Isofemale line technique

In this study, we applied the isofemale line technique (ILT), which is a method that is commonly used on wild populations of Drosophila sp. (review by [32]). The relationship between narrow-sense heritability (h²) and the coefficient of intraclass correlation (t) provided by ILT is not very obvious. Nevertheless, t does give a better heritability estimate than the half-sib design or the parent–offspring regression method. For example, the technique provides measures for about 95 ± 15% of the morphometric trait heritabilities in natural populations of D. melanogaster [32], thus yielding a close estimation of additive genetic variance. It is a useful tool for rapidly estimating genetic variability and the genetic correlation between traits, and can also compare these genetic parameters between different traits and different environments (or even different populations). Furthermore, ILT can be easily applied to C. elegans because of its androdioecious breeding system (i.e. self-fertilization of hermaphrodites and facultative outcross with males). Indeed, after mating, each isogenic line can be conserved as an inbred line. The term “heritability” has been used throughout this paper instead of “the coefficient of intraclass correlation”.

Nematode culturing

We used a stock population of C. elegans, composed of a mixture of 16 wild isolates [33], to obtain a large genetic diversity. The population, containing more than 30% males, was maintained under laboratory conditions for 140 generations, where recombination–selection equilibrium was mostly achieved without significant loss of genetic diversity [33]. We put 500 individuals in a 9 cm diameter Petri dish containing NGM-modified (i.e. nematode growth medium) agar and HEPES buffer [34]. We produced six replicated experimental populations. The plates were seeded with 1 ml of a 20:1 mixture of Escherichia coli strain OP50 (OD_600nm = 3) as the food source. Every 3 days we washed the nematodes (about 20,000 individuals) off the plates with an M9-modified solution (containing HEPES buffer) and kept each replicate in 15 ml Falcon tubes. The number of individuals in a tube was estimated using five sample drops of 5 µl, following Teotónio et al. [33]. Then a volume corresponding to 500 individuals, for all developmental stages, was placed in a fresh Petri dish. The nematodes were cultured throughout the experiment at 20°C and 80% RH.

Creation of lines

After repeating this protocol about 40 times (~40 generations despite the discrete overlapping of developmental stages), we selected 100 gravid hermaphrodites and placed them on a new Petri dish. This was repeated for all six replicates. Potential evolutionary changes caused by the changing environmental conditions should be similar in both the treatment and control environments. We allowed the hermaphrodites to lay eggs for a 2 h period, and then transferred approximately 200 eggs to a new Petri dish (200 eggs per replicate). After 40 h, when sexual differentiation was visually possible, but individuals were not yet mature (i.e. L3 larval stage [35]), we randomly picked a single hermaphrodite and a single male from the different Petri dishes and placed them separately (by couples) in six-well tissue-plates. We created 14 couples.

The NGM in the wells was seeded with 150 µl of a 5:1 mixture of OP50. After leaving the plates in a laminar flow hood for 1 h to allow them to dry, the plates were then top-exposed to UV doses for 90 s to stop bacterial growth (Bio-Link Crosslinker; λ = 254 nm; intensity = 200 µwatt.cm^-2). The main aim of this UV treatment was to avoid differential bacterial growth in the control and polluted plates. Reproduction by mating produces a high frequency of males in the progeny, whereas hermaphrodite self-fertilization produces only 0.1% males [36–38]. Therefore, 72 h after the separation of the couple, we checked to see if the ratio of males and hermaphrodites in the progeny was approximately 1:1, to guarantee that mating between the hermaphrodite and the male was successful. We then picked one L3 stage hermaphrodite in each of the progeny of 14 couples to initiate 14 isogenic lines. This was done 72 h after couple creation to avoid self-fertilization, which is often a risk in the first few hours after mating [39]. Each line was reproduced by self-fertilization for two generations in a 6 cm Petri dish, in the same environment as used above.

Contamination design and measurement of traits

When the second self-fertilization generation had appeared, we transferred 20 gravid hermaphrodites of each line to a new Petri dish. This was considered as t = 0. After 2 h of egg-laying, the eggs were individually placed in a 12-well tissue culture plate. Twelve eggs per line and per environment were placed in three different NGM media: control, uranium (addition of 1.1 mM U [uranyl nitrate: UO₂(NO₃)₂·6H₂O; Sigma–Aldrich, France]), and high salt concentration (addition of 308 mM NaCl). The wells were seeded with 75 µl of a 5:1 mixture of OP50 (same UV treatment as the six-well plates). In our experiment, the term “isogenic” does not correspond exactly to our lines, as there was heterozygosity in our population [33]. However, the individuals had been reproduced twice by self-fertilization before the experiment; so we expected a significant reduction in heterozygosity [40–42].

Individuals growing in salt showed a slower development rate than in the control individuals and those in the uranium. Therefore, individuals were transferred twice into a new well at 96 h in the control and uranium environment, or at 168 h in the salt environment and then again 36 h later. Hatched progeny were counted the day following each transfer to measure brood size, which is an index of fecundity. We photographed the individuals using a stereomicroscope (Olympus SZX12, 1.6 x 90 magnification) with a computer-connected camera (Nikon D5000). Body length was measured at 72 and 144 h using a micrometer scale measure and ImageJ software [43]. These two measurement times corresponded approximately to the population’s onset and to the end of the egg-laying period in the control environment, respectively. These two measures were called early growth and late growth in our study. Survival was estimated by counting the number of parental individuals still alive at 144 h.

Genetic parameter estimations

We estimated the genetic parameters by using a Bayesian model approach in the MCMCglmm package for the generalized linear mixed-effects model [44] in the R software [45]. We used the multivariate generalized linear mixed-effects models for different traits within the same environment (a quadrivariate model per environment) and for the same trait across different environments (a trivariate model per trait). The models did not contain fixed effects, but we included lines as random effects to estimate the between and within-line variance for each trait. We modelled survival using a binary error structure and the Gaussian model for the other traits. Expanding the binomial data into binary data allowed us to fit a model that could estimate the correlation between survival and the other traits. Estimations were generated using a Bayesian approach so that we could obtain the entire posterior distributions of (co)variance matrices of traits rather than only point estimates. To avoid any biased results caused by the fact that trait mean values differed, we rescaled the traits prior to analysis by subtracting each value by the mean of the sample and dividing it by twice the standard deviation [46]. After several priors had been tested, we obtained a proper prior [nu = k – 1 + 0.002] with a very low variance parameter [V = diag(k) * V_p * 0.05], where V_p is the phenotypic variance and k is the dimension of V (i.e. number of traits). There was one exception for survival. In this case we fixed the environmental (also called the residual) variance to one. For the multivariate analyses with four different traits, we used the models to estimate different random and environmental variances and covariances between the pairs of traits. In the models that analysed the same trait across the three different environments, the traits were measured on different individuals, which fixed the environmental covariance to zero (see S1 Table for comparisons between the models, with or without genetic covariance, between traits allowed in the prior). Despite the slightly informative priors used in the models, starting with a low variance and any kind of covariance allowed us to say that the generated (co)variances were tangible. After verifying the convergence of the parameter values (i.e. number of iterations, burn-in phase and thinning), and autocorrelation issues, we retained 2 500 000 iterations with an initial burn-in phase of 500 000, from 1000 samples per analysis [44].

We estimated heritability in the multivariate models for traits within environments using t = nVb / (nVb + (n – 1)Vw), where Vb and Vw were respectively the between and within-line variance and n was the number of lines [32]. Following David et al. [32], we also estimated genetic correlations without correcting for variance and covariance within-line, although n was less than 20. Using this correction generated the posterior distribution for correlations containing values greater than 1. We used the posterior mode of the distribution (i.e. the value that appears most often in the obtained distribution) to estimate the heritability, and genetic and phenotypic correlations as quantitative genetic parameters. For each trait, we also estimated the phenotypic [V_P = Vw + Vb], environmental [V_E = V_P—V_G] and genetic variances [V_G = tV_P], and the genetic covariance between traits using , where r_G corresponds to genetic correlation between traits 1 and 2; and V_G1 and V_G2 are the genetic variance for trait 1 and 2, respectively (see S1 Fig.). We assumed that each estimate was significantly different from zero when its 95% Bayesian highest posterior density intervals (HPDIs) were not zero. We also tested the differences between an individual’s traits and the genetic parameters in different environments. We assumed differences to be significant between two environments when 95% HPDIs produced by the subtraction between the posterior distributions of the trait in the two polluted environments did not include zero.

Effects on heritability

Survival heritability was absent or extremely low (thus difficult to detect) in the control environment, compared to the other traits. This may be due to its genetic regulation. Survival is regulated by a few genes that have large effects contrary to growth or reproductive traits, which are regulated by many genes that have small effects on heritability [1,49] (in C. elegans [50]). Heritabilities in the control environment were moderate for early growth and fecundity, and low for late growth. Gutteling et al. [31] studied broad-sense heritability in recombinant inbred lines of C. elegans for body mass at maturation and fecundity at different temperatures. Heritability was similar to our study’s results for fecundity, but was higher for growth. However the variations can be explained by the differences in temperature between our study (20°C) and theirs (12°C and 24°C), which affects average growth, reproduction [51], and heritability [31].

The decrease in early growth and fecundity heritabilities in both polluted environments was consistent with several previous laboratory or field studies, although some studies showed the opposite response (reviewed by [17,18]). Heritability reduction could be caused by a decrease in additive genetic variance or an increase in environmental variance. Our results showed that both environmental and genetic variances were higher in the control, resulting in a phenotypic variance decrease in both polluted environments (Fig. 1). Despite the decrease in phenotypic variance, the reduction in heritabilities was mainly due to the reduction in (additive) genetic variance. Although a reduction in heritability is commonly caused by an increase in the environmental variance component, there can be cases where it results from a reduction in the genetic variance components [17]. In this latter case, the decrease in heritability could be triggered by a limitation in the genetic potential due to constraining developmental conditions [21], e.g. despite strong additive genetic effects responsible for individual differences in body size, the potential size may not be attained because of stress-induced reductions in the growth rate [52].

Heritabilities were extremely low in the uranium environment. Several studies have suggested that phenotypic variation in stressful environments seems to be related to the type of new environment being experienced [53,54]. The original populations, from which the wild isolates used in our study were taken, may have previously experienced various stresses that had affected the same sets of genes as the uranium affected. Moreover, the evolutionary response of a trait to selection pressure cannot occur in an environment where the trait heritability is zero, no matter how strong the selection pressures are. [21] showed a decrease in maternal genetic variance in a stressful environment for wild sheep, although the selection strength was strongly and positively correlated to the quality of the environment. Cadmium induced a broad-sense heritability reduction in Daphnia magna at 20°C, but apparently this was mostly due to the decrease in dominance variance [55]. Nonetheless, Hendrickx et al. [56] observed a strong decrease in growth heritability in the wolf spider, Pirata piraticus, and they associated this to a decrease in additive genetic variance caused by the cadmium environment. They suggested that the expression of growth in that species involved different sets of genes in different environments. In a stressful environment, gene expression governing normal traits can be masked by other alleles involved in detoxification [1]. Indeed, for several heavy metals, including uranium, tolerance is due to one or a few major genes in C. elegans [57,58], which could mask the expression of genes for quantitative traits, such as life history traits. Nonetheless, if heritability is only hidden, but is still existent, selection may still act on it. More studies are required to confirm the effects of polluted environments on heritability reduction.

Cross-environment genetic correlations

We detected cross-environment genetic correlations between the control and uranium environments for fecundity and for late growth. This indicated a partial overlap between the genes involved in the expression of these traits in both environments [5], despite different genetic variances between the control and uranium samples. For the other measures, the 95% HPDIs largely overlapped zero, and therefore the correlation between the control and the polluted environments was probably negligible or the result may have been caused by a potential lack of power due to the number of lines in our experiment. Given that our aim was only to detect the presence of genetic correlations, rather than obtain an accurate measure of it, our method, based on the line mean, was applicable as it generally produced lower correlations than other methods, thus resulting in more conservative estimates (see [59]). All the posterior correlations with the control environment were either positive or close to zero, which implied that there was no trade-off for the measured traits between the uranium or salt environments and the control environment. Therefore, adaptation to uranium or to salt does not seem to be associated with a strong co-evolution between environments for these traits.

There was a negative genetic correlation for early growth between the uranium and salt environments, which indicated that there was a cost to adapting to the presence of these pollutants. Lopes et al. [60] found that once C. elegans had adapted to an environment polluted with a pesticide, there was no adaptation cost for reproduction when they were subsequently moved to the control environment or an environment polluted with a second pesticide. In contrast, Jansen et al. [61] suggested that there was an adaptation cost for D. magna populations that were exposed to a pesticide, and that it was conditional on the new environmental conditions. Adapted populations of D. magna suffered no costs in the control environment, but were more parasitized, which suggested that this population had a higher susceptibility to other stresses. In contrast, adaptation costs to cadmium were seen in the control environment, D. magna growth [62], and D. melanogaster fecundity [63]. The advantage of cross-environment estimates of genetic correlation, when measuring the potential adaptation cost to pollutants, is that the assessments can be performed over a short period of time compared to other approaches, such as experimental evolution. The absence or presence of an adaptation cost to pollutants revealed by this method can be confirmed by direct multigenerational experiments investigating adaptation to uranium and salt.

Effects on genetic correlation structure

The strong genetic correlation for the control environment indicated a co-evolution between growth and fecundity (Table 2). Genotypes that grew faster before sexual maturity (early growth) were more fertile, but had a reduced late growth. This evolutionary strategy is generally shown by small, short-lived organisms, such as C. elegans (r-selection [6,64]). C. elegans produces juveniles as soon as possible [65]. Sperm production takes place during the larval stages and then stops at the molting stage, which produces the adults. The remaining germ line cells generate exclusively oocytes until the end of the egg-laying period [66]. Investment in growth during the costly oocyte production period directly affected fecundity [67]. Consequently, the likely outbreeding depression caused by hybridization between diverse wild isolates [68], to create the populations used in our experiments, may be the cause of its low fecundity (about 174 larvae in the control population, Table 1) compared to other populations of C. elegans [65]. However, outbreeding depression may have happened when the wild isolates were crossed and did not break down the genetic structure of the study population [33].

There were a number of genetic correlations for the populations raised in the salt environment and they covered the same combination of traits. When compared to the control, the changes in sign for correlations involving late growth in salt could be explained by the considerable developmental delay in this environment. Unlike the other environments, most individuals started laying eggs after 72 h, and even after 144 h. We could therefore say that early and late growth in the control and uranium environments corresponded to early growth in the salt environment and that the genetic structure between growth and fecundity was close to that of the control environment. In contrast, the genetic correlations in the control became zero in the uranium environment, especially the values for fecundity. Interestingly, in the uranium environment, the phenotypic correlations were similar to the control environment because there were significant environmental correlations (see S2 Table). Previous research has shown that the environmental correlations can have opposite signs to the genetic correlations, particularly when considering the acquisition and allocation of resources for both traits [6,69]. This could also explain the presence of highly significant environmental correlations, in the two polluted environments, where individuals have probably suffered from resources acquisition and allocation effects [25,27,28], compared to the control. This highlights the importance of looking at the genetic structure, even though the phenotypic structure might be similar in different environments. The genetic structure may have broken down, which would lead to a different evolution of life history traits [6,70–72].

Tolerance mechanisms that are governed by gene expression ensure the survival of individuals under highly stressful conditions, such as pollution, but they also affect the expression of genetic structure [1]. Shirley and Sibly [63] analysed the processes by which D. melanogaster increased its tolerance to cadmium. They suggested that the links between genes involved in fecundity, survival, and development period in a control environment may be disrupted due to the addition of tolerance genes that are expressed in the presence of cadmium. It is also the case for the natural birch-feeding insect, Priophorus pallipes. The genetic correlation between body size and development time was positive when foliage quality was good and negative or low when the resource quality was close to zero [73]. Evolutionary trajectories depend on the effects of environmental conditions on organisms [19]. Hence, even if selection experiments in polluted environments mean that we need to consider the temporal heterogeneity of the G matrix, e.g. the emergence of trade-offs [12,74,75], this study showed that the genetic structure had changed when individuals were subject to sudden pollution. In contrast, a change in the genetic structure in the uranium environment by selection on heterozygosity, for example, is hardly possible given that mortality was similar in the uranium and the control environments. There were also different degrees of change in the uranium and salt environments, which was probably dependent on their mode of action. Consequently, the evolution of the G matrix was already affected from the moment pollutants appeared in the environment. This is why it is important to investigate why some pollutants or specific genetic conditions promote G matrix stability or instability.

Relationship between heritability and fitness

The decrease in heritability was higher in the uranium environment than in the salt environment (Table 3 and Table 4), although the salt concentration used in our study had a greater effect on individual fitness (Table 1). Consequently, changes in heritability were more complex than a simple positive relationship between stress effects and fitness reduction. The heritability decrease in polluted environments may also depend on how interference disrupts gene expression [3]. However this needs to be investigated by studies that focus on heritability in various polluted environments at different pollutant concentration ranges.