Do traits separated by metamorphosis evolve independently? Concepts and methods

Abstract

Despite the ubiquity of complex life cycles, we know little of the evolutionary constraints exerted by metamorphosis. Here, we present pitfalls and methods to answer whether animals with a complex life cycle can independently adapt to the environments encountered at each life stage, with a specific focus on the microevolution of quantitative characters. We first discuss challenges associated with study traits and populations. We further emphasize the benefits of using a combination of approaches. We then develop how multivariate methods can limit several issues by revealing genetic patterns that are invisible when only considering trait-by-trait genetic correlations. Finally, we detail how Lande's work on sexual dimorphism can be applied in measuring G matrices across life stages. The methods and tools described here will contribute towards building a predictive framework for trait evolution across life stages.

1. Phenotypes separated by metamorphosis

Many phenotypic traits are stable over animals' adult lifetime; for example, body size in birds, insects or eutherian mammals only marginally varies between maturity and death [1]. Contrastingly, juvenile characters are not always good predictors of the late-life phenotype. Lack of positive correlation between the early and late trait values may be due to developmental noise and other environmental effects, or because traits are controlled, at least partially, by different genes. The contribution of genetic factors to phenotypic correlations is measured by genetic correlations. When two traits are genetically correlated, artificial selection on one trait causes a change in the correlated trait, even if the latter trait was not subject to selection [2,3] (figure 1). In a pioneer study, Cheverud et al. [4] measured genetic correlations between morphological traits across time in lines of rats that were artificially selected according to their body weight gain, as juveniles or as adults. First, they found that genetic correlations existed between the different weight measurements during the lifetime of the rats. Interestingly, genetic correlations were stronger between measurements that were closer in time than between early and late stages of life. They also found higher genetic correlations between measurements that were closer in time for tail length, a trait genetically correlated to body weight that indirectly evolved from the selection on body weight. Furthermore, Atchley & Rutledge [5] found on the same artificially selected lines that the genetic correlations between some pairs of traits decreased during lifetime (tail length with body weight and with chest circumferences) while others remained constant across life (chest circumferences with body weight). Those genetic correlations were caused by pleiotropy and they determine the limits within which ontogeny can evolve [4].

Figure 1.

Illustration of the effect of genetic correlations on the evolution of the trait under selection and that of the correlated trait.

In most animals, in taxa as diverse as parasitic worms [6], aquatic invertebrates [7], arthropods [8] or amphibians [9], juvenile and adult stages are separated by a major development event: metamorphosis. In complex life cycles (CLC), metamorphosis separates discrete life stages that often exhibit morphologies so distinct that traits expressed in an early stage may have no equivalent in a later stage. For example, tadpoles harbour gills and fins while adult frogs have lungs and legs. Is metamorphosis an adaptation that reduces genetic constraints on ontogeny? According to the adaptive decoupling hypothesis (ADH) [10]—the prevailing adaptive explanation for the pervasiveness of CLC—metamorphosis facilitates the independent evolution of phenotypes expressed at different times in life, and therefore maximizes adaptation to their respective environments.

In opposition, if two traits separated by metamorphosis are genetically correlated, the operation of selection on one of the traits in one life stage alone generates an indirect response of the correlated trait in the other stage (figure 1). If genetically correlated traits are differentially selected through an individual's life, genetic correlations create antagonistic pleiotropy, where alleles have beneficial effects on some trait at a specific time of life, and correlated detrimental effects on another trait or at a different life stage [11] (figure 2). Antagonistic pleiotropy was first proposed to explain the evolution of senescence: alleles that confer a reproductive advantage at an early age at the price of senescence later in life would be favoured by natural selection [11].

Figure 2.

Example of unresolved antagonistic pleiotropy (a) and resolved antagonistic pleiotropy (b) between two life stages in a unique population (e.g. larva and adult). Coloured areas represent the fitness functions of each life stage. The white area shows the phenotypic distribution of the population, which represents a maladaptive compromise (a) or an adapted population where trait values are optimized in each stage (b). (Adapted from an illustration of intra-locus sexual conflict in [12]).

Genetic correlations act as evolutionary constraints, as they determine the directions where ontogeny and phenotypes can, or cannot, evolve; genetic correlations limit the ability for phenotypes to access some dimensions in the fitness landscape [2,3,13]. Nevertheless, genetic correlations do not necessarily impede adaptation, and they may even enable faster adaptation when selection on correlated traits is aligned with the genetic correlations [14]. Furthermore, if two traits are selected in the opposite direction to their genetic correlation, it may be possible for them to evolve independently, by forcing the genetic correlation to evolve itself [15] (figure 3b).

Figure 3.

Illustration of how natural selection can affect variance and covariance structures. (a) Natural selection depletes genetic variance in such a way that negative covariance between traits expressed in different stages dominates. This scenario is most likely when traits are fitness components under directional selection for greater value. (b) Natural selection on traits 1 and 2 is opposite to the genetic correlation and creates positive covariance between both traits.

Whether genetic correlations on traits expressed at different life stages correspond to evolutionary constraints depends on whether the direction of selective pressures at the different life stages is concordant or antagonistic to the genetic correlations [16,17]. Although some traits may be expected to be under concordant selection at several stages (e.g. resistance to insecticides or to temperature stress), the striking morphological differences between larval and adult environments suggest that different stages of life have different optimal trait values [16]. Additionally, the adult stage has the special property of being selected for a specific task: reproduction. Thus, selective pressures are likely to be different at different life stages. The differences in fitness landscapes at different life stages, although crucial to understand the evolutionary constraints of metamorphosis, are not the subject of this paper. Methods to measure selection on different components of fitness, such as larval survival and adult reproductive success, have been greatly described elsewhere [18–21], and we encourage these methods to be applied to the particular case of selection pressures applied at different life stages. Here, we describe the empirical challenges faced when investigating the evolutionary coupling or independence of traits separated by metamorphosis, and we describe the methods available to solve it. Genetic constraints on the evolution of quantitative traits in animals with CLC are poorly understood. Unravelling genetic correlations between traits separated by metamorphosis is crucial as they affect adaptation in animals with CLC [22] and ecological interactions with community members (e.g. [23,24]).

2. Empirical studies of genetic constraints between life stages

In most animals with CLC, each life stage inhabits a different environment. Evolutionary constraints between stages hence determine whether they can independently adapt to their specific habitat. At the macro-evolutionary scale, the contrasted morphologies (e.g. limb numbers, body plan) exhibited by successive life stages demonstrate the evolutionary lability of discrete (i.e. qualitative) characters permitted by metamorphosis. For example, mosquito larvae do not carry cumbersome wings and blood-sucking mouth parts, and adult mosquitoes do not harbour fins [25]. Thus, the presence and absence of discrete traits are largely congruent with independence between life stages. Nevertheless, comparisons between species with and without CLC within a taxon suggested that life-cycle complexity may also be a source of constraints: for example, loss of metamorphosis and cycle simplification associated with morphological diversification were estimated from quantitative (i.e. continuous) trait variation in salamanders [26].

Although key to understand constraints on short-term and local adaptation, little is known about the extent of the evolutionary independence of the different life stages of animals with CLC at the micro-evolutionary scale. When similar characters can be measured in distinct life stages, it is currently unknown whether to expect a correlation between trait values. Tests of the genetic independence of quantitative traits separated by metamorphosis produced variable results. On the one hand, it has been demonstrated in a variety of taxa that traits were genetically correlated between life stages. For example, the experimental evolution of trematode parasites, which infect snails and mammals as intermediate and definitive hosts, revealed indirect evolution of fitness components at a non-selected stage [24]. Similarly, yellow fever mosquito lines that were selected for early or late pupation revealed a significant negative genetic correlation between the seemingly unrelated traits of speed of larval development and immune response in adults [27]. One of the most comprehensive quantitative genetics studies to date, led by Aguirre et al. [28], found support for antagonistic genetic coupling for viability across four life stages in the marine invertebrate Ciona intestinalis. The situation is reversed in Pacific tree frogs as genetic correlations for locomotory traits are positive between the larval and the adult stage [29].

On the other hand, some traits that seemed to have similar functions at different life stages were found to be genetically independent. Selection on larval and adult ability to resist extreme temperatures showed these traits are independent in Drosophila buzzatii flies [30] and in the butterfly Bicyclus anynana [31]. Phenomenological independence between temperature–stress resistance in insect larvae and adults is congruent with the recent discovery that, at the genome level, different combinations of genes affect cold hardiness in larvae and in adults in Drosophila melanogaster [32]. Even functionally related traits measured in the same organism can exhibit very different levels of genetic correlation. In Drosophila melanogaster, the expression of an antimicrobial peptide (Diptericin; an immune component) shows a positive genetic correlation between life stages while the expression of another one (Drosomycin) in adults is largely independent of that of larvae [23].

Beyond anecdotal measures on a few arbitrary traits, we do not know whether genetic correlations constrain traits across life stages and, if pleiotropy is widespread, whether traits are more often negatively or positively correlated. However, it is possible to find other examples in evolutionary biology where genetic correlations were measured between different phenotypes that share the same genotypes (box 1). In particular, there are a number of studies measuring the inter-sexual genetic correlation of male and female traits. In the case of CLC, because different phenotypes (e.g. larval and adult) are expressed in the same individuals, all methods developed in other fields cannot directly be used in our case of interest. There are, however, many tools and experimental designs that can be applied to investigate pleiotropy among life stages. Below, we discuss experimental designs and data analyses that can resolve when and how much traits expressed across life stages can evolve independently.

Box 1. Genetic constraints between alternative phenotypes sharing a common genome.

A unique genome can develop into alternative organisms with different sets of traits (e.g. males and females, larvae and adults), which we call distinct phenotypes. Arguably the most studied case of separate phenotypes sharing genetic correlations are males and females in species where sexes are separated (i.e. gonochoric), both in plants and animals (see review in [33]). Here, separate male and female phenotypes share a common genome in all autosomes, potentially leading to intra-locus sexual antagonism when alleles beneficial in one sex are detrimental in the other sex (figure 2 with stage 1 and 2 corresponding to male and female). However, in the case of inter-sexual correlations in gonochoric species, the common genotype is expressed in different individuals with different fitness values (male fitness and female fitness), while, in the case of CLCs, total fitness depends on a unique individual that must get through all life stages. In addition to sexual dimorphism, there also are other cases of different phenotypes expressed by a common genotype that are relevant for the study of phenotypes expressed at different life stages.

Simultaneous hermaphrodites express a unique adult phenotype that ensures both male and female functions. Similarly to genetic correlations between different life stages, genetic correlations between male and female functions in hermaphrodites affect various components of an individual's fitness [34]. Mutations with pleiotropic effects that are antagonistic between sexual functions or life stages are selected in a similar way: they spread if the net fitness effect of the mutation is positive on the total fitness, all else being equal [35]. In sequential hermaphrodites, male and female phenotypes are expressed one after the other, thus presenting similar conditions to larval and adult phenotypes. In particular, an individual decision must be made about the timing of both the change of sex and metamorphosis [36]. However, whether hermaphrodites are simultaneous or sequential, if an individual invests all its resources in one component of fitness (male or female reproductive success), it would retain some total fitness. The mean success of an allele is the mean of its effects on male and female fitness (as in gonochoric species). By contrast, in the case of components of fitness at different life stages, mutations that benefit larvae but forbid adults to reproduce or increase adult reproductive fitness at the expense of larval survival cannot persist. A unique individual must get through all life stages to have any fitness; thus, the average success of an allele is close to null if it is very deleterious to any life stage, even if it provides great benefits to other life stages.

Social insect colonies contain discrete ‘castes’ specialized in contributing to distinct components of fitness: breeding individuals increase reproductive success while non-breeding phenotypes secure resources and defence for the colony. The dissociation of reproductive and growing functions in separated caste phenotypes is similar to the time dissociation between larval and adult stages. Because the breeding individuals produce all the phenotypes of the colony, the breeder's genotype is used in several phenotypes, potentially leading to caste-antagonistic pleiotropy (figure 1, with stage 1 and stage 2 corresponding to breeding and non-breeding castes [37]). Both breeding and non-breeding castes are necessary for the survival of a colony, like the necessary viability of all life stages; hence, the opportunity for alleles with detrimental effects on a caste/life stage to spread are limited, whatever their effect on other castes/life stages. Similarly to the growing attention towards potential for antagonistic pleiotropy in species with different life stages, there is growing attention towards the potential for antagonism between castes [38,39], yet more investigation is needed to improve our understanding of potential antagonisms or independence between phenotypes that share a unique genotype, and whether and how they can resolve.

Potential mechanisms to resolve antagonism

Genetic correlations can evolve quickly [40], and even negative inter-sexual genetic correlation for fitness could disappear in a few hundreds of generations [41]. Resolutions of antagonistic pleiotropy include all mechanisms where different phenotypes can reach their own fitness optima independently of other phenotypes, reducing genetic correlations between fitness components. Simultaneous hermaphrodites are constrained in their potential to reach a resolution as the male and female functions are not separated physically nor by time [34].

Gene duplication followed by phenotype-specific gene expression and adaptation of the paralogues to the different phenotypes can resolve some cases of antagonistic pleiotropy described here [38,42]. A potential example or conflict resolution through gene duplication and independent specialization comes from Drosophila melanogaster, where seven drosomycin genes are expressed at different life stages [43]. Other mechanisms that can resolve antagonistic pleiotropy differ according to the genetic architecture of the antagonistic phenotypes and whether phenotypes are physically separated or separated by time. Phenotypes that are physically separated (social insects and gonochoric species) can get alternative genetic imprinting to express the most beneficial alleles in each phenotype [44,45]. Phenotypes that have different chromosomal constructs (haplodiploidy in hymenoptera; sex chromosomes in numerous gonochoric species) can also express phenotype-specific dominance to increase their respective fitness optima [46]. Finally, separated phenotypes could resolve antagonistic pleiotropy with gene expression modifiers that would enable an optimal gene expression for each phenotype [47]. In particular, alternative gene splicing could reduce antagonism [48].

3. Experimental designs: necessity to combine population designs

Well-planned experimental designs need to include adequate breeding designs and populations, control the environment(s) and carefully choose the measured traits. A potential issue in measuring genetic correlations across life stages may arise because of the sequential nature of life cycles (e.g. larvae precede adults). Traits expressed in late stages depend on resources acquired in earlier stages, and therefore on the adequacy of the early phenotype with the early environment. The sequential nature of the studied phenotypes is a particular case of the ‘general vigour’ or ‘nutrition’ problem—an illustration of GxE interaction for traits correlated with fitness—that can conceal antagonistic pleiotropy between traits that depend on resource acquisition [16,49]. In addition to such resource acquisition effects, the morphology of adult organs may be influenced by larval environment (e.g. [50,51]) for the simple reason that the development of later-stage traits often begins before metamorphosis (e.g. [52]). Similarly, larval development may depend on resources deposited by adult parents [53]. Because larvae and adults create the environment of other life stages, it is important to study correlations between phenotypes at the genetic level, choosing experimental designs that limit environmental effects, including environmental effects of the different life stages on each other. A second issue is that long-established populations show the residual V_A and CoV_A after the action of selection (figure 3a). Alleles that cause positive genetic covariance (CoV_A) between fitness components across life stages are expected to be either quickly fixed if they are beneficial or discarded from the population if they are deleterious (figure 3a). Alternatively, variants causing negative genetic correlation between components of fitness at different life stages are more likely to retain genetic variation [46]. Although the validity of this intuitive model can be debated [3], it illustrates how selection can modify the standing genetic variance and covariance of long-established populations.

Solutions exist to lower the filtering effects of selection on study populations. Mutation accumulation lines—inbred lines maintained in selection-relaxed environments so that they only differ by newly arisen mutations—enable testing newly appeared V_A and CoV_A before the action of selection [54]. An alternative solution may be to study evolutionary trajectories in new environments where some traits will exhibit large genetic variation that did not show in ancestral environments, following the release of cryptic genetic variation [55] (figure 4). However, gene-by-environment (G × E) interactions, and therefore positive CoV_A due to resource acquisition variations (i.e. general vigour/nutrition), cannot be accounted for in mutation accumulation lines or in populations in a novel environment. Artificial selection experiments are the best-known tool to test for antagonistic pleiotropy (figure 1) with lower artefacts due to resource acquisition bias, because the response to selection on a correlated trait is the most direct measure of CoV_A between traits [49]. However, artificial selection mainly operates on the available V_A of the starting population, and if this V_A is depleted due to previous selective episodes, low response to selection will be observed. Thus, there is no single design that can directly test the generality of coupling between life stages. A combination of mutation accumulation lines and artificially selected lines can be used to test the extent of antagonistic pleiotropy between traits separated by metamorphosis.

Figure 4.

Illustration of how phenotyping a population in a new environment can release cryptic genetic variation and therefore increase the amount of genetic variance expressed. In the above example, covariance between traits A and B in both the original and new environments is negligible. This may not always be the case; for example, cryptic genetic variation may be pleiotropic and produce a genetic correlation between traits. (Online version in colour.)

4. Choice of traits and the benefits of multivariate approaches

In addition to the choice of population genetic structure, the choice of traits on which to measure genetic variance and covariance is critical. Quantitative traits fall in two categories: ‘components of fitness', also called life-history traits, and ‘metric traits’ [2,16]. Components of fitness are distinct from metric traits as they are always positively correlated with total fitness when all other traits are held constant; thus, by definition components of fitness are directionally positively selected. For example, if all other traits are held constant, an increase in larval survival causes an increase in fitness. Components of fitness vary themselves because of the underlying metric traits, which can be either positively or negatively selected. For example, an increase in body size is not synonymous with an increase in fitness; this correlation is only valid if body size increases a component of fitness such as survival or reproductive success [16]. Genetic correlations between metrics traits and between components of fitness have different evolutionary consequences. Although subject to debate [3], components of fitness may be more likely to be depleted in V_A and positive CoV_A (figure 3a), and therefore to share negative CoV_A (for example, due to functional trade-offs), than metric traits [56,57]. Fitness components are therefore potentially poor candidates to study the genetic architecture of traits separated by metamorphosis. By contrast, there are no expectations for the evolution of CoV_A in metric traits, whether they are directly correlated with fitness or not, because it depends on past and current selection pressures at different life stages. Nevertheless, pervasive stabilizing selection is expected to apply on most quantitative traits [58–61], also potentially limiting the available V_A. Ideally, constraints between traits separated by metamorphosis should be investigated in traits that were not under selection in recent population history, but rather at mutation–drift equilibrium. However, finding traits that are functionally and structurally independent from fitness is somewhat illusory in natural populations.

Multivariate approaches can reduce the effects of V_A depletion by revealing genetic patterns invisible when only considering trait-by-trait genetic correlations [62]. A multivariate approach enables to consider at the same time variance of multiple traits and how they covary in a unique G matrix. Intuitively, a G matrix for n traits should vary in n dimensions. However, because many traits typically covary, a G matrix of n traits usually has less than n dimensions of variance [13,63]. By diagonalizing G, it is possible to reduce the G matrix to the number of dimensions that actually vary. g_max, the first eigenvector of G, contains as much variance as its eigenvalue, necessarily exceeding or equal to the univariate variance of the trait with the maximum variance [13]. Because diagonalizing G matrices accounts for the covariance between all the traits in the G matrix, this approach removes spurious V_A that is included when considering several traits in independent univariate analyses. Multivariate approaches were successfully used in many studies that highlighted new biologically interpretable traits (see review [64]). A classic example of how multivariate analyses reveal biological functions that were not available with univariate analyses comes from a series of studies described by Blows [13]. In Drosophila serrata, females prefer mating with males showing a specific chemical signature; females exert directional selection on a cocktail of male cuticular hydrocarbons (CHCs). Univariate analyses revealed that all different CHCs contained V_A. The detection of directional selection on heritable traits suggests that a rapid evolution of CHCs should be observed. However, no such evolution occurred. Multivariate analyses revealed the evolutionary constraints limiting the evolution of male CHCs: there was almost no genetic variance available in the direction of female selection. Thus, despite univariate V_A for each CHC, genetic constraints prevented males from reaching the particular combination of CHCs preferred by females [13]. Here, the diagonalization of G was the only method to access meaningful biological variation (axes of variance that were not impacted by female choice) and revealed that the sexually selected traits did not vary. Furthermore, it is possible to consider together several traits of low genetic variance to form a unique trait that contains a higher level of genetic variance and biological significance, free from all spurious V_A due to covariances with other traits. Finally, multivariate approaches enable one to analyse together several functions and structures, and thus are a more integrative approach than two-trait studies [65].

5. New insights from older statistical methods

Testing the genetic independence of quantitative traits separated by metamorphosis requires studying genetic correlations between many quantitative traits measured in at least two life stages. G matrices allow measuring genetic covariances between several traits at the same time [66]. Lande [67] extended the G matrix framework to describe genetic covariance shared by different phenotypes, namely males and females. It is possible to apply this work to other examples of distinct phenotypes sharing genetic covariances; in our cases, different stages of life (box 1). Lande's equation applied to two life stages (here called l and a) allows the simultaneous quantification of genetic covariance between different traits measured within a life stage (traditionally contained in the off-diagonal values of G), genetic covariance between similar traits among stages and genetic covariance between different traits among stages using

$${\mathit{G}}_{\mathit{l},\mathit{a}}=\hspace{0.17em}\left(\begin{array}{cc}{\mathit{G}}_{\mathit{l}}& \mathit{B}\\ {\mathit{B}}^{\mathit{T}}& {\mathit{G}}_{\mathit{a}}\end{array}\right),$$ 5.1

where the matrix ${\mathit{G}}_{\mathit{l},\mathit{a}}$ includes four matrices ${\mathit{G}}_{\mathit{l}}$ and ${\mathit{G}}_{\mathit{a}}$ are the traditional genetic variance–covariance matrices for the different traits within life stages l and a. The B matrix (and its transpose B^T) includes in its diagonal the covariances between similar traits among stages l and a, and off-diagonal, the covariances between different traits among life stages. Thus, this method enables one to quantify genetic covariance between traits that are seemingly different, which is of particular importance in the case of measuring inter-stage correlations as it may be challenging to measure a ‘unique’ trait when life stages are morphologically different.

Simultaneous consideration of the same trait in more than two life stages can reveal more complex patterns than when only considering pairwise genetic correlations. Aguirre et al. [28] found that multiple positive and negative correlations between viability of three life stages in the marine invertebrate Ciona intestinalis could be reduced to a unique axis of variance; this means that some associations of genes underlie viability in embryo, larval and juvenile stages. Extending Aguirre et al.'s [28] approach to several traits using Lande's equation (5.1) could greatly improve our understanding of the levels of (in)dependence between life stages. However, it needs to be highlighted that an important limitation to the use of these multivariate tools is the power required to test all dimensions of a G matrix. Studying ${\mathit{G}}_{\mathit{l},\mathit{a}}$ requires almost four times as many parameters as separate ${\mathit{G}}_{\mathit{l}}$ and ${\mathit{G}}_{\mathit{a}}$ matrices (equation (5.1)); thus careful experimental planning is required.

Traditionally, studies testing evolutionary constraints between life stages determine whether there is a significant genetic correlation between similar traits separated by metamorphosis (i.e. is r ≠ 0) [23,24,30]. Significant correlations hence indicate correlated traits are not independent. However, some degree of genetic independence between traits separated by metamorphosis (i.e. |r| < 1) would already reveal relaxed constraints and the possibility of some independent evolution (figure 3b). Testing whether V_A in some life stages is independent of V_A in other life stages uses a null hypothesis of pure genetic correlation (i.e. r = 1). Most studies asking similar questions for intra-locus sexual conflict test whether genetic correlations between males and females differ from 1, and are positive or negative, using a trait measurable in both males and females (reviewed in [68], recently in [69]). Univariate tests of whether genetic correlations between different phenotypes are lower than 1 were extended to multivariate datasets, to test the genetic independence of traits in four life stages. Aguirre et al. [28] created an original multivariate test to detect whether the covariance between traits across life stages was null and whether there was a complete dependence between those traits. They simulated G matrices that retained the observed levels of variances for each trait and calculated the corresponding covariance resulting from null (all values of the diagonal of B = 0) or complete correlations between traits (all values of the diagonal of B corresponded to a correlation of 1), and compared their observed G matrices with the simulated ones to obtain a test of significance. Hence, methods are now readily available to assess how evolutionary independent are traits in different life stages.

6. Towards predicting evolutionary coupling

It is not yet possible to predict which traits separated by metamorphosis can evolve independently, largely because we lack comparable data. The first step towards a predictive framework requires standardized measures of many traits in the same organism. The chosen traits need to encompass a diversity of functions and have various degrees of effects on fitness. Ideally, measures should be carried out in model organisms to facilitate the identification of the genetic factors causing coupling. Good traits that cover a large array of functions, selective forces and genome locations are gene expression traits (i.e. transcriptomics). Transcriptomes, with costs constantly decreasing, produce thousands of quantitative phenotypes measured in a comparable unit (e.g. quantity of reads with RNAseq methods). Used in an adapted breeding design (see above), transcriptomes can reveal how the genotype–phenotype map operates, for example, by measuring pleiotropy and modularity [70] or the extent of epistasis [71] on an unprecedented number of traits. Gene ontology repositories can further be used to test the association between traits' functions and how they are constrained throughout different life stages.

To conclude, conceptual and technical tools are now available to solve a long-standing question: how evolutionarily independent are traits expressed in different life stages? Simultaneous study of numerous traits in the same organism will provide the basis for a long-needed predictive framework of evolutionary constraints between stages. A mechanistic understanding of when and why similar traits have distinct genetic bases when expressed in different stages will further be helpful to fully uncover the history of the (lack of) genetic coupling. Convergence between functional and quantitative genetics will result in an important advance in our understanding of the evolutionary ecology of organisms with CLC.

Competing interests

We declare we have no competing interests.

Funding

This work was supported by the French Institute of Agricultural Research (INRA- SPE) to S.F. and an Agreenskills + fellowship to J.C.