Stage- and thermal-specific genetic architecture for preadult viability in natural populations of Drosophila melanogaster

1 INTRODUCTION

The study of fitness-related traits provides a powerful tool to describe the organismal investment of resources in different biological functions, underlying particular adaptive strategies that depend on the ecological scenario (van Noordwijk & de Jong, 1986; Stearns, 1992). In this sense, elucidating the genetic architecture for developmental traits affecting fitness allows the characterization of the processes that lead to phenotypic diversity and determine evolutionary potential (Hansen, 2006).

Although the traditional framework for studying the relationships between genotype and phenotype focuses on traits’ means in certain environments, aspects relating to phenotypic variability are equally relevant to the understanding of developmental processes underlying ecological strategies and phenotypic evolution (Ørsted, Rohde, Hoffmann, Sørensen, & Kristensen, 2018; Paaby & Testa, 2018). One such parameter is phenotypic plasticity, the ability of a genotype to modify the phenotype according to the environment (DeWitt, 2016), which constitutes a strategy to deal with environmental heterogeneity (DeWitt & Scheiner, 2004): when phenotypic optimum for a given trait varies across environments frequently experienced by a population, plasticity for the trait is adaptive and is expected to arise (DeWitt & Scheiner, 2004). This phenomenon also implies that the effect of some genetic variants can be concealed to natural selection in some conditions, favouring the accumulation of genetic variability (Paaby & Gibson, 2016).

Phenotypic variation within environments, which has been shown to have a genetic basis (Blasco, Martínez-Álvaro, García, Ibáñez-Escriche, & Argente, 2017; Ørsted, Rohde, Hoffmann, Sørensen & Kristensen, 2018; Sørensen, de los Campos, Morgante, Mackay, & Sorensen, 2015), is also relevant to the potential evolutionary trajectory of populations. In fluctuating environments, genetic variants that favour a high phenotypic variance may maximize fitness; this phenomenon is known as “bet hedging strategy.” Conversely, in constant environments, stabilizing selection would promote the fixation of epistatic interactions promoting the robustness of phenotype to microenvironmental or genetic changes (environmental or genetic canalization, respectively [Debat & Le Rouzic, 2019; Ørsted et al., 2018]). It has been proposed that traits with a great impact on fitness are usually more strongly canalized (Stearns & Kawecki, 1994). Phenotypic canalization hides the effect of some genetic variants from natural selection, and thus, these variants are less likely to be selected against. Under subsequent genetic or environmental perturbation, a decanalization of the phenotype and the liberation of cryptic genetic variation as additive genetic variance may occur, with the possibility of rapid phenotypic evolution (Flatt, 2005; Paaby & Gibson, 2016). Although genetic canalization refers to genetic variance within a population, the degree of environmental canalization for a given trait, that is the genotypes’ ability to produce similar phenotypes within an environment, can be estimated by the environmental variation coefficient (CVE), which standardizes environmental variability according to phenotypic means. Estimating CVEs allows for the comparison of environmental canalization between different populations or even different traits. Environmental canalization may be plastic, meaning the phenotypic robustness across environments for a given genotype is not constant (Ørsted et al., 2018). Phenotypic plasticity, canalization and the plasticity of canalization for a trait emerge from its genetic architecture and are important for the maintenance of genetic variability, invasiveness and evolvability (DeWitt, 2016; Gilchrist & Lee, 2007; Ørsted et al., 2018; Paaby & Gibson, 2016). Relationships between phenotypic means and these components of environmental variation are apparently complex, and they may depend on the assessed trait and the range of environments (Blasco et al., 2017; Harbison, McCoy, & Mackay, 2013; Morgante, Sørensen, Sorensen, Maltecca, & Mackay, 2015; Reed et al., 2010; Sørensen et al., 2015).

Preadult viability, the proportion of individuals reaching maturity, is a trait with a direct effect on fitness (Fowler, Semple, Barton, & Partridge, 1997; Futuyma, 1998) and is correlated with other fitness-related traits, such as developmental time, pupation behaviour or resistance to different stressful conditions (Casares & Carracedo, 1987; Chippindale, Chu, & Rose, 1996; Folguera, Ceballos, Spezzi, Fanara, & Hasson, 2008; Prasad, Shakarad, Anitha, Rajamani, & Joshi, 2001; Prasad et al., 2000). Differences among and within populations reared in the same environmental conditions reveal genetic variation for this trait (Folguera et al., 2008; Gardner, Fowler, Partridge, & Barton, 2001; Horváth & Kalinka, 2016; Mackay, 1986; Rodríguez-Ramilo, Pérez-Figueroa, Fernández, Fernández, & Caballero, 2004) that can be maintained by genotype-environment interaction (Fanara, Folguera, Fernandez Iriarte, Mensch, & Hasson, 2006; Horváth & Kalinka, 2016).

The traditional framework evaluates preadult viability as the survival from the first instar larva until the emergence of the imago. However, both larval and pupal ontogenetic stages present anatomically, physiologically and behaviourally dissimilar characteristics, so that they could be analysed as distinct modules. It has been proposed that the modularity between life stages in holometabolous insects results in a greater intra- and interspecific diversity (Minelli, Brena, Deflorian, Maruzzo, & Fusco, 2006; Yang, 2001). Furthermore, modularity at different levels (e.g. genetic, anatomical) would restrict the impact of changes in one module over the rest, thus preventing the production of trade-offs (patterns of negatively correlated fitness effects) that act as evolutionary constraints (Chippindale et al., 1996; Hughes & Leips, 2017).Therefore, modularity would generally favour the maintenance of genetic variability and increase evolvability (Hansen, 2003; Hill & Zhang, 2012; Wagner & Zhang, 2011) and should be accounted for when studying developmental fitness traits (Bolker, 2000; Hoekstra & Coyne, 2007). For that reason, the decomposition of preadult viability in its larval and pupal components would be of interest not only to achieve a higher resolution when describing phenotypic variability but also to provide relevant information on the degree of modularity at the ontogenetic level.

In this work, we expect to characterize phenotypic variability for preadult viability and to demonstrate the hypothesis that its genetic architecture can change in response to environmental conditions and ontogenetic stage. Furthermore, we wish to determine whether the different aspects of phenotype behave independently and therefore can respond to different selective pressures and generate complex ecological strategies. With these objectives, we analysed the variability for Larval and pupal viability (LV and PV) means, their plasticity and canalization patterns at two different rearing temperatures (17° and 25°C) in isogenic lines derived from three natural populations of Drosophila melanogaster. This setting allowed us to i) test if there is phenotypic variability for viability traits (means and variances) between and within populations; ii) estimate the weight of genetic and environmental components underlying this variation; iii) test if the genetic architecture for viability changes along ontogeny; and iv) test if the genetic bases for means and environmental variances are decoupled.

Our results show differences between populations for both traits, their canalization patterns and plasticities. The genetic bases for these different aspects of phenotype were apparently decoupled, suggesting the possibility for their independent evolution and the subsequent emergence of distinct ecological strategies. A genotype-environment interaction was observed for LV and PV in the three populations, with a change in genotypic ranking across environments, contributing to the maintenance of genetic variability. Finally, our results point towards a differentiated genetic basis for viability not only between populations, but also according to developmental stage and rearing temperature. We therefore argue for a comprehensive characterization of phenotypic variability, including the study of environmental variance components, to achieve a more realistic view of the ecological strategies and evolutionary dynamics followed by natural populations and to understand changes in their ability to evolve.

2 MATERIALS AND METHODS

2.2 Viability assays

For each line of each population, large quantities of first instar larvae were obtained by placing batches of 100 pairs of sexually mature flies into egg-collecting chambers with Petri dishes containing an agar and yeast egg-laying medium (three chambers for each line). After larvae started hatching, batches of 30 first instar larvae were transferred from the Petri dishes to culture vials containing 5 ml of cornmeal–agar–molasses; this density prevents crowding effects. Vials from each line were kept in an incubator at two temperatures: (17 ± 1)°C and (25 ± 1)°C, respectively, representing the mean temperatures for summer months in the highland and lowland locations considered, under a 12-hr light : 12-hr dark cycle and at 70% humidity. We set up 4–5 replicates for each line–temperature combination, representing a total of 952 vials. Every 24 hr, new pupae were counted and emerged flies were collected.

For each replicate, LV was estimated as the percentage of pupae formed from the total of first instar larvae initially seeded (30), whereas PV was estimated as the proportion of hatched adults from the number of formed pupae. We also calculated, for each line–temperature combination, CVEs for viability traits as the ratio between standard deviation and average for the replicates.

Phenotypic plasticities for all traits’ means corresponded to the quotient of the means per line at 17°C and 25°C, whereas plasticities for CVEs were calculated by subtracting CVEs by line at 25°C from CVEs by line at 17°C.

The differences between populations for both viability traits were evaluated using the averages by line and temperature by means of two-way fixed model analysis of variance (ANOVAs) according to the model

where P and T are the fixed effects of population and temperature, and ε stands for the error (variance within P × T combination). The population effect tests for phenotypic differentiation among populations, whereas P × T stands for variation in thermal plasticity among populations. When P × T interaction was significant, partial ANOVAs by temperature were carried out.

The variance for viability traits within populations was evaluated using the averages by replicate, by means of two-ways analysis of variance (ANOVAs) for each population, according to the mixed model

where L and T are the effects of line (random) and temperature (fixed), and ε stands for the error (variance within L × T combination). In these ANOVAs, significant L and L × T effects are interpreted as the presence of genetic variation and genetic variation in thermal plasticity, respectively. A significant L × T effect, an estimator of the genotype-environment interaction, can arise as a consequence of differences in among-lines variance in separate environments (change in scale) and/or deviations from unity of the cross-environment genetic correlation (change in ranking order). The contribution of the two sources of variation to genotype-environment interaction was analysed by means of the equation (Muir, Nyquist, & Xu, 1992; Robertson, 1959):

where V_GEI is the genotype-environment interaction variance component, σ_17°C and σ_25°C are the square roots of the among-lines variance components at 17°C and 25°C (which were obtained after performing ANOVAs for each temperature separately), and rGxE_{(17°C 25°C)} is the cross-environment genetic correlation. rGxE_{(17°C 25°C)}, which reflects the degree in which the same genetic variants control trait expression across temperatures, was estimated for each trait as COV_{17°C 25°C} / σ_17°C σ_25°C; where COV_{17°C 25°C} represents the covariance of lines means measured at 17°C and 25°C and, σ_17°C and σ_25°C were defined above. This method for the estimation of rGxE_{(17°C 25°C)} is not equivalent to the computation of a product-moment correlation (Lynch & Walsh, 1998).

Quantitative genetic parameters were calculated for both larva and pupal viability in each population. Under our experimental design, genetic (), environmental () and phenotypic () variance was estimated as = + ; and , where , and are among-line, line by temperature and within-line variance components, respectively. Broad sense heritability (H²) for each trait was estimated as H² = /. Coefficients of genetic (CV_G = 100σ_G/mean) and environmental (CV_E = 100σ_E/mean) variance were also computed. We also estimated for each combination of population and temperature, as , where and are the among-line and within-line variance components obtained from the one-way ANOVA for each population performed for 17°C and 25°C. Correlation analyses were carried out between LV and PV within each temperature, between temperatures for each V trait, between viability traits and their CVEs and between phenotypic (mean) plasticity and CVE plasticity for each population.

All statistical tests were performed using STATISTICA 8.0 (StatSoft, Inc). Bonferroni correction for multiple tests was applied whenever results from multiple tests were combined in one final conclusion.

3 RESULTS

We scored a total of 18,322 adults derived from 21,408 pupae that developed from ~30,000 first instar larvae. The overall viability considering all populations was 61.1%, being the larval and pupal viabilities 71.4% and 85.6%, respectively. The populations did not differ significantly in the number of replicates excluded and no lines showed more than 1 replicate for each temperature with a total viability <25% (data not shown).

Correlation analyses (Table 2) revealed no significant associations between LV and PV in any of the three populations at either of the rearing temperatures. Also, our results showed no significant correlations between temperatures for any of the preadult viability traits. These results suggest that the genetic bases for LV and PV were decoupled and that different genetic factors were associated to development at different temperatures in both viability traits. Thus, we performed the analyses for LV and PV separately.

Our data showed a higher PV than LV at both temperatures with the sole exception of the Raleigh population at 17°C (Figure 1), although when considering the average duration of each stage, hourly mortality was higher for the three populations at 17°C and during the pupal stage (Figure S1). Furthermore, both mean viabilities and CVEs exhibited a greater dispersion at 25°C compared to the results observed at 17°C. On the other hand, distinct patterns were found for LV and PV, as expected by the lack of statistical correlations between both traits. We detected significant variation among populations only for LV (Table 3a), where the higher average viability corresponded to lines collected in Raleigh population. The populations’ mean viabilities responded to thermal variation differently (Table 3a, Figure 1a). Argentinean populations exhibited a similar pattern for LV means and CVEs, with higher viabilities at 17°C; LV for Raleigh population showed less plasticity. Conversely, for both means and CVEs of PV, lines from the Raleigh population were the most plastic on average, whereas those from Uspallata had the most constant phenotype across temperatures (Figure 1), in a pattern that mirrored the annual thermal amplitudes of the original locations. On the other hand, CVEs of LV did not change according to temperature (Table 3), adding evidence for the decoupling between both components of preadult viability.

To further evaluate the phenotypic differences between the three populations, we performed ANOVAs within each population to investigate whether the genetic and environmental factors affecting viability traits were homogeneous across populations and estimate their relative contributions. Results revealed significant natural genetic variation underlying variability for LV and PV in the three populations, since variation attributable to genetic factors (line and line × temperature interaction) contributed at least ~40% of the total variance (Figure 2). For all three populations and for both viability traits, the genotype-environment interaction, estimated by means of the line × temperature interaction, was always the main factor contributing to genetic variance, wherein changes in ranking order among environments accounted for at least 76% of the interaction (Table 4, Figure 3). The estimated cross-temperature genetic correlations (rGxE_{(17°C 25°C)}) for each trait and population (Table 4) serve as a measure of the extent to which the same genetic variants affect a trait in flies raised at both temperatures. This rGxE_{(17°C 25°C)} correlation ranked from −0.53 to 0.31 determining thermal-specific differences in the genetic bases of both traits in lines collected in Raleigh, Lavalle and Uspallata. This pattern implies that optimal phenotypes at both temperatures were not associated to the same genotypes, thus limiting the loss of genetic variability in heterogeneous environments. Certainly, the analyses indicate that Raleigh, Lavalle and Uspallata populations harboured genetic variation affecting all traits: differences among lines accounted for at least 22% of the total phenotypic variance within each temperature (data not shown).

To gain insights into the differences in genetic architecture between LV and PV in all populations, we calculated quantitative genetic parameters (Table 4). The data showed moderate broad sense heritabilities for LV and higher values for PV. It is of interest to note that the values for CV_G ranged from 16.07 to 27.75 (for LV) and from 9.47 to 14.09 (for PV) for the three populations, suggesting that LV had a greater potential for change under selection, at least in the conditions assayed.

Finally, negative correlations were found between viability and CVEs as well as between plasticity for means and CVEs for PV in the three populations (Table 5). These results could suggest a common genetic basis for these different components of pupal phenotype. On the other hand, no consistent relationships were found for LV means and CVEs and between plasticities for means and CVEs, although opposing patterns for these plasticities were found in the Raleigh populations with respect to the Argentinian ones (Table 6), with a higher robustness at 25°C compared to 17°C.

4 DISCUSSION

Studying the evolution of genetic architecture of developmental traits affecting fitness is essential to understand their evolutionary dynamics, including adaptation to different and heterogeneous environmental conditions (Hansen, 2006; Mackay, 2001; Mackay & Huang, 2018). Here, we used lines derived from three natural populations of D. melanogaster to characterize phenotypic variability and genetic architecture for preadult viability traits at two temperatures.

Our results revealed phenotypic differences between populations for larval and pupal viabilities, indicating changes in genetic architecture, as population-specific trends for preadult viability traits and distinct patterns of phenotypic canalization depending on the developmental temperature were found.

We also detected a high contribution of genetic components to total phenotypic variability within populations and a moderate (for larvae) to high (for pupae) heritabilities, suggesting that genetic variance for preadult viability is not exhausted by natural selection, as it would be expected for fitness traits (Falconer & Mackay, 1996; Lavagnino, Anholt, & Fanara, 2008; Mousseau & Roff, 1987). This finding can be explained by the genetic variation found for thermal reaction norms within populations (genotype-environment interaction). Indeed, several studies show that genetic variability can be maintained by genotype-environment interaction (Del Pino, Salgado, & Godoy-Herrera, 2012; Horváth & Kalinka, 2016; Satorre, Fanara, & Lavagnino, 2014). Moreover, we found that the genetic basis of each trait varied across the temperatures assessed, as the cross-environment genetic correlations for all traits in the three populations were far from unity. These results suggest that susceptibility to different causes of lethality changes according to rearing temperature, probably in response to the alteration of genetic networks involved in developmental processes, behaviour, immunity and/or resistance to different stressful stimuli. This change in genetic ranking across environments (crossover of reaction norms) would play an important role as contributor to genotype-environment interaction, implying the possibility of independent evolution of these traits at each temperature. A high contribution of changes in ranking order to the significant genotype-environment interaction has also been found for preadult viability in different studies performed with several Drosophila species considering diverse environmental factors (Dobzhansky & Spassky, 1944; Fanara et al., 2006; Horváth & Kalinka, 2016; our results). Changes in ranking order imply that no genotypes are associated to phenotypic optima across temperatures; given that natural populations face continuous temperature changes, this fact partly explains the maintenance of a high level of genetic variability associated to the trait.

Although Drosophila development encompasses successive developmental stages characterized by different biological processes and behaviours, in most population genetics and evolutionary studies, preadult stages are not differentiated to evaluate potential changes in the genetic basis and factors affecting the evolution of developmental traits. Here, we detected no significant correlations between LV and PV, and found differences in variance components and quantitative genetic parameters for these preadult traits. These results suggest that the genetic basis for viability differs between stages of development, and consequently, both traits (LV and PV) are able to evolve independently. Other studies have shown decoupling between larval and pupal traits (Chippindale, Alipaz, & Rose, 2004; Chippindale et al., 1996; Mensch et al., 2010; Partridge, 1992; Partridge, Barrie, Fowler, & French, 1994; Petino Zappala, Ortiz, & Fanara, 2018; Artieri & Singh, 2010), and substantial changes in transcriptional patterns during ontogeny have also been reported (Arbeitman et al., 2002; Gerstein et al., 2014). Moreover, studies on preadult viability performed in other holometabolous insects, when accounting for ontogenetic decoupling, showed similar results concerning the stage and also environmental-specific nature of phenotypic variability (Eskafi & Fernandez, 1990; Liu, Chen, & Zalucki, 2002; Rausher, 1979; Rueda, Patel, Axtell, & Stinner, 1990); nonetheless, the decomposition of traits according to developmental stages is frequently overlooked in studies using Drosophila. Accounting for this decoupling is important not only because measuring egg-to-adult viability represents a loss of information, but also because preadult viability condenses the differential susceptibility to a wide range of stressors (including internal, e.g. developmental problems, or external, such as infections) whose incidences vary throughout developmental stages. Furthermore, modularity entails important implications for the maintenance of genetic variability and evolvability, and therefore, it should be addressed (Le Rouzic & Carlborg, 2008; Wagner & Zhang, 2011; Yang, 2001).

We should note that preadult viability is a particularly complex trait affected by disparate factors, and therefore, it could be argued that our results may be not generalizable. However, studies on the genetic architecture of developmental time and pupation height performed using equal thermal treatments on the same D. melanogaster lines confirmed their genetic bases are also temperature- and stage-specific (Petino Zappala et al., 2018). These results suggest that ontogenetic decoupling and the effect of environmental factors should be accounted for when studying the genetic architecture of other quantitative developmental traits.

Previous studies have also addressed the importance of phenotypic canalization in the maintenance of genetic variability, allowing adaptation to sudden environmental change, promoting invasiveness and overall increasing evolvability (Davidson, Jennions, & Nicotra, 2011; DeWitt, 2016; Flatt, 2005; Pigliucci, 2008), but this aspect of phenotype is also frequently neglected. Here, we detected different patterns of genetic and environmental canalization (i.e. genetic and environmental variance) in the three populations according to rearing temperatures. Generally, temperature affected both components of canalization the same way (i.e. for each population, genetic variance and mean CVE behaved similarly according to the raising temperature). This result is consistent with an interplay between, or a single cause for, both components of canalization (Meiklejohn & Hartl, 2002). For example, under stable environmental conditions, the fixation of canalizing alleles or an accumulation of epistatic interactions “hiding” genetic variability could lower genetic and environmental variance within a population. In case of genetic or environmental perturbations, the alteration of genetic networks could expose previously cryptic genetic variability and also lower environmental robustness. Therefore, our results support the hypothesis that genetic canalization and environmental canalization for preadult viability traits are under genetic control and are probably subject to natural selection.

Whether means and different aspects of environmental variance for different traits share a common genetic basis is a matter of debate. Here, we observed negative correlations between PV means and its CVEs. This finding was expected, since it has been shown that phenotypic decanalization is usually associated to developmental instability and tends to correlate with low viability (Baer, 2008; Bouclier & Biémont, 1982; Møller & Manning, 2003), which could suggest a shared genetic basis for both aspects of phenotype. However, in the case of viability traits, it can be argued that lines with high viability (and therefore near the upper limit of this variable) would necessarily present a lower environmental variance, generating a spurious negative correlation. Furthermore, the correlation is never significant when mean phenotypic values drop below 70%. This suggests that indeed, the correlation between PV means and CVEs would be an artefact caused by its mean values near the variable's upper limit. On the other hand, no conserved correlations were found between LV means and its CVEs. This finding underscores the complex relationships between variation for phenotypic means and environmental canalization for different traits (Blasco et al., 2017; Harbison et al., 2013; Morgante et al., 2015; Reed et al., 2010; Varón-González, Pallares, Debat, & Navarro, 2019). This dependence on the considered trait also holds for the relationships between plasticity and phenotypic variance within environments (Ørsted et al., 2018; Siegal & Leu, 2014; Takahashi, Okada, & Teramura, 2012; Valladares, Balaguer, Martinez-Ferri, Perez-Corona, & Manrique, 2002). That most genetic variants only have an effect restricted to a particular aspect of phenotype (i.e. means or CVEs) would prevent the loss of genetic variability due to selective pressures acting on either aspect. This implication is particularly important for traits with a great impact on fitness, in which the pressure for phenotypic canalization is stronger (Stearns & Kawecki, 1994). All in all, we underscore the need to work on a realistic characterization of genetic architecture of complex traits, including the study of environmental variance components, and assessing the possible modularity in different levels (e.g. genetic, ontogenetic, anatomical), which is essential to accurately describe the robustness and flexibility of the genetic networks maintaining diversity and underlying relevant ecological strategies and evolutionary phenomena.

Further work is needed to confirm by means of Genome-wide Association Studies whether the genetic bases for viability traits are indeed population, stage and temperature-specific, and which biological processes underlie phenotypic variability for these traits.

DISCLOSURE

The authors declare no conflict of interest.