Shapiro–Wilk test, PROC UNIVARIATE, SAS), but no transformation made them homoscedastic (Bartlett test, PROC GLM, SAS). In order to account for heterogeneity of variance, weighted anovas were then performed.
Growth – experiment 2
Because no individual labelling was made, no individual weight data was available. Therefore, hypothesis testing for growth consisted of ancova of weight1/2 regressed against time with sire and dam/sire as random class effects, treatment as a fixed effect and all interactions (PROC GLM, SAS). As expected from the sampling procedure (see experimental protocol), initial weight did not differ between treatments: the treatment effect and the interactions between treatment and both sire and dam/sire were not significant and thus were dropped from the analysis. Nested ancovas (time fixed effect, sire and dam/sire random class effects and all interactions) were also performed in order to test for genetic effects on growth within each treatment. Weight was square root-transformed in order to achieve normality, but data were heteroscedastic. Thus, weighted ancovas were performed.
A detailed account of the estimation of reproductive effort is given in the Appendix. Here, we briefly describe the three steps of the procedure. First, the fecundity of each oyster sampled before spawning (the number of gamete produced by the individual) was estimated after sex determination and complete stripping of the gonad. Secondly, the dry weight of gametes was estimated as the product between the mean dry weight of a single gamete (spermatozoid or oocyte, accordingly) and fecundity. The third step investigated the relationship between the dry weight of gametes produced by an individual and its total dry weight, as fecundity is known to be size-dependent in bivalve molluscs. We estimated reproductive effort as the size-independent dry weight of gametes, i.e. the residuals of the ln-transformed dry weight of gametes regressed again the total dry weight using the common slope in an ancova where the regression slopes happened to be homogeneous among treatments, sires and dams/sire.
For both experiments, reproductive effort data were normally distributed, but not homoscedastic. They were then analysed in the same way as size-independent growth in experiment 1.
Genetic parameter estimates
Variance components and heritability estimates
For size-independent growth (experiment 1) and reproductive effort (both experiments), observed variance components were estimated by equating the mean square estimates with their expectations in the different anovas used for hypothesis testing (PROC GLM, SAS). In this case, however, the analyses were nonweighted because homoscedasticity is not required for the estimation of variance components (Lynch & Walsh, 1998). Causal variance components were then computed across treatments following Fry (1992) and causal components were computed within each treatment using classical quantitative genetic equations (Lynch & Walsh, 1998). Narrow sense (NSH) and broad sense heritability (BSH) of trait mean (mean) and trait plasticity (pl) were estimated as and , respectively, where and are, the additive and nonadditive genetic variance, and are the additive and non-additive genotype-by-treatment interaction component, and is the total phenotypic variance (Scheiner & Lyman, 1989).
In the absence of individual data for growth in experiment 2, genetic parameters were estimated using numerical resampling combined with bootstrapping, as Windig (1994) did for the quantitative genetics of reaction norm slopes. First, we randomly drew without replacement 30 pairs of initial and final weight per full-sib family per treatment. We then computed ‘individual’ growth increments as the difference in the square root-transformed initial and final weight. Square-root transformation was used for consistency with hypothesis testing on growth (see above). Finally, we used the pseudo-dataset obtained to compute genetic parameters as previously described. We repeated resampling 1000 times and used the mean of the values of interest as estimates. This method should inflate the error variance, but should leave the estimates of the other variance components unbiased.
Survival probability was treated using quantitative genetic methods for threshold characters (Lynch & Walsh, 1998). First, variance components and heritabilities were estimated on the ‘observed’ scale (i.e. 0/1 data) as described before. Then, these estimates were transformed to the ‘liability’ scale (i.e. the continuous scale underlying a threshold trait) according to Robertson's appendix in Dempster & Lerner (1950).
As the datasets were unbalanced because of mortality, no exact estimation of the standard error was available for variance components and heritabilities (Becker, 1984). Therefore, standard errors were estimated as the standard errors of the distributions obtained by bootstrapping data between sire and dam/sire and computing the genetic parameters 1000 times (Manly, 1997). In the case of growth in experiment 2, the bootstrapping procedure was coupled with resampling in order to account for the additional error coming from numerical resampling (Windig, 1994).
It should be noted that, because of the relatively small number of families used for the experiments, the precision in the computation of the variance components was low and this sometimes lead to negative estimates (Lynch & Walsh, 1998). These are explicitly indicated in the Results section when the corresponding effects were found to be significant.
As trait values were collected on different individuals, it was impossible to compute phenotypic correlations, and genetic covariances could not be estimated using individual data. Therefore, additive genetic covariances between traits, cova(x,y), were computed using full-sib family means as data in a nested half-sib mating design (Lynch & Walsh, 1998). Genetic correlations were then computed as ρa(x,y) = cova(x,y)/[σa(x) + σa(y)] and their standard error was estimated by bootstrapping (see above). In case of negative genetic variance estimates preventing the computation of genetic correlations using original data, the bootstrapping means were taken as estimates of the genetic correlations (values given in italics).
For each experiment, two sets of genetic correlations were computed. First, within-treatment genetic correlations, i.e. between different traits expressed in the same treatment, were estimated to test for potential genetically based trade-offs. In this case, genetic correlations were computed combining covariances obtained from full-sib family means and variances obtained from individual data. Secondly, genetic correlations between mean and plasticity of the different traits were computed in order to identify any structure in these two dimensions of the traits. For each trait, the family mean across treatments was computed by pooling family data over all treatments and the family mean plasticity was assessed as the difference in family means of the trait in each treatment (Scheiner & Lyman, 1989). In this case, genetic correlations were computed using covariances and variances obtained from full-sib family means.
Significance level of genetic correlations was obtained by testing whether genetic covariances differed from 0 (Lynch & Walsh, 1998) using randomization tests (Manly, 1997). Trait values were randomized across half-sib families and genetic covariances were computed 1000 times. Significance levels were then determined by comparing the distributions obtained to the original genetic covariances. Finally, the across-treatment stability of the within-treatment genetic correlations was tested by randomizing trait values across treatments, but within families, and computing the difference between the resulting within-treatment genetic correlations 1000 times. Significance levels were then determined by comparing the distributions obtained to the original differences between within-treatment genetic correlations.
Genetic variation in trait means: main family effects
Throughout the description of genetic variation in trait mean and plasticity, we focus on the NSH and refer to the BSH only when it differs strongly from NSH, indicating a strong nonadditive genetic variance.
Significance of the main family effects varied among traits, but was consistent among experiments. In both experiments, survival differed significantly among sires and dams/sire (Table 2A), whereas growth (Table 2B) and reproductive effort (Table 2C) varied significantly among dams/sire only. The genetic variance components of trait means were roughly in agreement with these results. Survival mean had a strong NSH in both experiments, 0.28 and 0.23 (Table 3), whereas growth mean had a very low NSH, 0.04 and 0.03 (Table 3). Negative estimates of nonadditive genetic variance, however, must be noted for survival in both experiments. For reproductive effort mean, the pattern of genetic variance was inconsistent among experiments. In experiment 1, the additive and the nonadditive genetic variance were moderate, leading to a moderate NSH, 0.11 and BSH, 0.14 (Table 3). In contrast, in experiment 2 the additive component was negative, whereas the nonadditive genetic variance was very strong, leading to a negative NSH, −0.08, and a strong BSH, 0.43 (Table 3).
Table 3. Causal variance components and heritability for mean and plasticity of survival, growth and reproductive effort (RE) in experiments 1 and 2.
and refer to additive genetic, nonadditive genetic, treatment, additive genotype-by-treatment interaction, nonadditive genotype-by-treatment interaction, and total phenotypic variances, respectively. NSH and BSH are narrow and broad sense heritability, respectively. Variance components for survival probability are given on the observed scale, whereas heritabilities are given after transformation to the liability scale.
Trait mean heritabilities
NSH ± SE
0.28 ± 0.18
0.04 ± 0.05
0.11 ± 0.13
0.23 ± 0.20
0.03 ± 0.05
−0.08 ± 0.18
BSH ± SE
0.10 ± 0.05
0.06 ± 0.04
0.14 ± 0.08
0.17 ± 0.09
0.08 ± 0.05
0.43 ± 0.23
Trait plasticity heritabilities
NSH ± SE
−0.01 ± 0.05
0.03 ± 0.03
0.58 ± 0.33
−0.02 ± 0.03
0.00 ± 0.02
0.30 ± 0.27
BSH ± SE
0.12 ± 0.06
0.01 ± 0.02
0.27 ± 0.13
0.09 ± 0.04
0.03 ± 0.02
0.35 ± 0.17
Plasticity of traits: bivariate reaction norms, treatment effect and family-by-treatment interactions
In both experiments, the three traits varied significantly among treatments indicating plasticity (Table 2). However, the pattern of plastic co-variation between traits differed between experiments. In experiment 1, survival decreased from the poor to the rich conditions whereas growth and reproductive effort increased (see bivariate reaction norm slopes in Fig. 1a–c), whereas, in experiment 2, the three traits increased altogether from intertidal zone to salt marsh (Fig. 1d–f). The degree of plasticity, i.e. the proportion of phenotypic variability involving plasticity, , was variable among traits. Survival was relatively buffered against environment with a low degree of plasticity, 0.13 and 0.09, compared with growth, 0.77 and 0.54, and reproductive effort, 0.60 and 0.55.
Genetic variability in plasticity also differed among traits. For survival, sire-by-treatment and dam/sire-by-treatment interactions were significant in both experiments (Table 2A). However, the additive and nonadditive genetic interaction variance components were negative and low, respectively, resulting in a negative NSH for survival plasticity, −0.01 and −0.02, and a low BSH, 0.12 and 0.09 (Table 3). For growth, regardless of which experiment, neither the sire-by-treatment nor the dam/sire-by-treatment interaction was significant (Table 2B.), suggesting that growth plasticity was not genetically variable. Indeed, the NSH for growth plasticity appeared to be very weak, 0.03 and 0.00 (Table 3). As expected from this lack of genetic variability in survival and growth plasticity, the bivariate reaction norms of survival with growth were nearly parallel (Fig. 1a, d). In contrast, the plastic response of reproductive effort to treatment was highly variable among half-sib and full-sib families as indicated by the significant sire-by-treatment and dam/sire-by-treatment interactions in both experiments (Table 2C.). Concordantly, in both experiments, the bivariate reaction norms involving reproductive effort did indeed cross, such that families with the highest reproductive effort in the poor conditions and the intertidal zone were those with the lowest in the rich conditions and the salt marsh (Fig. 1b, c, e, f). Despite a negative estimate for the nonadditive genetic interaction component in experiment 1, genetic interaction variance components roughly confirmed these findings (Table 2), NSH for reproductive effort plasticity being very strong indeed, 0.58 and 0.30 (Table 3).
Environmental effect on genetic parameters
The genetic variance components of the three traits varied extensively among treatments (Table 4). Survival displayed substantial significant NSH in every treatment (from 0.17 to 0.43, Table 4). Growth had high significant NSH, 0.60, in the poor conditions and low nonsignificant NSH in the other treatments (from 0.00 to 0.09, Table 4). Finally, apart from a negative estimate in the salt marsh, −0.16, reproductive effort globally exhibited high NSH (from 0.64 to 1.27, Table 4), though nonsignificant in the poor conditions.
Table 4. Heritability estimates in each treatment for survival, growth and reproductive effort in experiments 1 and 2.
*P < 0.05, **P < 0.01, ***P < 0.005.
and refer to additive genetic, nonadditive genetic, error and total phenotypic variance, respectively. NSH and BSH are narrow and broad sense heritability, respectively. For size-independent growth and reproductive effort, significance level of NSH and BSH are based on hypothesis testing for sire and dam/sire effect, respectively, in a nested anova within each treatment. For survival probability, significance levels are based on hypotheses testing in a nested logistic regression within each treatment.
NSH ± SE
0.17 ± 0.16***
0.43 ± 0.26***
0.60 ± 0.39***
0.04 ± 0.24
0.64 ± 0.70
1.27 ± 0.46**
BSH ± SE
0.18 ± 0.09***
0.28 ± 0.16***
0.11 ± 0.17
0.36 ± 0.27***
1.12 ± 0.16***
0.29 ± 0.17***
NSH ± SE
0.16 ± 0.19***
0.27 ± 0.18***
0.00 ± 0.09
0.09 ± 0.16
0.84 ± 0.49*
−0.16 ± 0.44
BSH ± SE
0.33 ± 0.16***
0.17 ± 0.09***
0.15 ± 0.10
0.26 ± 0.16***
0.72 ± 0.29***
1.20 ± 0.52***
Despite the pattern of plastic co-variation between traits differed among experiments (see previous section), a consistent pattern emerged concerning the environmental effect on the within-treatment genetic correlations (Fig. 1). In both experiments, the genetic correlation between survival and growth was stable across treatments (randomization test, P = 0.49 and P = 0.21 for experiment 1 and 2, respectively). It was positive in all treatments (Fig. 1a, d), but only significant in the salt marsh. In contrast, the genetic correlation between survival and reproductive effort varied significantly and changed sign across treatments (randomization test, P < 0.01 and P < 0.01). It was negative in the poor conditions and intertidal zone, though nonsignificant in the latter case, whereas it was significantly positive in the rich conditions and salt marsh (Fig. 1b, e). The genetic correlation between growth and reproductive effort also varied significantly and changed sign across treatments (randomization test, P < 0.05 and P < 0.05). It was negative in the poor conditions and intertidal zone, whereas it was positive in the rich conditions and salt marsh, but significance was only achieved in experiment 2 (Fig. 1c, f).
It should be noted here that most of the significant genetic correlations have an absolute value higher than one. This comes from the fact that genetic correlations are not computed as product–moment correlations and thus are not intrinsically confined between −1 and +1. In case of small mating designs, absolute values may then happen to be higher than 1 (Lynch & Walsh, 1998). In the present study, this has no great implications as we are more interested in the sign of genetic correlations, in order to detect potential genetically expressed trade-offs, than in their absolute value. Also note that genetic correlations involving growth must be considered with caution, as growth had low genetic variance.
Genetic correlations between mean and plasticity of traits
Only a few genetic correlations between mean and plasticity of traits appeared significant (Table 5). This might be explained by the small size of the mating design. However, three genetic correlations were consistently significant across experiments. First, the correlation between growth mean and survival mean was significantly positive (Table 5), confirming the positive within-treatment genetic correlations already observed between the two traits (Fig. 1). In addition, the genetic correlations between reproductive effort plasticity and both growth mean and survival mean were significantly positive, suggesting that the genotypes with the best survival and the fastest growth across treatments were the most plastic ones in terms of reproductive effort. Once again, the genetic correlations involving growth should be considered with caution, as almost no genetic variation was detected for it.
Table 5. Genetic correlations between mean and plasticity of traits.
*P < 0.05, **P < 0.01.
S, G and RE refer to survival, growth and reproductive effort respectively, whereas m and pl refer to the mean and the plasticity value of these traits. Each genetic correlation, together with its bootstrapping standard error, is presented for experiments 1 and 2 on two consecutive rows. When some variance component estimates were negative, thus preventing the estimation of genetic correlations using original data, the bootstrapping means were taken as an estimate (values in italics).
1.05 ± 0.68*
0.23 ± 0.88
−0.62 ± 1.07
−0.49 ± 0.65
1.14 ± 0.51**
1.30 ± 0.67*
−0.46 ± 0.89
−0.79 ± 0.91
1.59 ± 0.48
1.08 ± 0.48*
1.28 ± 0.76
−1.02 ± 0.48
−1.02 ± 0.79
1.42 ± 0.63**
−0.89 ± 0.93
−0.47 ± 1.05
1.48 ± 0.51
1.27 ± 0.42*
−1.07 ± 0.85*
0.09 ± 0.79
0.62 ± 0.83
−0.20 ± 1.18
−0.51 ± 1.25
−0.81 ± 1.03*
0.23 ± 0.98
−0.82 ± 0.93*
−0.79 ± 1.01
−0.48 ± 1.08
−0.53 ± 0.64
1.67 ± 0.73
Survival, growth and reproductive effort in C. gigas responded plastically to spatial variability in food abundance. Under controlled conditions, survival was favoured against growth and reproductive effort when food was scarce, whereas growth and reproductive effort improved at the expense of survival when food was more abundant (treatment effect). This plastic response was genetically polymorphic. Polymorphism mostly originated from strong genetic variability in reproductive effort plasticity (family-by-treatment effect), whereas genetic variability in reproductive effort mean was moderate (main family effect). In contrast, we found substantial heritability for survival mean, whereas it was weak for survival plasticity. Growth roughly displayed no genetic variability in either of the two dimensions. Genetic polymorphism in plasticity was related to a strong environmental effect on genetic parameters. Specifically, the genetic correlations between reproductive effort and both survival and growth were negative with low food abundance and positive with high food abundance. Finally, positive genetic correlations between reproductive effort plasticity and both survival and growth means suggested some structure in the two dimensions of these life history traits. Results obtained in the field experiment were roughly consistent with those found in controlled conditions.
Three potential limitations of our experimental design should be noted. First, because of the infrastructures required for the simultaneous production of oyster families and the necessary amount of technical work, our mating design was relatively small (15 full-sib families nested in five half-sib families). Consequently, the power of some statistical tests and the precision of the estimates of genetic parameters were low. Secondly, we did not distinguish between sexes in our study. Indeed, it would have been impossible to sex the individuals monitored for growth and survival without sacrificing them. Moreover, a preliminary analysis showed no difference in reproductive effort between sexes. Thirdly, we did not perform any multivariate analysis on the three traits, mainly because we are not aware of any multivariate analyses combining both binary (survival) and continuous data (growth and reproductive effort). Nevertheless, despite these limitations, most of the results were clear cut and consistent among experiments.
Resource allocation and physiological constraints on plasticity
The pattern of plastic co-variation between traits observed in experiment 1 (Fig. 1) has two implications. First, the plastic response is constrained by a resource-based physiological trade-off between survival (linked to maintenance) and both growth and reproductive effort. Increasing the former trait did indeed result in decreasing the latter ones and vice versa. Fecundity being size-dependent in C. gigas, this trade-off actually balances survival against both current and future reproduction (the latter through growth). Such physiological trade-off between survival and reproductive traits, revealed by environmental manipulation, is often considered as evidence for reproductive costs (see among many authors Partridge et al., 1987; Kaitala, 1991; Chippindale et al., 1993; Chippindale et al., 1997).
Secondly and more importantly, the plastic response reveals a shift in resource allocation from survival to growth and reproductive effort as food abundance increases. Indeed, if the proportion of resource allocated to the different traits had stayed unchanged, increasing food abundance would have led to an increase in every trait despite the physiological trade-off (Fig. 2a). It follows that the pattern of plastic co-variation between the three traits is not a passive consequence of variation in resource abundance, but an active change in resource allocation strategy that may well be adaptive. Elaborating on Schaffer's (1974) model about resource allocation between adult survival and fecundity, we can indeed show that, if the trade-off between survival and both growth and reproductive effort were concave (from the point of view of the axis origin) and its steepness decreased as resource abundance increased, a shift in resource allocation from survival to both growth and reproductive effort as resource abundance increased would maximize fitness (Fig. 2b). Concavity of the trade-off curve is rather likely as linear or convex trade-offs would always result in the allocation of all the resource to either survival or reproductive effort and growth (Schaffer, 1974). The decrease in the steepness of the trade-off curve as resource abundance increases comes from the simple fact that, whereas resource abundance increases, survival tends toward 1, whereas growth and reproductive effort increase in a much less restricted space. As these two assumptions are realistic, a shift in resource allocation from survival to reproductive effort and growth as resource abundance increases may have been selected for in C. gigas.
Resource allocation and the acquisition of resource
Genetic polymorphism in bivariate reaction norms involving reproductive effort led to a sign reversal in the genetic correlations between reproductive effort and both survival and growth. These were negative in the poor conditions and intertidal zone, again suggesting trade-offs (Fig 1b, c, e, f) and became positive in the rich conditions and salt marsh. Although still rare, such disappearance of genetic evidence for trade-offs between life-history traits across environments has already been observed in a few cases (Gebhardt & Stearns, 1988; Newman, 1988a, b; Leroi et al., 1994a, b).
Sign reversal of genetic correlations reflecting fitness trade-offs are generally interpreted as revealing different degrees of genotypic specialization across environments. However, in case of traits competing for a limited resource, a subtler explanation can be envisaged. If the amount of acquired resource were fixed among genotypes, those investing more in one trait would invest less in others, generating a negative genetic correlation between traits. In contrast, if the resource allocation strategy were fixed among genotypes, genotypes acquiring more resource would have proportionally more resource available for any trait, leading to a positive genetic correlation. Then, if genetic variation in resource acquisition is larger than in resource allocation, one observes a positive genetic correlation despite the trade-off and vice versa (Houle, 1991; de Jong & van Noordwijk, 1992). This theoretical insight suggests that, in our experiments, oyster families differed in their capacity to gather food and that this difference was only expressed in treatments where food was abundant (rich conditions and salt marsh), generating the sign reversals observed. This is consistent with the feeding behaviour of oysters. In suspension-feeders, food acquisition depends on the filtration rate of the individual, but also on the concentration of algae in the water. An increase in the concentration of algae would multiplicatively increase the difference in the amount of resource acquired between individuals with different filtration rates. As a consequence, such a difference should be mainly expressed in cases of high concentrations of algae. It remains to be demonstrated that genotypic differences in resource acquisition actually originate from differences in filtration rate.
The maintenance of genetic variability in resource acquisition, despite the obvious advantage of genotypes with better resource acquisition, is also an issue. It may result from the fact that this variability is not expressed in poor environments. Because of wide larval dispersal and a low capability to assess food abundance before settling, adult sessile oysters encounter environments with varying food availability. Genotypes with lower resource acquisition may then be maintained because, in poor environments, they do not present any disadvantage compared with genotypes with better resource acquisition.
The two dimensions of life history traits: genetic structure and potential implications for the maintenance of genetic polymorphism in plasticity
The positive genetic correlations between reproductive effort plasticity and both survival and growth means (Table 5) suggested some genetic structure between the two dimensions of traits. We know only one study (Newman, 1994) that documented such genetic correlations between mean and plasticity of traits. Such genetic correlations might be related to costs of phenotypic plasticity (van Tienderen, 1991; Newman, 1994; DeWitt et al., 1998), which are expected to be incurred for maintaining and utilizing the physiological machinery needed for plasticity. However, in our experiments an increase in plasticity does not diminish fitness, as mean survival increases with reproductive effort plasticity.
One potential explanation for this relationship then relates to the sensitivity of fitness to survival. Using Schaffer's (1974) model again, it can be shown that the sensitivity of fitness to survival increases with survival itself. This implies that as mean survival improves, the impact of any variation in survival on fitness becomes stronger. As a consequence, genotypes with a high survival mean will need stronger variation in reproductive effort to compensate for variation in fitness because of survival plasticity. This is also consistent with the fact that mean survival is negatively correlated with survival plasticity, although not significantly. Genotypes with high survival mean tend to be more buffered against environmental variability in terms of survival itself and compensate for the increase in fitness sensitivity to survival variation through wider reproductive effort plasticity. The positive genetic correlation between survival mean and reproductive effort plasticity may then result from a genotypic specialization to differential sensitivity to survival. Genetic polymorphism in reproductive effort plasticity may be supported by genetic variability in survival mean. Finally, it should be noted that the families with the highest survival means were those with supposedly better resource acquisition and conversely. Whether maintenance of genetic variability in survival mean is related to genetic variability in resource acquisition is a question to be investigated.
On the need for both environmental manipulation and genetic studies, and for both controlled and field experiments
Our results suggest different trade-off structure according to the approach considered. According to bivariate reaction norms, survival trades off with both growth and reproductive effort, but reproductive effort and growth do not. In contrast, according to genetic correlations, reproductive effort trades off with both growth and survival, but growth and survival do not. The reason for this inconsistency cannot be inferred from our experiments and relates to some debate on the relevance of environmental manipulation and genetic correlations to reveal costs or trade-offs (Partridge, 1992; Reznick, 1992a, b). Yet these results suggest that physiological constraints acting on plasticity and genetic constraints acting on evolution may differ. The relevancy of the two approaches, environmental manipulation or genetic correlations, then depends on the question that one wants to tackle.
The results obtained in controlled and field conditions were roughly similar with respect to genetic parameters. In contrast, the pattern of plastic co-variation between traits observed in field conditions did not support the resource-based physiological trade-offs suggested in controlled conditions. Though the intertidal zone and salt marsh differed according to food abundance, survival, growth and reproductive effort all increased together from intertidal zone to salt marsh pond, i.e. with food abundance (Fig. 1). This discrepancy may arise from the effects of other environmental parameters masking the plasticity of resource allocation. In the intertidal zone oysters faced exposure to air and strong increases in temperature during low tides whereas in the salt marsh oysters were always immersed. These stressful conditions could have generated additional mortality in the intertidal zone counteracting the potential increase in resource allocation to maintenance because of low food abundance (Widdows et al., 1978 J.P. Baud personal communication). These observations emphasize the need for both controlled and field experiments in order to evaluate the respective effect of specific environmental parameters.
The results reported in this paper highlight the importance of a multi-trait and multi-dimension (mean and plasticity of traits) approach to living organisms in order to gain insights into their adaptation to variable environment under genetic and physiological constraints. The potential adaptive nature of the plastic decrease in survival as food abundance increases can only be appreciated knowing that growth and reproductive effort increase at the same time. Genetic differences in resource acquisition can only be inferred by considering genetic correlations between traits sharing a limited resource. The potential support of genetic polymorphism in reproductive effort plasticity by genetic variability in survival mean can only be detected by inspecting genetic correlations between mean and plasticity of traits.
These results give a coherent picture of how C. gigas deals with and may have adapted to spatial variation in food abundance, a critical environmental feature for sessile organisms. However, natural environment also fluctuates temporally, which could affect plastic responses and may explain some unresolved issues such as the maintenance of genetic variability in survival mean and its relationship with genetic polymorphism in reproductive effort plasticity. In a future paper, we will address the plastic response of C. gigas in the face of temporal variability in food abundance.
We would like to thank the technical teams of the Ifremer laboratories at La Tremblade and Bouin for their assistance with the production and maintenance of the animals used in this study in their hatchery and nursery facilities, and the technical team of the Centre de Recherche en Ecologie Marine et Aquaculture for their help during estimation of fecundity and dry weights. Our thanks also to André Gérard, Emmanuel Goyard and Serge Heurtebise for discussion of the experimental design and to Helen McCombie for advice on the English. This work was made possible by a PhD grant from the French Ministère de l'Education et de la Recherche to B.E. and was conducted as part of the European Union Funded project ‘GENEPHYS’ (FAIR PL 95.421).
Reconstruction of weight data in experiment 1
The procedure was restricted to individuals that were still alive at spawning time in order to obtain growth data consistent with reproductive effort estimates, and also to increase consistency with survival data. First, the general shape of individual growth curves was determined using weight data of the oysters that survived until the end of the experiment (PROC NLIN, SAS). The shape was best described by the equation:
This nonlinear model was then individually fitted to the weight data of any labelled oysters that had at least survived until spawning and used to estimate the weight that they would have reached if they had survived until the end of the experiment. The reconstructed weight dataset consisted of 293 and 388 individuals in the rich and poor conditions, respectively, and was then used to compute growth increments.
Estimation of reproductive effort
The estimation of reproductive effort consisted of three steps. The first step was to estimate individual fecundity. For each oyster sampled, the internal tissue (somatic tissue + gonad) and the shell were separated and individually labelled. Tissue was conserved in 4% formaldehyde artificial seawater (distilled water + 30 g L−1 NaCl). Gametes were gathered by stripping of the gonad and sex was checked by observation under the microscope. Gametes were diluted in 60 mL of 4% formaldehyde artificial seawater and the solution was filtrated (vacuum pump + 30 and 200 μm mesh for spermatozoids and oocytes, respectively). After further dilution (to 1/1000 for spermatozoids and 1/150 for oocytes), the size distribution of the particles present in the gamete solution was determined using a Coulter Counter (TA II model) fitted with a 50 or 280 μm probe for spermatozoids and oocytes, respectively. Measurements were made with a coincidence coefficient of particles below 5% and all sizes were expressed as equivalent spherical diameter. Fecundity, defined as the number of gametes produced by the individual, was then estimated using the particle size distribution.
However, calibration was first needed to determine the part of the size distribution that exactly corresponded to gametes. We randomly chose 30 spermatozoid and 30 oocyte solutions for which, in addition to the particle size distribution, the number of gametes was determined by taking a census (cell counter + microscope + Neubauer and Mallassez slides for spermatozoids and oocytes respectively). Fecundity was then determined by census (fecunditycensus) and linearly regressed against fecundity estimated as the number of particles comprised between two arbitrary size boundaries in the particle size distributions of the solutions (fecunditycounter). A numerical iterative procedure was used to identify the boundaries that maximized the coefficient of determination of this regression. These were 1.136–2.303 μm for spermatozoids and 47.469–71.161 μm for oocytes, leading to R2 = 0.97 and R2 = 0.99, respectively. As shown by the regression equations:
for spermatozoids and oocytes, respectively, fecunditycounter predicted fecunditycensus very well. The boundaries obtained by calibration were then used to estimate fecundity from the particle size distribution for each sampled oyster. In the case of hermaphrodites, spermatozoids and oocytes were isolated via differential filtration and numbered separately.
The second step consisted in estimating the dry weight of gametes produced by each individual, Wgametes. This was estimated as the product between fecundity and the mean dry weight of a single gamete (ws and wo for one spermatozoid and one oocyte, respectively):
To estimate ws and wo, we dried 200 random samples of a known number of gametes at 50 °C for 2 days, weighed them to 0.0001 g, divided their weight by the number of gametes, and took the mean of the 200 values obtained. We obtained ws = (1.61 ± 0.05) × 10−10 g and wo = (9.30 ± 0.07) × 10−08 g. In the case of hermaphrodites, the dry weight of spermatozoids and oocytes were summed in order to obtain a single estimate of the dry weight of gametes.
The third step consisted in investigating the relationship between the total weight of an individual and the weight of gametes it produced. The total dry weight, Wtotal, of an individual was computed as the sum of the dry weights of its shell, Wshell, soma, Wsoma and gametes, Wgametes:
Shell was weighed to 0.01 g after drying at 60 °C for 48 h and soma was weighed to 0.01 g after gonad stripping and drying at 50 °C for 72 h. An ancova analysis showed that ln(Wgametes) increased linearly with Wtotal and that the regression slopes were homogeneous among sex, sire, dam/sire and treatments. Therefore, reproductive effort was estimated as the size-independent dry weight of gametes produced by an individual, which was obtained as the residuals of ln(Wgametes) regressed against Wtotal using the common regression slope.