Genetic polymorphism and trade-offs in the early life-history strategy of the Pacific oyster, Crassostrea gigas (Thunberg, 1795): a quantitative genetic study


Correspondence and present address: Bruno Ernande, Adaptive Dynamics Network, International Institute for Applied Systems Analysis, A-2361 Laxenburg, Austria. Tel.: +43 2236 807 242; fax: +43 2236 807 466; e-mail:


We investigated genetic variability and genetic correlations in early life-history traits of Crassostrea gigas. Larval survival, larval development rate, size at settlement and metamorphosis success were found to be substantially heritable, whereas larval growth rate and juvenile traits were not. We identified a strong positive genetic correlation between larval development rate and size at settlement, and argue that selection could optimize both age and size at settlement. However, trade-offs, resulting in costs of metamorphosing early and large, were suggested by negative genetic correlations or covariances between larval development rate/size at settlement and both metamorphosis success and juvenile survival. Moreover, size advantage at settlement disappeared with time during the juvenile stage. Finally, we observed no genetic correlations between larval and juvenile stages, implying genetic independence of life-history traits between life-stages. We suggest two possible scenarios for the maintenance of genetic polymorphism in the early life-history strategy of C. gigas.


Most marine invertebrates have ‘complex’ life cycles that are divided into discrete successive stages (Strathmann, 1993; Giangrande et al., 1994). Two nonmutually exclusive theories are invoked for the evolutionary emergence of complex life cycles. The first one presents them as an adaptive decoupling between specializations at different stages allowing fitness trade-offs to be broken (Moran, 1994). For instance, in marine invertebrates with pelagic larvae and benthic adults, the larval stage could result from an adaptation to dispersal and habitat selection, whereas the adult stage could represent specialization to growth and reproduction (Wray & Raff, 1991; Pechenik, 1999). In this scenario, complex life cycles are then expected to prevent any genetic correlation between life-history traits across stages (Moran, 1994). The second theory interprets complex life cycles as an adaptive size-specific shift in ecological niche through metamorphosis (Wilbur, 1980). Then, at small sizes, a planktonic larva may be advantageous over a benthic form in terms of maximizing growth and survival (Strathmann, 1993). This theory gives no insights about the relationship between life-history traits at different stages but stresses that size at metamorphosis is a critical trait for species with complex life cycles.

These two general theories are meant to provide unifying hypotheses for the emergence of any type of complex life cycle. Marine invertebrates provide examples of diverse complex life cycles (Giangrande et al., 1994), although their specific characteristics demand more detailed descriptions. In this paper, we focus on marine benthic invertebrates with a pelagic larval stage. Here, it is known that age and size at metamorphosis are pivotal for fitness (Wray & Raff, 1991; Strathmann, 1993; Pechenik, 1999), as in many other taxa (Wilbur & Collins, 1973; Wilbur, 1980; Werner & Gilliam, 1984; Forrest, 1987; Ebenman, 1992; Moran, 1994), and should therefore be under strong selective pressure. Indeed, an early age at metamorphosis supposedly favours larval survival by decreasing larval predation and hazard, lengthens adult reproductive life span and decreases age at maturity. In addition, size at metamorphosis should increase with larval growth rate relative to adult growth rate and increased size should improve post-metamorphic survival (if positively body size-dependent) and enhance size at maturity (Wilbur, 1980; Werner & Gilliam, 1984). However, it might be difficult to optimize both age and size at metamorphosis simultaneously. Three different scenarios are possible. First, if larval growth rate is constant, there is an inherent fitness trade-off (positive correlation) between age and size at metamorphosis since early metamorphosis inevitably leads to a small size whereas late metamorphosis entails large size. Second, if larval growth rate varies, some authors (Wilbur & Collins, 1973; Werner, 1986) argue that only one of the two traits could be optimized depending on the growth rate. This would be size when growth rate is high and age when it is low, again generating a positive correlation between age and size at metamorphosis. A third possibility, however, is that, in case of time constraints such as seasonal reproduction or resource availability, fast growing individuals could optimize both traits (metamorphose early and at a large size) whereas slow growing individuals would perform badly in terms of the two traits, leading to negatively correlated age and size at metamorphosis (Rowe & Ludwig, 1991; Abrams et al., 1996). An additional degree of complexity comes from the fact that metamorphosis itself might be a critical period in terms of survival, as it is in marine bivalve molluscs (Roegner, 1991). Although this aspect is generally not well appreciated because of the difficulty of experimental studies on metamorphosis and the estimation of related mortality, the probability of success at metamorphosis could affect the age and size at which it occurs. To summarize, optimal age and size at metamorphosis are expected to result from their mutual interactions as well as those with pre- and post-metamorphic traits (essentially growth rate and survival) and the success at metamorphosis. A clear picture of the strategy for age and size at metamorphosis can then only emerge from studies on the entire early life of the species considered and specifically on the correlation structure between the different early life-history traits.

Most of the scenarios presented above do not distinguish between genetically fixed and plastic (or condition- dependent) strategies for age and size at metamorphosis (Werner & Gilliam, 1984). Numerous experimental studies have documented environmental causes of intraspecific (co)variation in metamorphic traits (age and size at metamorphosis or related traits) in amphibians (e.g. Berven & Gill, 1983; Pfennig et al., 1991; Newman, 1994; Morey & Reznick, 2001), insects (e.g. Via, 1984a; Forrest, 1987; Karlsson, 1994; Stevens et al., 2000) and marine benthic invertebrates (e.g. Pechenik, 1984; Hadfield & Strathmann, 1996; Pechenik et al., 1996a,b). In contrast, potential genetic bases for intraspecific (co)variation have seldom been studied in amphibians (Berven, 1987; Travis et al., 1987; Newman, 1988; Semlistsch, 1993) or insects (Robertson, 1960, 1963; Via, 1984a,b; Gebhardt & Stearns, 1988; Moller et al., 1989; Zwaan et al., 1995; Nunney, 1996; Tucic et al., 1998), and we are not aware of any quantitative genetics studies on metamorphic traits in any marine benthic invertebrates.

In this paper, we present a quantitative genetics experiment focusing on the early life of the Pacific oyster, Crassostrea gigas (Thunberg, 1795). This marine bivalve is characterized by a planktotrophic pelagic larval stage capable of wide dispersal followed by a sessile adult stage. Metamorphosis directly follows settlement, that is, definitive attachment of larvae to a substratum, which determines habitat selection for the adult stage. Using a nested half-sib mating design, we investigate the genetic bases for variation in pre- and post-metamorphic growth and survival, as well as in metamorphic traits themselves – larval development rate that determines age at settlement, size and success at settlement and subsequent metamorphosis. We then focus on the genetic correlation structure between these traits. Specifically, we test for the absence of genetic correlations between traits at different stages as well as for the presence of trade-offs between metamorphic traits themselves and between metamorphic traits and larval/juvenile traits. Finally, we discuss the different results in the light of life-history theory.

Materials and methods

The Pacific oyster, Crassostrea gigas

Crassostrea gigas is a marine bivalve mollusc originating from the Asian coasts of the Pacific ocean. On the French Atlantic coast, C. gigas reproduces in summer. Mature adults lay oocytes and sperm in water where fertilization occurs. Fertilization is followed by a planktotrophic pelagic larval stage that lasts for about 20 days. Larvae undergo differentiation characterized by successive morphological states: ‘D’, veliger, ‘eyed’ and pediveliger larvae ( Quayle, 1988 ). In the pediveliger state, larvae acquire the competence for settlement that can be delayed for a few days ( Tamburri et al., 1992 ; Turner et al., 1994 ) in order to find a convenient substratum. Metamorphosis directly follows settlement and leads to a sessile suspension-feeding adult stage. At this point, remodelling of pre-existing organs occurs as well as the differentiation of new ones. Moreover, metamorphosing individuals cannot feed and have to rely on stored energetic reserves ( Bartlett, 1979 ). Metamorphosis is, therefore, a critical period in terms of survival ( Roegner, 1991 ).

Parental population

One hundred oysters originating from the Baie de Marennes-Oléron (Charente Maritime, France) were randomly chosen as the parental population. They were brought to sexual maturation in laboratory raceways with a flow-through system. Common standard maturation conditions (Robert & Gérard, 1999) ensured sufficient fecundity to perform the crosses. After 2 months, 6 sires (males) and 24 dams (females) were randomly chosen to perform crosses in the hatchery of the Laboratoire de Génétique et Pathologie, IFREMER (Ronce-les-Bains, Charente Maritime, France).

Crosses and larval rearing

Families were produced according to a nested half-sib mating design: each of the 6 sires was mated with 4 different dams from the total 24, producing 6 half-sibs families and 24 full-sibs families. 1 500 000 oocytes per dam were fertilized at a ratio of 200 spermatozoids per oocyte following the procedure described in Collet et al. (1999). Three hours after fertilization, the percentage of developed embryos (=fertilization success) was determined in 2 samples per cross as described in Gérard et al. (1994).

Full-sib families of larvae were reared in 24 glass-reinforced polyester (GRP) 30-L tanks filled with 1-μm filtered seawater (temperature 20.95 ± 0.95 °C and salinity 30.6 ± 1.1‰) and were fed Isochrysis galbana and Extobocellulus criberiger. The amount of algae provided, 60 cells μL−1 day−1, was in excess to avoid competition for food (Nascimento, 1983). To further limit competition, larval concentration was progressively reduced from 50 larvae mL−1 at day 1 after fertilization down to 7 larvae mL−1 at day 8, by discarding a randomly chosen part of the full-sib family.

Every 48 h, the larvae were collected in 1-L beakers by sieving. Densities were then estimated by counting (cell counter + microscope) larvae in 3 water samples per full-sib family. These estimates were multiplied by the backward multiplicative proportion of retained larvae in order to correct for the discarded parts of each full-sib family. At the same time, 30 larvae per full-sib family were measured for shell length (Profile projector, Nikon). Due to technical difficulties, only 21 of the 24 full-sib families successfully reached the end of the larval period.

Settlement procedure and micro-nursing

From the day when the first pediveliger larvae were observed in a full-sib family, indicating competence for settlement, 4 water samples were counted daily to estimate their proportion. When this proportion exceeded 10%, these larvae were put to settle. Because pediveliger larvae are larger than others, we selected them by sieving on a 200-μm mesh (Robert & Gérard, 1999). To assess the developmental state of the selected larvae, 4 water samples were examined by microscope. For every full-sib family, 100% of larvae selected were pediveliger. The larvae in the water samples were then counted to estimate density and 30 individuals were measured for size at settlement. The larvae selected were put to settle in a 200-μm mesh-bottomed tray with ground oyster shell. The remaining larvae were kept in the larval rearing tank. The whole procedure was repeated 2 days later to create a second settlement cohort and the unselected larvae were this time discarded. Each full-sib family, therefore, had two settlement cohorts.

To assess the accuracy of our method for isolating pediveliger larvae, we compared the density estimates of pediveliger larvae before and after sieving. We described these density estimates as Poisson data using the following log-linear model: ln(density) = sieving + sire + dam/sire + sieving × sire + sieving × dam/sire (McCullagh & Nelder, 1989; PROC GENMOD, SAS Software, SAS Institute, 1995). The sieving effect and its interactions with sire and dam nested within sire (dam/sire) were not significant (sieving: inline image = 1.44, P = 0.2308; sieving × sire: inline image = 5.94, P = 0.3124; sieving × dam/sire: inline image = 17.93, P = 0.2665), indicating that, in each full sib-family, almost all pediveliger larvae present were selected by sieving.

Each settlement cohort was randomly distributed into raceways for micro-nursing, with flow-through 20-μm filtered seawater enriched with phytoplankton (Robert & Gérard, 1999). Mesh size was changed regularly following juvenile growth, and trays were washed and moved at random daily. To limit potential micro-environmental effects, the 2 settlement cohorts were grouped together after 10 days. When necessary, the densities were reduced down to 30 000 individuals per full-sib family to limit competition.

For each cohort, three estimates of the density of juveniles were computed 7 days after settlement (before grouping the cohorts) by dividing the total weight of the cohort considered by 3 estimates of the mean weight of an individual in that cohort. The 3 estimates of the mean weight of an individual were obtained by taking three samples of a random number of individuals, weighing each sample and dividing their weight by the number of individuals in each of them. In addition, 30 individuals per full-sib family were measured for shell length at days 39 and 61. For technical reasons, only 19 of 21 full-sib families were successful in settlement and metamorphosis.

Juvenile rearing (nursing)

At day 70, juveniles were transferred to the Laboratoire Côtier, IFREMER (Bouin, Vendée, France). Full-sib families were reared until day 185 in 50-cm diameter mesh-bottomed tubs in a single concrete tank with a seawater upwelling system (Bacher & Baud, 1992). Skeletonema costatum was provided to oysters ad libidum in order to prevent food competition. The density of juveniles was estimated every 2 weeks, following the same method as during micro-nursing. The density was gradually reduced down from 30 000 individuals at day 70 to 3000 at day 185 and estimates were corrected backwards as above. Thirty individuals per full-sib family were weighed at days 98, 138 and 185. At day 98, an allometric relationship between length and weight was determined using 100 randomly chosen individuals among the 19 full-sib families in order to convert length measures taken during micro-nursing into weights according to the following equation:


where r2 = 0.96, P < 0.0001.

Data analyses

The different variables monitored during the experiment are summarized in Fig. 1.

Figure 1.

Summary of the different life-stages and variables monitored for each full-sib family. 1st and 2nd correspond to the first and second settlement cohorts. x is the date of settlement of the first cohort, i.e. the date at which more than 10% of the larvae were at the pediveliger state in the full-sib family considered. There were two settlement groups of full-sib families, i.e., x  = 24 or 27 given the full-sib family considered (see Results). Frequencies of measurements during the juvenile stage are approximate (see Fig. 3 for precise dates).

Statistical inferences for quantitative traits

We treated quantitative traits – larval shell length (=larval length) and juvenile weight (Fig. 1) – with classical linear models (PROC GLM, SAS). We analysed these two traits using ancovas with time or functions of time as co-variables, in order to test for genetic variation in larval growth rate and juvenile growth rate, respectively (Table 1). We also tested for genetic variability in larval length at settlement (=size at settlement), juvenile weight at day 39 (=weight after metamorphosis) and juvenile weight at day 185 (=final weight) using nested anovas (Fig. 1, Table 1).

Table 1.  Statistical analyses and models used.
  1. Variables were transformed before analysis using the function f. Quantitative traits: transformation of variables was intended to achieve normality (Shapiro–Wilk test, PROC UNIVARIATE, SAS) and homoscedasticity (Bartlett test, PROC GLM, SAS). However, no transformation made larval length homoscedastic. Therefore, we performed weighted analyses in order to account for heterogeneity of variance. Hypothesis testing was made using the following mean squares ratios: continuous versus continuous × sire, sire versus dam/sire, dam/sire versus errors, continuous × sire versus continuous × dam/sire and continuous × dam/sire versus errors (Dagnelie, 1998), the continuous effect being timex or lnx (time), with x from 0 to 3. Type III sums of squares were used to compute mean squares since data were unbalanced. Binary data: transformation was made in accordance with generalized linear model assumptions: ‘ln’ for Poisson data and ‘logit’ for binomial data (see McCullagh & Nelder, 1989 for details). Data were overdispersed for larval and juvenile densities (deviance/d.f. = 2.12 and 2.33, respectively). In these cases, we performed scaled analysis to correct for overdispersion (PROC GENMOD, SAS; McCullagh & Nelder, 1989). Hypothesis testing relied on likelihood ratios, which asymptotically follow a χ2 distribution. In the case of scaled analyses, F statistics were also produced to assess significance of the different effects. As results were the same for χ2 and F statistics, only χ2 statistics are presented in the Results.

Quantitative traits   
 Larval lengthWeighted ancovalnTime + sire + dam/sire + time × sire + time × dam/sire
 Juvenile weightancovalnln2 (time) + sire + dam/sire + ln2 (time) × sire + ln2 (time) × dam/sire
 Size at settlementanovalnSire + dam/sire
 Weight after metamorphosisanovalnSire + dam/sire
 Final weightanovalnSire + dam/sire
Binary data   
 Larval densityLog-linear modellnln(time) + ln2 (time) + ln3 (time) + sire + dam/sire + ln(time) × sire +  ln(time) × dam/sire + ln2 (time) × sire + ln2 (time) × dam/sire +  ln3 (time) × sire + ln3 (time) × dam/sire
 Juvenile densityLog-linear modellnln(time) + sire + dam/sire + ln(time) × sire + ln(time) × dam/sire
 Fertilization successLogistic regressionlogitSire + dam/sire + sample/dam
 Development rateLogistic regressionlogitSire + dam/sire + sample/dam
 Metamorphosis successRandomization testlogitSire + dam/sire + sample/dam

Statistical inferences for binary data

We treated binary data with generalized linear models (McCullagh & Nelder, 1989; PROC GENMOD, SAS). In order to test for genetic differentiation in larval and juvenile survival rate, we analysed the trend of the corrected density estimates of larvae and juveniles with time (Fig. 1) using log-linear models for Poisson data (Table 1). The proportion of developed embryos 3 h after fertilization (=fertilization success) and the proportion of pediveliger larvae at day 24 (Fig. 1) estimated probabilities and, therefore, were analysed as binomial data using logistic regression (Table 1). The proportion of pediveliger larvae at day 24, which estimates the probability for a larva to reach the pediveliger state at day 24, was an indirect measure of larval development rate (and thus age at settlement), as day 24 corresponds to the first occurrence of pediveliger larvae in any of the full-sib families. For convenience, we will refer to it as ‘larval development rate’.

Numerical resampling and randomization test for metamorphosis success

Probability of success at settlement and metamorphosis (=metamorphosis success) was estimated as the ratio of the density of live juveniles 7 days after settlement to the density of pediveliger larvae put to settle (Fig. 1). Success of each cohort and total success over the 2 cohorts were analysed, but we will only present total success results because separate cohorts do not offer additional information. The estimates of density for pediveliger larvae and for juveniles were independent for a given cohort and between cohorts for a given full-sib family (Fig. 1). To analyse these data, we performed a resampling procedure coupled with a randomization test (Manly, 1997). We randomly paired without replacement the estimates of the density of pediveliger larvae of the 2 cohorts to obtain 4 estimates of the total density of pediveliger larvae in a full-sib family. Using the same procedure, we obtained 3 estimates of the total juvenile density. Then, we randomly paired without replacement the estimates of the total density of pediveliger larvae with those of the total juvenile density to obtain 3 estimates of total metamorphosis success, and we analysed the data obtained using the same nested logistic model as for fertilization success and larval development rate (Table 1). We repeated resampling 1000 times to obtain the distribution of the observed statistic P. Following Massot et al. (1994), we directly declared the effect significant if the 95% confidence interval of the observed statistics was below 0.05 and nonsignificant if the whole confidence interval was higher than 0.05. For intermediate cases (i.e. a part of the confidence interval is below or beyond 0.05), we performed randomization tests as described in Massot et al. (1994) with sire and dam within sire (=dam/sire) as treatment effects.

Replicate tests

Due to the limitation of rearing infrastructures, it was not possible to produce replicates for all the full-sib families. Although they were reared in the same water with controlled temperature and the same amount of food, we used separate tanks and trays and, thus, the dam/sire effect could be confounded with micro-environmental effects. To test for this possibility, we used the same sires and dams to produce 2 pools of larvae in addition to the full-sib families. The first pool resulted from the same mating design but with the embryos pooled just after fertilization, whereas the second came from a common fertilization event by mixing all oocytes and spermatozoids at the same time. Each pool of larvae was then divided into 2 separate tanks for larval rearing and maintained in 2 different trays during micro-nursing and nursing. We performed all the analyses previously described with pool and tank (or tray) nested within pool replacing sire and dam/sire random effects, in order to test for possible micro-environmental variance. No trait was significantly affected by the rearing tanks or trays (Table 2). Therefore, we can reasonably assume that in the other analyses the dam/sire effect is not significantly influenced by micro-environmental effects.

Table 2.  Statistical tests for potential micro-environmental effects.
TraitAnalysisMicro-environmental effectStatisticP
Larval traits    
Larval densityLog-linear modelTank/poolinline image  = 0.09 0.95
ln(time) × tank/poolinline image  = 0.01 0.99
ln2 (time) × tank/poolinline image  = 0.02 0.99
ln3 (time) × tank/poolinline image  = 0.06 0.97
Larval lengthWeighted ancovaTank/poolF2,1432  = 0.12 0.89
Time × tank/poolF2,1432  = 0.04 0.96
Metamorphic traits    
 Development rateLogistic regressionTank/poolinline image  = 0.22 0.90
 Size at settlementanovaTank/poolF2,236  = 0.08 0.93
 Metamorphosis successRandomization testSieve/pool>0.05
Juvenile traits    
 Juvenile densityLog-linear modelSieve/pool0.05inline image = 1.430.49
ln(time) × sieve/poolinline image  = 0.06 0.97
 Weight after metamorphosisanovaSieve/poolF2,116  = 2.04 0.14
Juvenile weightancovaSieve/poolF2,592  = 1.20 0.30
ln2 (time) × sieve/poolF2,592  = 0.38 0.68
 Final weightanovaSieve/poolF2,116  = 0.08 0.92

Variance components and heritability estimates

Genetic parameters for size at settlement, weight after metamorphosis and final weight were estimated after transformation of the data (Table 1). We estimated sire and dam/sire components of variance by comparing type III mean squares estimates with their expected values in a nested anova (PROC NESTED, SAS; see Lynch & Walsh, 1998 for details). We then computed additive inline image and nonadditive inline image genetic variance using classical quantitative genetic relationships between causal genetic components and observed components of variance (Lynch & Walsh, 1998). We estimated narrow sense (NSH) and broad sense (BSH) heritability as NSH = inline image /inline image and BSH = (inline image + inline image)/inline image, where inline image is the total phenotypic variance.

It was technically impossible to obtain individual measurements for larval and juvenile growth rate. To obtain genetic parameters for growth rate, we therefore performed numerical resampling combined with bootstrapping, as did Windig (1994) for the quantitative genetics of reaction norm slopes. We randomly drew one size measurement per full-sib family per day, in order to obtain a ‘individual’ growth curve over time. We then estimated growth rate as the slope of the log-transformed measurement of size (shell length and weight for larvae and juveniles, respectively) regressed over time (time and ln(time) for larvae and juveniles, respectively). We repeated this procedure 30 times without replacement for each full-sib family to obtain 30 individual growth rate measurements per family. We then estimated the genetic parameters as described before. We resampled 1000 times and used the mean of the different values of interest as estimates. This method should inflate within-full-sib family variance of growth rate and thus the error term, but should give unbiased estimates of sire and dam/sire variance components. Heritability estimates for growth rate should, therefore, be downwardly biased. This additional source of error was taken into account in the computation of the bootstrapped standard error presented below.

We computed genetic parameters for binary data (larval and juvenile survival, larval development rate and total metamorphosis success) using quantitative genetics methods for threshold characters (Lynch & Walsh, 1998). We first calculated variance components (PROC GLM, SAS) and heritabilities on the ‘observed’ scale (i.e. 0/1 data) as described before. Then, we transformed the estimates to the ‘liability’ scale (i.e. the continuous scale underlying a threshold trait) according to Robertson's appendix in Dempster & Lerner (1950). Although we tested for genetic effects on survival rates (the trends of the density of individuals with time) for both larval and juvenile stages, no individual survival rate could be defined. Therefore, we computed the heritability of survival ‘probability’ over each stage considered. To do this, we paired the independent sample estimates of density at the initial and final dates of larval and juvenile stages using the same resampling procedure as for statistical analysis of metamorphosis success. ‘Individuals’ that were still alive at the final date were classed 1 and ‘individuals’ that were dead were classed 0, and we computed the genetic parameters as described above. We repeated this procedure 1000 times and used the means of the different genetic parameters as definitive estimates. We used the same procedure for total metamorphosis success. Finally, for larval development rate (=proportion of pediveliger larvae at day 24), no resampling was needed since the density of pediveliger larvae and the total larval density were estimated using the same water samples.

Due to mortality, data were unbalanced and no exact estimation of the standard error is available for variance components and heritabilities in this case (Becker, 1984). Therefore, we used bootstrapping methods (Efron & Tibshirani, 1993) to estimate standard errors for each trait. Specifically, we randomly drew sire and dam/sire 1000 times with replacement and used the standard error of the distribution obtained as the standard error estimate. When numerical resampling of the data at the lowest level (individuals or full-sib family samples) was already needed for variance components and heritability estimates, we coupled the bootstrapping procedure with resampling to account for errors from numerical resampling.

Genetic correlations

As the different traits were measured on different individuals, no phenotypic correlation between traits could be estimated. For the same reason, we used family means to compute genetic covariances between traits according to the formulae covh.s(x, y) = 1/4 cova(x, y) and covf.s(x, y) = 1/4 cova(x, y) + 1/4 covna(x, y), where covh.s(x, y) and covh.s(x, y) are the covariances between half-sib or full-sib family means of the traits x and y, respectively, and cova(x, y) and covna(x, y) are the additive and nonadditive genetic covariances, respectively. The estimates of genetic covariance should be unbiased since the different traits were measured on independent samples of individuals within each full-sib family (Lynch & Walsh, 1998). Narrow and broad sense genetic correlations were then computed as ρa(x, y) = cova (x, y)/(σa(xa(y)) and ρa + na(x, y) = [cova(xy) + covna(x, y)]/[inline image (x) + inline image (x)]1/2[inline image (y) + inline image (y)]1/2, respectively, using the genetic variances obtained from individual data. Significance level for genetic correlations was obtained by testing whether genetic covariances were different from 0, which in our case was the same as testing whether family mean covariances were different from 0. These tests were performed using classical linear regression analysis (PROC REG, SAS).


Fertilization success and larval traits (Table 3)

Table 3.  Descriptive statistics.
TraitMeanSEHalf-sib varianceFull-sib variance×10XX
  1. Mean : mean value across all families. SE : standard error. Half-sib variance : variance among half-sib family means. Full-sib variance : variance among full-sib family means. × 10 XX : power of ten for half-sib and full-sib variance. Descriptive statistics for larval survival probability, larval growth rate, metamorphosis success, juvenile survival probability and juvenile growth rate were determined using the data obtained by numerical resampling for the computation of the components of variance. Standard errors for binary data, i.e., larval and juvenile survival probability, larval development rate and metamorphosis success, are not based on variation between individuals as for quantitative traits but on variation between samples. Descriptive statistics for larval and juvenile survival are given for the survival probability over the entire stage considered and not for the survival rate (see Materials and methods).

Larval traits     
 Larval survival0.1200.1514.217.3−03
 Larval growth0.0595.192 × 10−032.913.3−06
Metamorphic traits     
 Development rate0.2890.1427.318.6−03
 Size at settlement281248512400
 Metamorphosis success0.5500.1479.319.3−03
Juvenile traits     
 Juvenile survival0.1350.0522.325.8−04
 Weight after metamorphosis0.0016.474 × 10−042.85.2−08
 Juvenile growth0.3590.0325.511.6−04
 Final weight4.0872.2852.410.4−02

Fertilization success depended on both parental effects whereas sample/dam effect was not significant (Table 4).

Table 4.  Hypothesis testing.
TraitSimple effectsStatisticPInteractionsStatisticP
  1. dam/s.  = dam/sire, sample/d.  = sample/dam. Interactions between both sire and dam/sire and continuous effects ( timex or ln x ( time ) with x ranging from 0 to 3) in polynomial models were used to test for genetic variation in the different rates, i.e., survival rate and growth rate.

Fertilization successSireinline image  = 168.34 <10−3   
Dam/s.inline image  = 185.97 <10−3   
Sample/d.inline image  = 11.96 0.980   
Larval traits      
Larval densityln(time)inline image  = 865.96 <10−3ln(time) × sireinline image  = 79.96 <10−3
ln2 (time)inline image  = 450.54 <10−3ln2(time) × sireinline image  = 37.07 <10−3
ln3 (time)inline image  = 370.33 <10−3ln3(time) × sireinline image  = 32.34 <10−3
Sireinline image  = 437.95 <10−3ln(time) × dam/s.inline image  = 95.56 <10−3
Dam/s.inline image  = 296.67 <10−3ln2(time) × dam/s.inline image  = 63.46 <10−3
   ln3(time) × dam/s.inline image  = 57.02 <10−3
Larval lengthTimeF1,5  = 4759.51 <10−3   
SireF5,15  = 3.91 0.018Time × sireF5,15  = 1.05 0.426
Dam/s.F15,7368  = 26.31 <10−3Time × dam/s.F15,7368  = 35.60 <10−3
Metamorphic traits      
Development rateSireinline image  = 214.38 <10−3   
Dam/s.inline image  = 416.46 <10−3   
Sample/d.inline image  = 127.71 <10−3   
Size at settlementSireF5,15  = 4.11 0.015   
Dam/s.F15,1237  = 8.74 <10−3   
Metamorphosis successSireRandomization<0.05   
Juvenile traits      
Juvenile densityln(time)inline image  = 1219.23 <10−3   
Sireinline image  = 70.05 <10−3ln(time) × sireinline image  = 51.50 <10−3
Dam/s.inline image  = 229.80 <10−3ln(time) × dam/s.inline image  = 146.29 <10−3
Weight after metamorphosisSireF5,13  = 2.50 0.085   
Dam/s.F13,551  = 3.53 <10−3   
Juvenile weightln2 (time)F1,5  = 13207.96 <10−3   
SireF5,13  = 1.48 0.262ln2(time) × sireF5,13  = 1.81 0.179
 Dam/s.F13,2812  = 2.09 0.012ln2(time) × dam/s.F13,2812  = 2.01 0.017
Final weightSireF5,13  = 0.73 0.616   
Dam/s.F13,551  = 0.71 0.753   

Survival rate, given by the trend of the logarithm of larval density ln(nL) with time (Fig. 2A), was very low for 2 days after fertilization, then increased and finally approximately stabilized. Therefore, we modelled ln(nL) as a third-order polynomial of ln(time) (Table 1, model; Table 4, analysis results). Increasing the polynomial order did not improve the model. As for fertilization success, initial larval density depended on both sire and dam/sire (Table 4). Moreover, the significant interactions between sire or dam/sire and the different powers of ln(time) (Table 4) suggested genetic bases for survival rate. The large genetic variance for larval survival probability confirms these findings and results in relatively high NSH (0.55) and BSH (0.81) (Table 5).

Figure 2.

Larval traits. (A) Survival: curve of the mean logarithm of larval density ln(nL) with time . (B) Growth: curve of the mean logarithm of larval shell length ln(lL) with time .

Table 5.  Narrow sense and broad sense heritabilities (NSH and BSH, respectively).
Trait NSH ± SE BSH ± SEStatistical analysisSire effectDam/sire effect
  1. *P < 0.001; †P < 0.005; NS, non significant.

  2. SE: bootstrapping standard error. Sire effect and dam/sire effect: significance of the sire and dam/sire effect, these can be considered as statistical tests for the significance of narrow and broad sense heritability, respectively. For traits modelled as polynomials of time or ln(time) (larval survival rate, larval growth rate, juvenile survival rate, juvenile growth rate), sire or dam/sire effect are indicated significant as soon as one interaction with a polynomial term is significant. Note that for larval and juvenile survival, hypothesis testing is made on survival rate whereas heritability is computed for survival probability over the entire stage considered (see Material and methods).

Larval traits     
 Larval survival0.55 ± 0.400.81 ± 0.29Log-linear model**
 Larval growth0.24 ± 0.570.91 ± 0.65Weighted ancovaNS*
Metamorphic traits     
 Development rate0.10 ± 0.170.37 ± 0.19Logistic regression**
 Size at settlement0.41 ± 0.290.41 ± 0.16anova*
 Metamorphosis success0.10 ± 0.170.31 ± 0.14Randomization test
Juvenile traits     
 Juvenile survival−0.02 ± 0.070.24 ± 0.09Log-linear model**
 Weight after metamorphosis0.19 ± 0.180.30 ± 0.13anova
 Juvenile growth0.05 ± 0.180.32 ± 0.20ancovaNS
 Final weight−0.01 ± 0.02−0.04 ± 0.03anovaNSNS

The logarithm of larval length ln(lL) increased linearly and significantly with time (Fig. 2B, Tables 1 and 4). Significant sire and dam/sire effects on initial larval length (24 h after fertilization; Table 4) implied genetic control for growth during the first 24 h. In contrast, the slopes of the regression of ln(lL) over time varied with dam within sire but not sire (Table 4), suggesting that larval growth rate had a low additive genetic variance relative to the nonadditive component. Indeed, NSH (0.24) appeared to be much lower than BSH (0.91), due to a large nonadditive component (67%) (Table 5).

Metamorphic traits (Table 3)

Sixteen of the full-sib families exceeded the 10% threshold of pediveliger larvae at day 24 (second cohort at day 26) whereas the five remaining families only exceeded it at day 27 (second cohort at day 29). We detected some genetic bases for larval development rate (the proportion of pediveliger larvae at day 24) as shown by the significant sire and dam/sire effects (Tables 1 and 4). The sample/dam effect was also significant (Table 4), resulting in a high error term. NSH was therefore low with a high standard error (0.10 ± 0.17) whereas BSH was much higher (0.37) (Table 5).

Size at settlement was significantly affected by sire and dam/sire (Tables 1 and 4), indicating genetic variance for this trait. Both NSH and BSH were high (0.41), but equal (Table 5). This arose because the nonadditive genetic component of variance was 0 despite the significant dam/sire effect. Such low or even negative estimates often arise when other variance components are high and the number of families used in the mating design is small (Lynch & Walsh, 1998).

Some genetic variance was also detected for metamorphosis success since all the effects in the model were significant (Tables 1 and 4). NSH (0.10) and BSH (0.31) were moderate but the standard error for the NSH was high (0.17) (Table 5), perhaps due to the resampling.

Juvenile traits (Table 3)

The survival rate of juveniles was almost constant: the logarithm of the density of juveniles ln(nJ) decreased linearly with ln(time) (Fig. 3A, Tables 1 and 4). The regression slope depended significantly on sire and dam/sire (Table 4), revealing genetic variance for juvenile survival rate. However, we found a slightly negative NSH for juvenile survival probability (−0.02) and moderate BSH (0.24) (Table 5). Thus, the negative estimate of NSH is certainly due to the small number of families.

Figure 3.

Juvenile traits. (A) Survival: curve of the mean logarithm of juvenile density  ln (nJ) with time . (B) Growth: curve of the mean logarithm of juvenile weight  ln (WJ) with time .

The trend of the logarithm of juvenile weight ln(wJ) with time was best described as a function of ln2(time) (Fig. 3B, Tables 1 and 4). Weight after metamorphosis, treated either as initial weight in the ancova of ln(wJ) with ln2(time) or directly analysed using an anova (Table 1), significantly differed among dams within sire but not among sires (Table 4). Juvenile growth rate also varied among dams within sire, as shown by the significant interactions between dam/sire and ln2(time), but not among sires (Table 4). Finally, neither sire nor dam/sire effect significantly affected final weight (Tables 1 and 4). Heritability estimates of juvenile growth rate and final weight were low (Table 5). In contrast, and despite a nonsignificant sire effect, weight after metamorphosis had a substantial NSH (0.19) and BSH (0.30) (Table 5). This pattern could be linked to the strong genetic variance in size at settlement.

Genetic correlations

Overall, we found consistent genetic correlations (i.e., significant with the same sign across broad and narrow sense values) between traits within the same stage, between larval and metamorphic traits and between metamorphic and juvenile traits but not between larval and juvenile traits, even when these traits were qualitatively the same (e.g. larval and juvenile growth rates). In this section, we mostly describe significant correlations. When we refer to nonsignificant correlations, we indicate them explicitly (Table 6 and Fig. 4 give detailed genetic correlation results). Note that some genetic correlations have an absolute value larger than 1 as a consequence of low heritability and the small mating design (Hill & Thompson, 1978). In addition, as seen before, some traits had negative genetic variance estimates, so that no genetic correlation could be computed. In that case, we only refer to genetic covariances.

Table 6.  Genetic correlations.
  1. *P < 0.01; †P < 0.05.

  2. Broad and narrow sense genetic correlations are presented above and below the diagonal, respectively. For traits with negative genetic variance estimates, no genetic correlation could be computed. In this case, only the sign and the significance level of the genetic covariance are presented. LS, larval survival probability; LG, larval growth rate; DR, larval development rate; SSet., size at settlement; MSuc., metamorphosis success; JS, juvenile survival probability; WM, weight after metamorphosis; JG, juvenile growth rate; FW, final weight.

LS 0.89†0.68†0.85†−0.33−0.310.57−0.64
LG0.55 1.55*1.33*−0.600.091.39*−1.22*
DR0.451.82* 1.30*−0.37−0.151.17*−1.10*
SSet.0.341.33†1.57* −0.46−0.79*1.65*−1.41*
MSuc.−0.54−1.83*−1.95†−1.54† −0.23−0.84†0.58+
JS−†−†+ −0.070.13+
WM−†−1.45 −1.50*
JG0.16−1.23−1.39−1.501.79+−1.89* +†
Figure 4.

Genetic correlations between the different life-history traits: (A) broad sense genetic correlations; (B) narrow sense genetic correlations. Positive significant correlations are shown by solid lines, negative significant correlations by dashed lines.

First, larval growth rate was positively correlated with larval development rate and size at settlement, and the genetic correlation between larval development rate and size at settlement was positive, so that the fastest growing genotypes during the larval stage settled and metamorphosed at the earliest age and the largest size.

Second, additive genetic correlations were negative between metamorphosis success and both larval development rate and size at settlement, suggesting that genotypes settling and metamorphosing early and large had poor survival during metamorphosis, whereas those settling and metamorphosing late and small had a better survival probability. In this case, the broad sense genetic correlations were also negative but nonsignificant. In addition, size at settlement and larval development rate negatively covaried with juvenile survival, with different significance levels for broad and narrow sense (Table 6). These covariances also suggest costs for metamorphosing early and large.

Third, the genetic correlation between size at settlement and juvenile weight after metamorphosis was positive, suggesting an early size advantage of settling and metamorphosing large. However, both traits were negatively correlated with subsequent juvenile growth rate (significantly for broad sense only in case of size at settlement, Table 6) and the early advantage of settling and metamorphosing at a large size disappeared with time as the covariance between size at settlement/weight after metamorphosis and final juvenile weight was negative and nonsignificant.


Early life-history traits in C. gigas have genetic bases. Specifically, we found substantial significant heritabilities in larval survival and metamorphic traits whereas genetic variation in larval growth rate and juvenile traits was weak and/or nonsignificant. Genetic variability was associated with genetic correlations. Among these, larval development rate and size at settlement were positively correlated, contrasting with the expected trade-off between age and size at metamorphosis. On the other hand, both larval development rate and size at settlement were negatively correlated with metamorphosis success and covaried negatively with larval survival, suggesting some costs of metamorphosing early and large. In addition, the size advantage of settling large (positive correlation between size at settlement and weight after metamorphosis) disappeared with time as shown by the absence of covariance between size at settlement and final weight. Finally, we observed no genetic correlations between larval and juvenile stages, suggesting a decoupling between the life-stages.

Experimental design

Two potential limitations of the experimental design should be noted. First, due to the sophistication of the infrastructures and the amount of technical work necessary for larval rearing and oyster farming, our mating design was relatively small (24 full-sib families distributed in six half-sib families). Consequently, the precision of the estimates of genetic parameters and the power of statistical tests were low. Second, by producing only two settlement cohorts per full-sib family and discarding the remaining larvae, we counter-selected slow growing and developing larvae, so that variability within families was truncated at settlement. However, the induced bias should be weak since the maximum proportion of individuals discarded among all the full-sib families was only 9.8%. Despite these limitations, most of the results were clear-cut and produce a consistent integrated picture.

Genetic variation and parental effects in early life-history traits of Crassostrea gigas

The amount of genetic variation differed strongly across stages and across traits within stages. We found a strong heritability for larval survival probability (NSH 0.55 and BSH 0.81, Table 5), which contradicts previous results of Lannan (1980) who observed no additive genetic variance for larval survival. The comparison, however, might be unreliable because Lannan did not analyse survival as a threshold character as we did. In contrast with larval stage, we observed no additive genetic variation for juvenile survival probability and only moderate BSH (0.24, Table 5). Such a decrease in the amount of genetic variation across stages could indicate homogenization of survival probability, across families and over time, due to the death of the less resistant individuals at earlier ages.

NSHs of larval and juvenile growth rates (0.24 and 0.05, respectively) were lower than values reported in other marine bivalve molluscs (Strömgren & Nielsen, 1989; Rawson & Hilbish, 1991; Hilbish et al., 1993; Toro & Parades, 1996). However, in contrast with our case, these heritabilities are not based on actual growth rate but on genetic variation in shell length at a given age. Weight after metamorphosis meanwhile displayed substantial significant NSH and BSH (0.19 and 0.30, respectively, Table 4), which could be related to the observed heritability in size at settlement and the positive genetic correlation between these two traits (see below). However, genetic variation in size disappeared with age as shown by the lack of genetic variation in final weight (Table 5).

The three metamorphic traits exhibited significant additive and nonadditive genetic variation (Table 5). Larval development rate and metamorphosis success had the same NSH (0.10), which was low relative to the NSH of size at settlement (0.41). This latter value might seem surprising compared to the moderate nonsignificant NSH of larval growth rate. However, size at settlement does not result from larval growth rate alone but rather from the functional relationship between larval development and growth rates, which may well be heritable (Smith-Gill & Berven, 1979; see also below). To our knowledge, no study on the genetic basis of metamorphic traits in marine benthic invertebrates is available for comparison. Nevertheless, in holometabolous insects and in amphibians, larval development rate and/or size at metamorphosis (or equivalent traits) have been shown to have heritable bases (Via, 1984a; Berven, 1987; Travis et al., 1987; Frankham, 1990; Tucic et al., 1998).

Phenotypic parental effects, rather than real genetic variation, can be highly influential on fertilization success as well as on larval growth rate. Indeed, the significant sire and dam/sire effects on the probability of fertilization and successful embryo development after 3 h could result from parental differences in gamete quality such as spermatozoid mobility and/or oocyte maturity (Ramirez et al., 1999; Boudry et al., 2002). By the same line of argument, the strong nonadditive component (67%) of larval growth rate could reflect maternal effects such as differences in stored lipid reserves among oocytes. These have already been shown to positively influence larval growth in some marine bivalves (Gallager & Mann, 1986; Gallager et al., 1986). The fact that the nonadditive component of variance for juvenile growth rate was much lower (27%) also supports this hypothesis since maternal effects are expected to become less pronounced with age (Dufty et al., 2002).

When and how large to settle and metamorphose?

Contrary to the theoretically expected fitness trade-off between age and size at metamorphosis (Wilbur & Collins, 1973; Werner, 1986), we found a positive genetic correlation between larval development rate and size at settlement (Table 6, Fig. 4). Therefore, early settlement and metamorphosis were associated with large size. This result contrasts with most of the genetic studies in amphibians and insects, which reported genetic evidence for a trade-off between age and size at metamorphosis (amphibians: Travis et al., 1987; Newman, 1988; insects: Via, 1984a,b; Moller et al., 1989; Zwaan et al., 1995; Nunney, 1996; Tucic et al., 1998). Two exceptions, however, should be noted. Berven (1987) and Gebhardt & Stearns (1988) reported, in a wood frog (Rana sylvatica) and a fruitfly (Drosophila mercatorum) genetic correlations between traits related to age and size at metamorphosis that changed sign according to the environment. The genetic expression of the trade-off might then be environment-dependent. In addition, the theoretical studies by Rowe & Ludwig (1991) and Abrams et al. (1996) showed that, in case of time constraints such as seasonal reproduction or resource availability, individuals with fast larval growth could optimize both age and size at metamorphosis. As we observed a positive genetic correlation between larval growth rate and both larval development rate and size at settlement (Table 6, Fig. 4), the possibility that time constraints affect larval life in C. gigas and could explain the absence of negative genetic correlation between larval development rate and size at settlement should be explored.

An alternative explanation can be found in the theoretical study by Smith-Gill & Berven (1979), who argued that metamorphic climax is a fixed developmental state and, thus, that larval development (or differentiation) rate fixes age at metamorphosis. Size at metamorphosis would then be a simple by-product of the relationship between larval development and growth rates. Specifically, if larval development and growth rates are variable and positively correlated, individuals can metamorphose early and large. Larval development and growth rates could then evolve relative to each other so as to optimize both age and size at metamorphosis. The fact that we found some genetic variation in larval development and growth rates (at least in the broad sense for the latter), positive genetic correlations between these two traits and between larval development rate and size at metamorphosis also supports this hypothesis. However, this assertion should be tempered because of the potential maternal effects acting on larval growth rate highlighted above.

Some trade-offs could induce costs of metamorphosing early and large

Some trade-offs were suggested by the fact that both larval development rate and size at settlement were genetically and negatively correlated with metamorphosis success and negatively covaried with juvenile survival probability (Table 6, Fig. 4) in the narrow sense. Broad sense estimates were negative but nonsignificant. These correlations mean that fast developing genotypes settle and metamorphose early and large, but have a low survival probability during metamorphosis and the juvenile stage. This could be interpreted as costs of metamorphosing early and large. However, it must be considered that juvenile survival had low or negative genetic variance (Table 5), so that any genetic covariance detected may be irrelevant.

We know of no study assessing genetic evidence for the same trade-offs, but Chippindale et al. (1997) and Borash et al. (2000) found a negative genetic correlation between development time and larval viability in Drosophila melanogaster. Moreover, some known physiological processes could explain the cost in terms of metamorphosis success in marine bivalve molluscs. Metamorphosis is a critical period in terms of survival because it requires a high level of energy and individuals are unable to feed during this period (Holland & Spencer, 1973; Crisp, 1976; Bartlett, 1979; Rodrigez et al., 1990). A high larval development rate consumes most energy to produce new cells and organs leaving low levels of stored resources, whereas a low development rate would allow individuals to save a part of these resources. It would, therefore, be likely that fast larval development induces higher mortality.

The size advantages of settling large

The positive genetic correlation between size at settlement and weight after metamorphosis confirmed the early size advantage of settling large. However, larval development rate, size at settlement and weight after metamorphosis were negatively correlated with juvenile growth rate (Table 6), suggesting that individuals settling early and large grew slower after metamorphosis. However, this result must be considered with caution since juvenile growth rate exhibited low genetic variance (Table 5). Yet, confirming the previous result, the early advantage of settling large disappeared with time (no genetic variation in final weight and no genetic covariance between size at settlement/weight after metamorphosis and final weight). These findings contrast with previously observed phenotypic correlations in several marine molluscs. Indeed, the length of the larval period was found to be phenotypically and negatively correlated with the juvenile growth rate in C. gigas (Collet et al., 1999) and with shell length at 5 months in Ostrea edulis (Newkirk & Haley, 1982). However, as in our study, Newkirk & Haley (1982) observed that this phenotypic correlation vanishes with time in O. edulis.

Decoupling of larval and juvenile stages

We observed no consistent genetic correlation between pre- and post-metamorphic traits (Table 6, Fig. 4). Again, this result differs from certain phenotypic correlations observed in some marine molluscs: larval and juvenile growth rates appeared to be positively correlated at the phenotypic level in Crassostrea virginica (Newkirk et al., 1977) and in Crepidula fornicata (Pechenik et al., 1996a,b). However, our findings are consistent with the previous observations that some correlations or covariances (phenotypic or genetic) disappear with time over the juvenile stage. Moreover, this absence of correlation between pre- and post-metamorphic traits supports the hypothesis of a decoupling between life-stages in order to break fitness trade-offs between adaptations to different tasks (Moran, 1994). This reinforces the interpretation of the larval pelagic stage as mainly dedicated to dispersal and habitat selection and of the adult benthic stage as specialized for growth and reproduction (Wray & Raff, 1991; Pechenik, 1999).

Conclusion: genetic polymorphism in the early life-history strategy of Crassostrea gigas

Some genetic polymorphism in the early life-history strategy of C. gigas seems to be maintained along a gradient between two extremes. Fast growing and developing genotypes are the first extreme. They settle and metamorphose early and large, but the optimization of both age and size at settlement incurs a cost in terms of success at settlement and metamorphosis. Moreover, post-metamorphic performances of these genotypes in terms of survival and growth rate are diminished. Specifically, the size advantage at metamorphosis is effective during early post-metamorphic life but vanishes with time. At the other end of the gradient are slow developing and growing genotypes. They will settle and metamorphose late and at a small size. However, their success at settlement and metamorphosis as well as their post-metamorphic growth rate and survival are higher.

Two possible scenarios for the maintenance of such polymorphism are, firstly, that some trade-offs underlie the different negative genetic correlations that we observed. These could be responsible for the maintenance of genetic variation in the different early life-history traits through antagonistic selection (Rose, 1982; Barton & Turelli, 1989; Roff, 1992; Stearns, 1992) and, therefore, would underlie the maintenance of genetic polymorphism in the early life-history strategy as a whole. This would be possible if the trade-offs balanced fitness costs and benefits perfectly, so that genotypes located at different points along the gradient of strategies previously described would have equal fitness. The second possibility is that genetic polymorphism results from the different strategies being favoured in different environmental conditions (environment describing here all abiotic and biotic factors including other species). The random dispersal of larvae across environments could allow the maintenance of genetic polymorphism in early life-history strategy through genotype-by-environment interactions in fitness components resulting in differential success of genotypes across environments (Felsenstein, 1976; Gillespie & Turelli, 1989; Stearns, 1992). The very high fecundity and the random larval dispersal characterizing C. gigas could then be considered as a parental bet-hedging strategy. Parents would spread the risk of offspring mal-adaptation over all possible environments, so that at least some of their progeny encounter environmental conditions to which they are adapted. It remains to be explored under which environmental circumstances genotypes settling and metamorphosing early and large but with a small probability of survival during metamorphosis and juvenile stage could be favoured as opposed to genotypes settling late and small but with a high probability of survival.


We would like to thank the technical teams of the IFREMER Laboratories at Rouce-les-Bains and Bouin for their assistance with the production and rearing of the animals used in this study. Our thanks also to André Gérard, Emmanuel Goyard and Serge Heurtebise for discussion of the experimental design. This work was made possible by a PhD grant from the Ministère de l'Education et de la Recherche to Bruno Ernande and was conducted as part of the European Union Funded project ‘GENEPHYS’ (FAIR PL 95.421).

Received 28 August 2002; revised 5 December 2002; accepted 3 March 2003