Developmental plasticity is pervasive in nature, with many environmental factors affecting the expression of different traits in a variety of species. Emblematic examples of developmental plasticity include temperature-dependent sex determination in reptiles (Merchant-Larios and Díaz-Hernández, 2013; Mitchell et al., 2018; Noble et al., 2018), nutrition-dependent caste determination in social insects (Maleszka, 2008; Smith et al., 2008), and density-dependent production of dispersing morphs in swarming locusts (Pener and Simpson, 2009; Ernst et al., 2015) and other insects (e.g. aphids; Braendle et al., 2006).

In the study of the regulation and evolution of developmental plasticity, it is often useful to distinguish between the environmental factor(s) that can induce changes in development, hereafter called inductive environmental cue(s), and the environmental factor(s) responsible for fitness differences between induced phenotypes, hereafter called the selective environment ( Figure 2A ). The relationship between inductive and selective environments is, in fact, of fundamental importance for the evolution of plasticity (Nijhout, 2003a). In some cases, the main inductive and selective environmental factors are the same, such as with the thermally induced changes in body size that influence the thermo-regulation of Drosophila melanogaster adults (Ghosh et al., 2013; Figure 1A ). In others, the inductive cue is predictive of future environmental conditions but not the main selective agent. This is often the case with polyphenisms, which are environmentally induced alternative discrete phenotypes (see Nijhout, 2003a; Simpson et al., 2011) that are common in relation to alternating seasons (see Kivelä et al., 2013). In the polyphenism of Bicyclus anynana butterflies, for example, the temperature experienced during development determines adult pigmentation and life histories ( Figure 1A ) and anticipates seasonal changes in background foliage coverage, favoring season-specific anti-predatory and reproductive strategies (Brakefield et al., 2009; Monteiro, 2015; Beldade and Peralta, 2017). In such seasonal polyphenisms, inductive and selective environments reflect temporal heterogeneity (Brakefield et al., 2007). External factors can also reflect spatial heterogeneity, such as that associated with different levels of predation between ponds inhabited by Rana temporaria tadpoles (Van Buskirk, 2017), with the alternative environmentally induced phenotypes being produced on different locations. In other cases of plasticity, the alternative morphs co-occur. For instance, with the nutritional-dependent caste determination in social insects, the heterogeneity in adult “environment,” rather than in any external factor, exists in terms of task allocation inside the colony (Jeanson and Weidenmüller, 2014; Gordon, 2016).

Typically, one same inductive cue will simultaneously affect different traits (e.g. predator presence affects head morphology and body size in Poecilia reticulata guppies; Torres-Dowdall et al., 2012), which are often part of the same “plasticity syndrome”. In these cases, different traits can be integrated into functional suites that respond to environmental influences in a concerted manner and that are typically also selected in concert (e.g. Mateus et al., 2014; Oostra et al., 2014; van Bergen and Beldade, 2019). Such extent of integration (or, conversely, independence) among plastic traits, some of which might be adaptive, whereas others might be maladaptive, has important implications for phenotypic variation and diversification as it influences responses to selection. A classical example of correlated plastic responses is the effect that temperature has on different phenotypes, including development time (e.g. diapause), body size, and other life-history traits in many arthropods. Although diapause is thought to be an adaptive plastic response, this may not be true for correlated traits whose developmental rates are affected by the availability of energy resources (Gotthard and Nylin, 1995). It is also common that one same trait can be simultaneously affected by different environmental cues (e.g. production of the winged morph in aphids affected by tactile stimulation, nutrition, and other factors; Braendle et al., 2006). Exploring how organisms integrate information from different external cues is a topic of much current interest, as studies of plasticity start to tackle what is the typical complexity of natural environments, where organisms are exposed to multiple and highly dynamic environmental factors (e.g. Saastamoinen et al., 2013; Fischer et al., 2017; Saxon et al., 2018).

Reaction norms ( Figures 1B–D ), where variation in phenotype is displayed as a function of variation in environment, are a common and very useful way to graphically represent developmental plasticity (Schlichting and Pigliucci, 1998; DeWitt and Scheiner, 2004; Sultan, 2017). These representations reflect the extent and the effects of environmental cues on phenotype expression ( Figure 1B ). Horizontal (flat) lines represent traits that are invariant, or robust (Félix and Barkoulas, 2015), to environmental conditions. For example, vulval cell fate patterning in Caenorhabditis elegans has been shown to be unresponsive to changes in temperature, salinity, or nutrients, which affect other aspects of worm development (Félix, 2012). Diagonal lines, in contrast, reflect a gradual relationship between environment and phenotype, such as temperature-induced differences in insect body size (Nijhout, 2003b; Mirth and Shingleton, 2012); the steepest the line is, the more plastic is the trait. In other cases, such as the polyphenisms described earlier, responses to environmental variation can be discontinuous, with discrete phenotypes below and above some environmental thresholds (e.g. Pfennig, 1992; Kamakura, 2011; Casasa and Moczek, 2018). Well-described examples of non-linear reaction norms resulting in polyphenims include honey bees, where only larvae that are fed high quantities of royal jelly develop into queens (Kamakura, 2011), and the diet-dependent size of horns in dung beetle males, which is disproportionally large for large males (Moczek, 1998). Some of such discontinuous reaction norms may result from cases where organisms can only produce two alternative phenotypes or from cases where organisms have never been exposed to intermediate environmental conditions, which would otherwise reveal gradual reaction norms.

The degree and type of plastic responses can vary between species, as is the case with the density-dependent swarming responses in locusts and grasshoppers (Song et al., 2017) or with the thermal effects on pigmentation in different mycalesine butterflies (van Bergen et al., 2017). Plastic responses can also differ between populations of the same species ( Figures 1B, C ). For instance, the effects of oxygen on brain and gill size differ between populations of cichlid fish from different geographical locations (and oxygen regimes; Crispo and Chapman, 2010). Reaction norms typically describe how one environmental factor affects one specific trait for one particular genetic background, and they usually differ between traits, environmental cues, and genotypes. For instance, dung beetle horn size increases more with food quantity than does body size (Moczek, 2002; Casasa and Moczek, 2018). Different melanin-based traits respond in independent (sometimes even opposing) manners to temperature and photoperiod in butterflies (Stoehr and Wojan, 2016). The different responses that genotypes can have to environmental inputs correspond to non-parallel reaction norms and significant genotype-by-environment (GxE) interactions (Saltz et al., 2018; Figures 1C, D ). The genes responsible for those inter-genotype differences in reaction norms can presumably fuel the evolution of plasticity and will be further discussed in Genes for Variation in Plasticity.

External environmental cues affect adult phenotype by altering developmental rates and/or trajectories. For this to happen, organisms must be able to somehow sense external conditions and provide information about those conditions to the developing tissues where changes in developmental cascades will result in changes in phenotype ( Figure 2A ). The cellular and molecular players involved in the sequence of steps that connect variation in external conditions to variation in developmental outcomes have been characterized in a number of cases, such as dauer formation in C. elegans nematodes (see Fielenbach and Antebi, 2008; Allen et al., 2015; Figure 2B ). The perception of external conditions can involve specific neurons (e.g. temperature-sensitive neurons in D. melanogaster; Peng et al., 2007) and/or specific molecules (e.g. ascaroside pheromones that indicate high density in C. elegans; Ludewig and Schroeder, 2013). The information about the external conditions is then conveyed to the tissues developing internally, a process that typically involves one or various hormones whose synthesis, degradation, and/or activation depend on environmental conditions (Mirth et al., 2009; Mirth et al., 2014; Vellichirammal et al., 2017). These hormones will then affect gene expression and/or function in the target plastic tissues (e.g. Koyama et al., 2013; Monteiro et al., 2015). In some cases, direct effects of environmental factors on gene expression have been reported, for instance, with temperature presumably directly regulating the transcription of clock genes in zebrafish (Lahiri et al., 2005). Because developmental plasticity refers to the production of different phenotypes from the same genotype, it necessarily involves epigenetic mechanisms, i.e. that are beyond the nucleotide sequence in genomic DNA. Mechanisms such as methylation of DNA or acetylation of histones, for example, are capable of mediating changes in gene expression without changes in gene sequence and have been implicated in plastic development (e.g. caste determination in honeybees and ants; Kucharski et al., 2008; Lyko et al., 2010; Bonasio et al., 2012; Simola et al., 2013b, Simola et al., 2016). The cascade of events from environmental inputs to alternative phenotypes will involve “effector genes” (cf. Figure 2 ) whose expression and/or function depend on environmental conditions and underlie developmental changes. These effector genes will be further discussed in Environmentally Responsive Genes.

Theoretical and in silico studies have identified a series of conditions that can influence the evolution of plasticity (e.g. Tufto, 2000; Lande, 2009). These include the predictability of environmental fluctuations and the reliability of inductive environmental cues in predicting the future selective environment (Leimar et al., 2006; Reed et al., 2010), both of which should favor plasticity. The availability and effective assessment of such cues will determine the evolution of plasticity in relation to other possible responses to environmental variation, such as bet-hedging (Simons, 2011; Herman et al., 2014; Tufto, 2015). The evolution of plasticity can also be influenced by other factors, such as potential costs associated with maintaining the sensory and regulatory mechanisms needed for plastic responses (DeWitt, 1998; Tufto, 2000; Callahan et al., 2008; Snell-Rood, 2012; Lande, 2014; Murren et al., 2015). Trade-offs between plasticity and different fitness-related traits have been documented that can presumably constrain the evolution of plasticity. For example, in freshwater snails, the levels of predator-induced plasticity in shell shape have been shown to be negatively correlated with growth rate (DeWitt, 1998) and predator-induced phenotypes have been shown to have lower survival in the presence of a different predator (Hoverman and Relyea, 2009). When the ecological conditions favor plasticity, and because plasticity is itself a heritable trait (Via and Lande, 1985; Scheiner and Lyman, 1989, Scheiner and Lyman, 1991; Thompson, 1991; Carter et al., 2017), plasticity can and does evolve (Box 1). Several studies have documented transitions between plastic and robust development in both natural populations (e.g. Cahan et al., 2004; Aubret and Shine, 2009) and laboratory populations (e.g. Suzuki and Nijhout, 2006; Bento et al., 2010). The loci carrying allelic variation responsible for variation in plasticity that can fuel its evolution will be discussed in Genes for Variation in Plasticity.

It is believed that, in most cases, the ancestral condition in trait development is some level of environmental sensitivity, with selection then favoring the ability to buffer environmental effects (Newman and Müller, 2000; Nijhout, 2003a; Nijhout et al., 2017). Selection can act on the regulation of environmentally sensitive phenotypes and adjust the properties and/or magnitude of the plastic responses by a process called genetic accommodation (West-Eberhard, 2003; West-Eberhard, 2005; Crispo, 2007). Well-documented examples include the evolution of thermal plasticity in larval pigmentation in experimental populations of Manduca sexta (Suzuki and Nijhout, 2006) and the evolution of developmental rate in response to aridification in spadefoot toads (Gomez-Mestre and Buchholz, 2006; Kulkarni et al., 2017). When genetic accommodation leads to the (genetic) fixation of what was previously an environmentally induced phenotype, we talk about genetic assimilation (Schlichting and Wund, 2014; Ehrenreich and Pfennig, 2016; Schneider and Meyer, 2017). The occurrence and underlying genetic mechanism of genetic assimilation have what is arguably its classical example in the fixation of a bithorax phenotype originally induced by the exposure of D. melanogaster to ether (Waddington, 1953; Gibson and Hogness, 1996). More recent studies of genetic assimilation have illustrated how transitions from environmental to genetic control of adaptive traits may happen at a (relatively) fast pace (e.g. loss of plasticity in head size in Notechis scutatus snakes within a few thousand years; Aubret and Shine, 2009) and how it can lead to complex interdependence between environmentally and genetically induced phenotypes (e.g. with caste determination in Pogonomyrmex harvester ants; Cahan et al., 2004).

Whereas, in the previous section, we discussed how plasticity is a trait under selection and can evolve, here we will focus on how plasticity might impact adaptive evolution (Box 1). Evolved plasticity is thought to be able to help populations face challenges posed by changing environments (e.g. de Jong, 2005; Xue and Leibler, 2018; Fox et al., 2019), allow them to cope with environmental heterogeneity (e.g. West-Eberhard, 2003; Lande, 2009; Levis et al., 2018), and aid the colonization of novel environments (e.g. Lande, 2015; Snell-Rood et al., 2018). More than plasticity potentially enabling persistence under novel conditions and thereby “buying time” for adaptation to occur (Corl et al., 2018; Pennisi, 2018), it has been suggested that plasticity may actually promote adaptive evolution and diversification (e.g. see Wund, 2012; Smith and Ritchie, 2013; Gilbert et al., 2015). This idea has been proposed with a number of variations, sometimes subtle, that emphasize different mechanisms and/or outcomes. Modeling work has sustained that plasticity might be able to foster an increase in the levels of genetic variation in a population (e.g. Draghi and Whitlock, 2012), thereby impacting adaptive evolution. A particular type of genetic variation often associated with plasticity is called cryptic genetic variation, i.e. genetic variation that is normally not expressed but can be uncovered by environmental or genetic perturbation (e.g. Gibson and Dworkin, 2004; McGuigan et al., 2011; Paaby and Rockman, 2014; Schneider and Meyer, 2017). Plasticity has been argued to favor both the accumulation of cryptic genetic variation (e.g. in genes not expressed under certain environmental conditions) and its release (e.g. upon environmental conditions outside the range typically experienced by the population). Such genetic variation can then fuel evolutionary change (Gibson and Dworkin, 2004; Ledón-Rettig et al., 2010; Paaby and Rockman, 2014; Schneider and Meyer, 2017). Furthermore, the evolutionary potential of plasticity has been illustrated by the contribution of plasticity to reproductive isolation, which includes cases where plastic traits affect mating (e.g. pigmentation patterns in butterflies; Westerman et al., 2014) or timing of organismal life-events (e.g. phenological shifts in grasshoppers; Buckley et al., 2015).

The range of phenotypic variation generated by plasticity has been proposed to be able to initiate and accelerate the pace of adaptive evolution and to promote morphological as well as phylogenetic diversification (West-Eberhard, 2003, West-Eberhard, 2005; Pfennig et al., 2010; Wund, 2012). This relies on the suggestion that plasticity could be an immediate source of initial inter-individual differences in phenotypes (and in fitness) that could then become genetically fixed by genetic assimilation (West-Eberhard, 2003; Wund et al., 2008; Susoy et al., 2015; Levis and Pfennig, 2016; Gibert, 2017; Levis and Pfennig, 2019), leading to increased phenotypic diversification among populations. This same idea is at the basis of what have been called the “plasticity-first” model (Levis and Pfennig, 2016; Levis and Pfennig, 2018) and the “flexible stem” hypothesis, with the latter specifically suggesting that a “plastic ancestor” more readily originates phylogenetic divergence and adaptive radiations (West-Eberhard, 2003; Standen et al., 2014; Schneider and Meyer, 2017). This distinction between the two models is not always made; the terms have sometimes been used inter-changeably but are also often discussed separately (Wund et al., 2008; Muschick et al., 2011; Levis and Pfennig, 2016; Gibert, 2017). For instance, the invasion of new niches by amphibian species is proposed to represent a natural example of the “plasticity first” model, with nutrition-induced plasticity fostering the origin of carnivore morphs in some species (Levis et al., 2018). In contrast, evolutionary diversifications in threespine stickleback fish (with head and mouth shape variation across ecotypes; Wund et al., 2008) and tetrapods (with the origin of “terrestrialization” traits; Standen et al., 2014) have been discussed as examples supporting the “flexible stem” hypothesis. The study of plasticity and its (potential) impact on adaptive evolution has generated substantial interest but also much discussion (e.g. Pigliucci, 2007; Laland et al., 2014; Forsman, 2015; Nicoglou, 2015; Turcotte and Levine, 2016; Charlesworth et al., 2017; Futuyma, 2017). This includes controversy about definitions for plasticity as well as some level of lack of clarity about distinction and integration of different models proposed to account for the role of plasticity in adaptation and diversification.

Context-dependent gene expression is at the heart of development, transforming a single cell into spatially organized different cell types that make up multicellular organisms (Gilbert, 2000). In the case of developmental plasticity, the environmental regulation of development requires gene expression and/or function to depend also on external context. Here, we will focus on effector genes whose expression changes in plastic tissues to alter developmental fate ( Figure 2 ). Efforts to identify such genes have documented differences in expression between environments for whole developing organisms or specific developing organs and either targeting specific candidate genes/pathways or doing transcriptome-wide scans ( Figure 3A ).

Candidate gene approaches to probe the genetic basis of environmentally regulated development have ranged from studies of gene expression levels (e.g. using quantitative polymerase chain reaction; e.g. Dalton et al., 2015) to studies of spatial expression patterns on specific organs (e.g. using in situ hybridization methods; Gibert et al., 2016). They have also included both in vivo (e.g. kairomone-driven regulation of patterning genes during production of the anti-predator defense morph in Daphnia water fleas; Miyakawa et al., 2010) and in vitro (e.g. temperature-dependent expression of sex-determining genes in turtle gonad cultures; Shoemaker-Daly et al., 2010) approaches as well as single-species (e.g. allatotropin expression under starvation in M. sexta; Lee and Horodyski, 2006) and multi-species (e.g. wing-patterning genes in polyphenic ants; Abouheif and Wray, 2002) analyses. Detailed studies of specific and well-known developmental pathways in emblematic plasticity models, such as pigmentation in Drosophila (e.g. De Castro et al., 2018), body and organ size in different insects (e.g. Lee and Horodyski, 2006; Mendes and Mirth, 2016), and wing development in ants (e.g. Abouheif and Wray, 2002), have been very insightful. The systematic investigation of melanogenesis enzymes underlying pigmentation development in D. melanogaster revealed how the environmental regulation of gene expression can happen via modulation of enhancer activity (Gibert et al., 2016; Gibert et al., 2017b; De Castro et al., 2018) or via effects on chromatin regulation (Gibert et al., 2007). The investigation of the hormonal mechanisms that underlie the nutritional regulation of body and organ size in different insects, on the other hand, demonstrated tissue and stage specificity of the effects of nutrition on the expression and sensitivity of several players from the insulin and target of rapamycin pathways (Mirth and Shingleton, 2014; Koyama et al., 2016; Mendes and Mirth, 2016). Finally, the analysis of the expression of wing development genes in workers versus queens in different ant species revealed dissociation (and modularity) of wing-patterning genes, which presumably influenced the evolutionary lability of this polyphenism (Abouheif and Wray, 2002; Béhague et al., 2018).

Genome-wide scans of different types have allowed for a more unbiased search of (putative) effector genes, including for traits for which the underlying developmental genetic basis is not well understood. Transcriptomic differences due to differences in developmental environments have been characterized for various species and in relation to various environmental cues, for instance, nutrition in beetles (Kijimoto et al., 2014) and ant castes (Evans and Wheeler, 2000), temperature in fish (Wellband and Heath, 2017), and multiple environmental cues in flies (Zhou et al., 2012). External environmental factors can affect the expression of a large number of genes, with 15% of the D. melanogaster genome being differentially expressed (Zhou et al., 2012) and 10% of the expressed genes being differentially spliced (Jakšić and Schlötterer, 2016) between temperatures. Aside from providing valuable quantitative insights into the distribution of environmental effects on gene expression levels (e.g. Zhou et al., 2012) and allowing researchers to draw transcriptomic reaction norms (Aubin-Horth and Renn, 2009; Gao et al., 2015; Oomen and Hutchings, 2017), transcriptomic scans have been very valuable at identifying candidate effector genes or pathways (see Figure 3 ) for further detailed analysis, which can be more or less obvious from the onset. Examples include specific gene classes/families differently expressed between castes in social insects (Evans and Wheeler, 2000), the down-regulation of growth and metabolism genes influencing the duration of larval stage in Strongylocentrotus droebachiensis sea urchins under food scarcity (Carrier et al., 2015), the differential expression of endocrine and pigmentation-related genes that underlie the seasonal pigmentation morphs in Junonia coenia butterflies (Daniels et al., 2014), and the caste-specific expression of chemoreception genes in termites (Mitaka et al., 2016). Whereas most studies have used transcriptomic approaches (e.g. Levine et al., 2011; Zhou et al., 2012; Chen et al., 2015) to investigate environmental effects on the levels and regulation of mRNAs and different types of non-coding RNAs (e.g. Garrett and Rosenthal, 2012; Jakšić and Schlötterer, 2016; Healy and Schulte, 2019), others have focused on environmentally induced changes in levels of protein (using proteomic approaches; e.g. Silvestre et al., 2012) and metabolites (using metabolomic approaches; see Bundy et al., 2009). Although such whole-genome scans can provide valuable insights onto the magnitude and nature of environmental effects on gene expression, it is crucial to remember that the identified differences in gene-product quantity do not necessarily translate into differences in organismal phenotypes (see Evans, 2015).

The environmental sensitivity of development is a property of a genotype (e.g. Gockel et al., 2002; Nussey et al., 2005; Lardies, 2008; Saltz et al., 2018). For any specific trait and cue, genotypes can differ in various properties of the corresponding reaction norms ( Figure 1D ). This includes variation in intercept (e.g. thermal reaction norms for life-history traits in Ischnura elegans damselflies; Bouton et al., 2011), shape (e.g. thermal reaction norms for pigmentation in Drosophila mediopunctata flies; Rocha et al., 2009), and slope (e.g. thermal reaction norms for growth rate in Orchesella cincta springtails; Driessen et al., 2007). Genotypes can also differ in the environmental cue that triggers change or in the environmental threshold for the induction of phenotypic change (e.g. for hormesis in C. elegans; Sikkink et al., 2014). The genes responsible for variation in reaction norms provide the raw material for selection to drive the evolution of plasticity (see DeWitt and Scheiner, 2004). There is evidence for both the polygenic nature of artificially selected changes in shape and height of reaction norms (e.g. Wijngaarden and Brakefield, 2000) and the single allelic variants that cause loss (e.g. Bento et al., 2010; Ragsdale et al., 2013; Ragsdale and Ivers, 2016) or gain (e.g. Suzuki and Nijhout, 2006) of environmental sensitivity.

To identify loci contributing to inter-genotype variation in plasticity, researchers have compared reaction norms between allelic variants of specific candidate genes or used genome-wide mapping scans of different types. Studies of thermal plasticity in D. melanogaster development, for example, illustrate both approaches: differences in reaction norms for abdominal pigmentation for mutant versus wild-type alleles of different melanogenesis genes (Gibert et al., 2007; Gibert et al., 2017b) and mapping of loci with allelic variation associated with variation in the slope of reaction norms for body size (Lafuente et al., 2018). Although, in the past couple of decades, we have seen great progress in unraveling the genetic basis of phenotypic variation and adaptation, covering many different traits and species (e.g. Barrett and Hoekstra, 2011; Martin and Orgogozo, 2013; Savolainen et al., 2013; Pardo-Diaz et al., 2015), relatively little is known about the genetic basis of variation in plasticity (see Pigliucci, 2005). In fact, most quantitative trait loci (QTL) mapping studies have tracked phenotypic variation under one single environmental condition, precluding the assessment of environment-specific or gene-by-environment effects. Studies that did include phenotyping under different environments (e.g. survival after heat stress in zebrafish; Hosseini et al., 2018), as is necessary to characterize the genetic basis of variation in plasticity, revealed a prevalence for environment-specific QTL (Dilda and Mackay, 2002; Green et al., 2014), which are QTL that are expressed differently in different environments. Studies mapping variation in D. melanogaster at different temperatures showed large proportions of QTLs exhibiting QTL-by-environment interactions, for example, 70% (in Gurganus et al., 1998) and 33% to 50% (in Dilda and Mackay, 2002) for bristle number and 83% for reproductive performance (Fry et al., 1998). Environment-specific QTL also includes cases with alleles having antagonistic effects in different environments (e.g. up to 59% of QTLs for variation in life span at different temperatures and nutritional regimes in D. melanogaster; Vieira et al., 2000). Although QTL-by-environment interactions certainly reflect inter-genotype differences in reaction norms, they are not necessarily QTLs contributing to inter-genotype variation in plasticity ( Figure 4 ). To unravel such loci, mapping has used the slope of reaction norms as the target quantitative trait. Examples include the identification of QTLs associated with the slope of reaction norms for thermal regulation of life-history traits in C. elegans (Gutteling et al., 2007) and of body size (Lafuente et al., 2018) and cold tolerance (Ørsted et al., 2018) in D. melanogaster. These studies identified QTLs corresponding to different types of genomic regions (e.g. regulatory and coding) and different putative biological functions (e.g. regulation of development, components of the nervous system, and environmental stress responsive genes), including regulatory genes (Ørsted et al., 2018), which are thought to play key roles on the genetic control of plasticity (Schlichting and Pigliucci, 1993).

More than identifying specific genes and describing their function, studies of the genetic basis of plasticity have shed light onto old debates about the existence and nature of plasticity genes (Via, 1993; DeWitt and Scheiner, 2004; Nicoglou, 2015). The proposed models had distinguished between scenarios where plasticity is a side effect of selection on the (plastic) trait versus where plasticity is itself a direct target of selection (Via and Lande, 1985; Scheiner and Lyman, 1991; Schlichting and Pigliucci, 1993;). Whereas positive genetic correlations between trait values and trait plasticity and environmentally dependent allelic effects are presumably consistent with the former (e.g. Fry et al., 1998), examples where selection on plasticity is independent of selection on trait mean (e.g. thermal plasticity in the timing of egg-laying in Ficedula albicollis collared flycatchers; Brommer et al., 2005, and in thorax size in D. melanogaster; Scheiner and Lyman, 1991) and/or where selection on trait mean does not constrain plastic responses (e.g. for cold tolerance in B. anynana butterflies; Franke et al., 2012) are consistent with the latter. In fact, several recent mapping studies have reported little overlap between genes contributing to variation in trait means with those contributing to variation in trait plasticity (e.g. Lafuente et al., 2018; Ørsted et al., 2018). This suggests a distinct genetic basis for trait plasticity versus the genetic basis for phenotypic variation in the trait itself (at any given environment) and points to the potential for independent evolution of the two.

Several studies in animals have combined the high-throughput gene expression analysis under different environments with forward genetic approaches to identify the genetic basis underlying environmentally dependent gene expression differences (Runcie et al., 2012). These studies, which involve assessing gene expression under different environmental conditions in different genotypes [for the so-called eQTL analysis or genome-wide association studies (GWAS) of gene expression; e.g. Li et al., 2006], allowed to link gene expression differences to genomic markers or allelic variants (see Grishkevich and Yanai, 2013). By associating variation in genotype to variation in environmentally regulated gene expression, researchers have not only characterized GxE interactions in gene expression and their underlying genetic basis but also identified genomic features distinguishing genes with GxE interactions from other types of genes. Work in C. elegans, for example, revealed that genes that exhibited GxE interactions tend to show distinct promoter architecture (e.g. long promoter with a high concentration of regulatory motifs) as well as mid-range expression level (Grishkevich et al., 2012), both of which are characteristic of tightly regulated genes (Promislow, 2005).

We have explicitly focused on genomic work studying genes involved in the regulation of plastic responses, i.e. those that respond to environmental inputs and affect trait expression, and genes potentially involved in the evolution of plastic responses, i.e. those that harbor allelic variants contributing to differences in reaction norms. There is, of course, some overlap in the genes involved in trait development, trait plasticity, and plasticity variation (e.g. orange gene in Figure 3 ). A compelling example comes from studies of abdominal pigmentation in Drosophila. The gene tan, encoding one of the melanogenesis enzymes (True, 2003), has been implicated in inter-individual variation in pigmentation within a given temperature (e.g. allelic variants of tan underlie differences in pigmentation; Dembeck et al., 2015), regulation of pigmentation plasticity (e.g. flies with reduced expression of tan have lighter pigmentation; Kalay et al., 2016), and variation in pigmentation plasticity (e.g. different tan alleles correspond to different reaction norms; Gibert et al., 2016). However, there does not need to be a complete overlap between the genes involved in plasticity regulation and those that contribute to plasticity variation/evolution ( Figure 3 ). Indeed, a gene that is differentially expressed between developmental environments might not harbor allelic variants contributing to variation in reaction norms (e.g. blue in Figure 3 ). Similarly, a gene contributing to variation in reaction norms does not need to differ in expression across environments (e.g. red in Figure 3 ).

In the previous sections, we highlighted different genomic studies to identify the genes involved in the regulation and evolution of plasticity, including transcriptomics or genetic mapping approaches. Although extremely powerful at identifying candidate genes, these genome-wide scans are also very prone to false positives and not very informative about the actual role of those genes. As such, it is crucial to independently run functional analysis (Gibert et al., 2017a) to validate candidate genes, something not always easily accessible. Tools for the manipulation of gene function (e.g. transgenic knock-downs or knock-outs to reduce and abolish gene expression) as well as the manipulation of specific sites (e.g. via allelic replacements) were typically only easily accessible in established models. However, fast advances in analytical tools, for both genomic and functional studies, are now enabling us to move from model organisms to less established models (e.g. Marinković et al., 2012; Santure and Garant, 2018; see Russell et al., 2017). The progress in methods is also aiding researchers to move from laboratory strains to natural populations, often including more realistic scenarios in terms of the diversity in genetic backgrounds and demographic parameters and of the complexity of external environments. All these factors are bound to affect phenotype expression and/or evolution (Braendle and Félix, 2008; Chevin, 2013; Piggott et al., 2015; Bretman et al., 2016; Matsui and Ehrenreich, 2016; Fischer et al., 2017). How organisms perceive and integrate information from complex environments, with multiple factors that can vary more or less independently, including within the time it takes to complete development (Warkentin, 2011; Fischer et al., 2017; Rodrigues et al., 2017), is a fascinating topic in need of deeper characterization.

Studies investigating the genes underlying plastic responses are rapidly increasing. Previous research using a variety of systems illustrated that environmentally induced changes in gene expression are pervasive and can include a variety of molecular and functional genes classes (e.g. Zhou et al., 2012; Kijimoto et al., 2014), as reviewed here. Future work exploring changes in gene expression beyond a single species (or a single population), and beyond a single environmental cue, could potentially help to identify general patterns about the effector genes underlying developmental plasticity, whether they belong to particular functional classes or are shared between cues. Moreover, as much more is known about the molecular mechanisms and the genetics of environmentally induced phenotypes in plants (see Sultan, 2000; Sultan, 2017), these are exciting times to start integrating studies on plasticity from animals and plants, which could provide insights on the potential commonalities in the mechanisms controlling whether phenotypes would respond to (or buffer) environmental variation.