In recent years it has come to attention that studying the genetic architectures of developmental adaptive traits and their variations through ontogeny and phylogeny is essential to understand evolutionary dynamics (Mackay 2001, Hansen 2006). However, despite the recognition of the complexity in the relationship between genotype and phenotype for these traits, in which phenotypic plasticity, robustness and modularity at different levels hold important implications for the maintenance of genetic variability and for evolutionary dynamics, most genetic analyses devote little attention to these issues. Here we focused on Developmental Time, preadult Viability and Pupation Height in D. melanogaster, which contribute to preadult fitness. We performed genetic analyses considering their phenotypic variability, plasticity and within-line canalization for two rearing temperatures in different stages along ontogeny, in order to understand the developmental processes behind the observed variation and the complex relationships between these fitness components. Although previous work allowed to analyze phenotypic variability for these traits in natural populations or their responses to artificial selection (Casares and Carracedo 1987, Welbergen and Sokolowski 1994, Chippindale et al. 2004, Yadav and Sharma 2014, Narasimha et al. 2015), and some studies have been carried out to study their genetic bases (Mensch et al. 2008, 2010; Riedl et al. 2007, Petino Zappala et al. 2018, 2019), these are, to our knowledge, the first GWA analyses for these traits performed on lines derived from a natural population, providing a more representative sample of natural genetic variability (Rockman 2012, Gasch et al. 2016). Moreover, our analyses also assessed pre-adult fitness traits in different environments and took into account within-line canalization patterns, while simultaneously decomposing preadult traits into their larval and pupal components.
We detected significant phenotypic variability for these traits and a strong genotype-environment interaction in this subset of lines of the DGRP population; a substantial amount of this variation could be explained by genetic factors. We also detected a reduction in phenotypic robustness at 17°C relative to 25°C, which was significant for DT traits and PV. It may be noted that although 17°C cannot be considered as a stressful temperature for the species, this finding was expected for a population adapted to a subtropical climate, which could have evolved epistatic interactions favoring a canalization of adaptive traits around phenotypic optima at higher temperatures. In this sense, a change in the genetic networks at lower temperatures would imply the loss of this buffering. The apparent uncoupling between the genetic architectures of different traits, their plasticity and within-line canalization and also their temperature and stage specificity, which was previously observed in these and other lines derived from natural populations (Lavagnino and Fanara 2016, Mensch et al. 2010, Petino Zappala et al. 2018, 2019, Ørsted et al. 2018), led us to further characterize the nature of genetic variation behind these findings.
By means of GWA Studies in DGRP lines, whose genomes are sequenced, we found 924 polymorphisms and 333 genes as candidates for underlying phenotypic variability for all traits’ means and CVEs at both temperatures. It is worth noting that, even when several candidate polymorphisms were found for most traits, the number of assessed lines limited the detection of candidate variants to those with a MAF of 10% or higher; less common variants would have a greater phenotypic effect but were undetectable with the sample size used. Indeed, we found that alleles with a lower MAF were associated with greater effects, in this case corresponding to longer LDT, lower viabilities and higher CVEs for all traits. The latter was also found by Harbison et al. (2013) in a GWAS for sleep traits; the presence of decanalizing alleles in low frequencies can be explained because some individuals bearing them would present the same phenotype as those carrying the more frequent alleles, allowing them to escape the effects of stabilizing selection. However, compared with viability and PH, decanalizing alleles showed lower effect sizes for DT traits, as expected for characters under stabilizing selection (Gibson and Wagner 2000, Gibson 2009). Regardless of this limitation, the focus of our work lies in characterizing how the genetic bases of pre-adult traits vary according to environmental conditions, ontogenetic stage and for different components of phenotype; therefore, we do not claim to have fully characterized the genetic architecture of the traits assessed.
Most of the candidate variants found were found in noncoding regions, a common result in GWAS studies in Drosophila and other organisms (Maurano et al. 2012, Harbison et al. 2013, Massey and Wittkopp 2016) which signals the importance of genetic variability affecting gene expression on phenotypic variation. Moreover, the abundance of variants located near or within the same gene on binding sites for only a few transcription factors suggests that individual variants did not affect the global patterns of expression, but entailed a temporally or spatially restricted effect. This modular characteristic of transcription factor action is already a recognized phenomenon (Wagner et al. 2007) which could also explain the independent effect of different loci within the same gene over different traits that we have found in this work.
The phenotypic analyses and GWA studies performed here point towards a complex and contingent genetic basis for developmental adaptive traits: as was suggested by the low cross-environment genetic correlations and the lack of conserved significant correlations between traits, most variants related to phenotypic variability were trait, temperature and developmental stage specific, and/or were only related to a particular aspect of phenotype (means, CVEs, plasticities for means or CVEs). Moreover, instead of a general mechanism underlying robustness for all traits, we found that genetic bases for CVEs were trait-specific, a result that is consistent with previous findings by Morgante et al. (2015).
Even when we found pleiotropic genes, the specific polymorphisms underlying variability for each trait were usually located in different regions. This phenomenon indicates a great degree of modularity at the genetic level, as distinct variants within the same genes may become selected independently. Also, their effect on phenotype may be probably affected by epistasis and the alteration of genetic networks by external or even internal conditions, which means that the same genetic variants are not necessarily expected to influence phenotype in different genetic backgrounds, environments or along ontogeny (Mensch et al. 2010, Krimsky and Gruber 2013, Chandler et al. 2017, Mackay and Huang 2018, Varón-González et al. 2019). Phenotypic plasticity, within-line decanalization of phenotypes at 17°C and decoupling between larval and pupal traits found here serve as examples.
Interestingly, when analyzing variants that could account for the decoupling of larval and pupal traits, widespread antagonistic pleiotropic loci were found. While almost all of the candidate loci obtained at 25ºC were associated with longer larval and shorter pupal DTs and with lower larval and higher pupal viabilities, the effects of the candidate loci affecting both traits at 17ºC were not consistent in sign and more pronounced.
The study of the functional architecture of developmental traits could also explain how they affect each other. Statistical correlations between the developmental adaptive traits assessed here have been found in several occasions (Casares and Carracedo 1987, Chippindale et al. 2004, Narasimha et al. 2015), and some were also observed here for the DGRP lines. However, it has come to attention that relationships between adaptive developmental traits are complex, statistical correlations do not hold universally (Chippindale et al. 2003), and according to our work, they may vary within the same population, depending on the environmental factors and ontogenetic stages. According to enrichment analyses, changes in the genetic networks affecting processes related to Central Nervous System development accounted, at least in part, for the phenotypic variation found in these lines. Several genes related to pathways (e.g. Insulin Receptor, Ecdysone Receptor, Octopamine Receptor or Short Peptide F Receptor pathways) controlling resource acquisition and/or allocation (i.e., metabolism, feeding, growth) were also obtained as candidates for all traits. Moreover, some of the candidate genes have been reported as pleiotropic in regard to the pathways they participate in, which raises the possibility that they mediate crosstalks between different cascades and coordinate biological functions. These complex genetic and environmental factors affecting the genetic networks behind developmental processes would explain phenotypic variation in all the traits considered but also the intricate relationships found between them.
Our results entail significant implications concerning evolvability for adaptive developmental traits. Modularity, either at the genetic, ontogenetic or anatomical level, limits the loss of genetic variability and favors adaptive radiation (Yang 2001, Hansen 2006, Mäkinen et al. 2018). Given the disparate behaviors and processes that characterize developmental stages and the varying challenges posed by different environments, the possibility of a somewhat independent evolution for traits corresponding to distinct periods of ontogeny in response to environmental heterogeneity would imply a greater phenotypic flexibility in the face of changing conditions and limit the loss of genetic variability (Yang 2001, Del Pino et al. 2012, Petino Zappala et al. 2018). In this sense, even when loci presenting antagonistic pleiotropy across ontogenetic stages were found, they were temperature specific; such trade-offs are expected to maintain natural phenotypic variability in the context of heterogeneous environmental conditions and epistatic effects (Huang et al, 2020). So does the decoupling between the genetic bases of means, plasticities and within-line canalization for adaptive traits when phenotypic optima vary across environments. If these aspects of phenotype are affected by different genetic variants which can evolve independently, different combinations of them may arise and become selected as ecological strategies as populations deal with environmental heterogeneity (Simons and Johnston 1997, DeWitt and Scheiner 2004, Simons 2014, Dewitt 2016).
All these factors favor the maintenance of genetic variability within natural populations and thus increase evolutionary potential (Hansen 2006, Schlichting, 2008). We therefore argue that, in order to understand the evolutionary dynamics underlying biological diversity, forthcoming studies should account for these phenomena emerging from the genetic architecture of developmental adaptive traits.
Our findings suggest that the search for specific variants or genes universally associated with complex traits or diseases may be futile. However, focusing on the genetic networks or processes in which candidate genes are involved would probably provide a more representative picture which could be extrapolated to different contexts. Our results support this hypothesis, and future works on lines derived from other natural populations could confirm it if the same genetic networks and developmental processes also underlie phenotypic variability for these traits through the effects of different genetic variants.