Natural or artificial selection that favors higher values of a particular trait within a given population should engender an evolutionary response that increases the mean value of the trait. For this prediction to hold, the phenotypic variance of the trait must be caused in part by additive effects of alleles segregating in the population, and also the trait must not be too strongly genetically correlated with other traits that are under selection. Another prediction, rarely discussed in the literature, is that directional selection should favor alleles that increase phenotypic plasticity in the direction of selection, where phenotypic plasticity is defined as the ability of one genotype to produce more than one phenotype when exposed to different environments. This prediction has received relatively little empirical attention. Nonetheless, many laboratory experiments impose selection regimes that could allow for the evolution of enhanced plasticity (e.g. desiccation trials with Drosophila that last for several hours or days). We review one example that involved culturing of Drosophila on lemon for multiple generations and then tested for enhanced plasticity of detoxifying enzymes. We also review an example with vertebrates that involves selective breeding for high voluntary activity levels in house mice, targeting wheel-running behavior on days 5+6 of a 6-day wheel exposure. This selection regime allows for the possibility of wheel running itself or subordinate traits that support such running to increase in plasticity over days 1–4 of wheel access. Indeed, some traits, such as the concentration of the glucose transporter GLUT4 in gastrocnemius muscle, do show enhanced plasticity in the selected lines over a 5–6 day period. In several experiments we have housed mice from both the Selected (S) and Control (C) lines with or without wheel access for several weeks to test for differences in plasticity (training effects). A variety of patterns were observed, including no training effects in either S or C mice, similar changes in both the S and C lines, greater changes in the S lines but in the same direction in the C lines, and even opposite directions of change in the S and C lines. For some of the traits that show a greater training effect in the S lines, but in the same direction as in C lines, the greater effect can be explained statistically by the greater wheel running exhibited by S lines (`more pain, more gain'). For others, however, the differences seem to reflect inherently greater plasticity in the S lines (i.e. for a given amount of stimulus, such as wheel running/day, individuals in the S lines show a greater response as compared with individuals in the C lines). We suggest that any selection experiment in which the selective event is more than instantaneous should explore whether plasticity in the appropriate (adaptive) direction has increased as a component of the response to selection.
Natural selection tends to act most strongly on aspects of the phenotype(traits) at relatively high levels of biological organization because they are the most strongly correlated with Darwinian fitness (e.g. lifetime reproductive success). Components of life history, behaviors and aspects of organismal performance (for reviews, see Ketterson and Nolan, Jr, 1999; Irschick and Garland, Jr,2001; Kingsolver and Huey,2003; Costa and Sinervo,2004) are `complex traits' in that they are composed of many subordinate traits at lower levels of biological organization(Swallow and Garland, Jr,2005) (Fig. 1). Thus, the evolutionary response to selection on such complex traits necessarily entails associated changes in aspects of morphology, physiology and biochemical pathways (Ghalambor et al., 2003; Sinervo and Calsbeek, 2003). In addition, complex patterns of trade-offs and constraints are expected to occur, and the genetic architecture underlying these may itself evolve in response to selection (e.g. Chippindale et al., 2003; Rose et al., 2005).
Regardless of position in the biological hierarchy, most if not all traits are affected by alleles segregating in the population at many loci, whose expression is affected by numerous environmental factors, including both abiotic (e.g. temperature) and biotic (e.g. social interactions). Environmental factors can influence development by acting at any time after formation of the zygote, or in some cases even before (e.g. maternal effects acting on the unfertilized egg). Moreover, organisms often exert some level of choice with respect to environmental conditions that may affect their own development, such as through selection of habitat or diet (e.g. Geiser et al., 1997; Kupferberg, 1997). Whenever they act, the consequences of environmental effects are often termed developmental or phenotypic plasticity.
Directional natural selection is predicted to have various effects, some fairly obvious but others less so. If natural selection in a given population favors individuals with higher values of a particular trait, then the population mean value of that trait is predicted to increase from generation to generation (Fig. 2A),assuming that some additive genetic variance exists and that the trait is not too strongly genetically correlated with other traits under selection. Evolutionary biology is replete with empirical examples illustrating the validity of this prediction (Endler,1986). Moving from phenotype to genotype, a second prediction is that alleles with `appropriate' pleiotropic effects will be favored, which would facilitate the coordinated evolution of components of complex phenotypes. For example, if selection were to favor individuals that foraged widely to find food, then alleles that increased motivation for high locomotor activity might be favored most directly, and the subset of those alleles that also tended to increase ability for high activity would be particularly favored. This sort of process, in which the genetic architecture of the traits involved (especially the additive genetic variance–covariance matrix)evolves to become more consistent with the prevailing pattern of multivariate selection, could facilitate further evolution and adaptive radiation (e.g. Garland, 1994; Schluter, 1996). A somewhat more subtle genetic prediction is that directional selection should tend to favor alleles that exhibit phenotypic dominance in the direction of selection,and this has also received empirical support (e.g. Broadhurst and Jinks, 1974; Henderson, 1981; Hewitt et al., 1981; Mather and Jinks, 1982; Falconer, 1989; Garland et al., 1990; Lynch, 1994; Lynch and Walsh, 1998). For example, under a selective regime that favored high activity levels, alleles that promoted high activity and were dominant to alleles with neutral or negative effects on activity would be the most favored among the spectrum of`high-activity alleles'.
A fourth hypothesis, not mutually exclusive with the previous three, is that the average plasticity of the population should also evolve if the selective agent imposes more than instantaneous `stress' (sensuHarshman et al., 1999; Wilson and Franklin, 2002; Gabriel, 2005) on the population. More specifically, plasticity of the trait under selection or of subordinate traits that contribute to that trait should increase in the direction that would be adaptive (tend to confer higher organismal performance and/or higher Darwinian fitness) under the prevailing selective regime(Fig. 2B). To clarify this hypothesis, we need to distinguish between selective agents that (1) impose selective events that are virtually instantaneous relative to the time course of possible plastic responses and (2) impose relatively prolonged selective events. As an example of an `instantaneous' selective event, a cheetah(selective agent) chases an antelope (selective event) and the outcome (life or death for the antelope) occurs so quickly (a matter of seconds) that the exercise physiology of the antelope has no chance to `train'. In addition to its own behavior, the motivation and abilities of the cheetah, and some element of chance (e.g. tripping over a rock), what determines the outcome of the selective event are the `innate' (constitutive or intrinsic) exercise abilities of the antelope at the instant the cheetah began its pursuit. As an example of a `prolonged' selective event, global warming trends that occur over years would allow long-lived organisms to acclimatize to the higher temperatures in a way that might be beneficial (e.g. upregulation of heat-shock proteins). Thus, the Darwinian fitness of individuals in a crocodilian population experiencing environmental warming might depend on both innate and induced components of heat tolerance. Of course, the distinction between instantaneous and prolonged selective events is not always clean. For example, an antelope might learn from a failed close encounter with a cheetah and became better at avoiding capture in subsequent encounters by altering its own behavior. The classic selective event of an ice storm that killed some(but not all) sparrows in a fortuitously collected sample(Bumpus, 1899; Lande and Arnold, 1983; Endler, 1986) lasted overnight, such that individual sparrows may have varied in thermal tolerance because of both innate and induced individual variation in components of heat tolerance.
The actual mechanisms for the evolution of increased plasticity could be several, of which we will mention two. First, referring to Fig. 2A, in addition to individuals whose phenotypes are intrinsically high, individuals that exhibit plasticity in the direction favored by selection will tend to be among the`survivors' each generation; thus, appropriately plastic individuals will be favored by phenotypic selection [see p. 67(Falconer, 1990)]. If plasticity is heritable, then it will evolve in response to such selection(assuming genetic correlations with other traits under selection are not too strong). Second, the genes that affect the constitutive value of the phenotype might also have pleiotropic effects (in the appropriate direction) on the plastic response of that phenotype when the organism experiences chronic (more than instantaneous) exposure to the selective agent (e.g. gradual warming,repeated encounters with predators). The genetics of plasticity are discussed further elsewhere (Scheiner,1993; Pigliucci,2005). Here it is also worth noting that the evolution of plasticity is related to the concept of genetic assimilation, a process in which environmentally induced phenotypic variation that is favored by selection (natural or artificial) gradually (across many generations) comes to be constitutively produced [reviewed elsewhere in this issue(Pigliucci et al., 2006)].
The foregoing ideas about evolutionary processes would seem to be implied by the `beneficial acclimation hypothesis' (see below), but to our knowledge they have not been discussed so explicitly in the literature. In any case, we hypothesize that the mean plasticity of a population under directional selection (Fig. 2A) should evolve from being neutral (or possibly deleterious) to being beneficial or adaptive (Fig. 2B). This evolutionary hypothesis has received little direct empirical attention (but see Falconer, 1990; Scheiner, 2002). Nonetheless,as discussed below, many laboratory experiments impose selective regimes that could allow for the evolution of enhanced plasticity (e.g. desiccation trials with Drosophila that last for several days), and the few that have tested for evolutionary changes in plasticity have found some evidence for it.
When behavioral performance traits are the subject of directional selection, the role of phenotypic plasticity in evolutionary response may be particularly interesting [other perspectives on the importance of behavior have been discussed elsewhere (Huey et al., 2003; Price et al.,2003; Price,2006)]. The term `self-induced adaptive plasticity' was proposed(Swallow et al., 2005) for situations in which a behavior induces plastic changes in morphological or physiological traits that in turn enhance the ability to perform the behavior. For example, animals that migrate altitudinally might make `trial runs' that would cause cardiovascular, pulmonary or metabolic changes that would improve their ability to function at high altitude. Similarly, animals that begin feeding on a new type of food may experience changes in digestive enzymes that increase efficiency of nutrient extraction and/or detoxification [examples of related effects of diet are reported elsewhere, including references therein(Geiser et al., 1997; Kupferberg, 1997)].
The first purpose of this paper is to provide a brief introduction to phenotypic plasticity from an ecological and evolutionary perspective [see also elsewhere in this issue (Fordyce,2006; Pigliucci et al.,2006; Price,2006)]. Second, we discuss how the evolution of plasticity can be studied, with an emphasis on the experimental evolution approach. Finally, we review some results from a study on the experimental evolution of high voluntary activity levels in house mice, including examples of self-induced adaptive plasticity.
Defining phenotypic plasticity
From the perspective of evolutionary biology, classic and dramatic examples of phenotypic plasticity in animals include wing polymorphisms in some insects, the timing of metamorphosis in amphibians, and alternative reproductive tactics in male vertebrates – all of which exhibit complex neuro-endocrine control mechanisms that are sensitive to various environmental factors (Ketterson and Nolan, Jr,1999; Sinervo and Calsbeek,2003; Boorse and Denver,2004; Knapp, 2004; Zera, 2004). From the biomedical perspective, well-known examples of plasticity include effects of intentional physical conditioning (exercise training)(Flück, 2006) such as weight lifting, on human morphology and physiology. Various biomedical subfields use additional terminology, such as `metabolic plasticity' or`cardiac remodeling', and the molecular mechanisms underlying such processes as muscular and neuronal plasticity are the subject of intensive study [for reviews, see other articles in this issue(Flück, 2006; Hood et al., 2006; Johnston, 2006; Magistretti, 2006; Swynghedauw, 2006)]. (Many environmental insults, e.g. excessive alcohol consumption, smoking, inhalation of coal dust, can lead to `plastic' changes in organs and organ systems, but when such changes are clearly pathological they are not typically included under the rubric of phenotypic plasticity.) In plants, basic growth form is notoriously plastic, and many readers will be familiar with the differences between dandelions growing in shade versus sun [although genetic differences among clones may also be involved(Collier and Rogstad,2004)].
As with the term `adaptation' (see below), phenotypic plasticity can refer both to a process and to the outcome of that process. Phenotypic plasticity can be defined formally as the ability of one genotype to produce more than one phenotype when exposed to different environments, as the modification of developmental events by the environment, or as the ability of an individual organism to alter its phenotype in response to changes in environmental conditions (Gordon, 1992; Scheiner, 1993; Via et al., 1995; Futuyma, 1998; Freeman and Herron, 2004; Pigliucci, 2005; Rezende et al., 2005; Stearns and Hoekstra, 2005; Pigliucci et al., 2006). The range of phenotypes that a given genotype (possessed by an individual organism or by an entire clone or inbred line) may produce when exposed to a range of environmental conditions is termed its norm of reaction, and non-parallel reaction norms of different genotypes indicate the presence of genotype-by-environment interaction.
The sequence of events involved in phenotypic plasticity often includes the following components: (1) something in the environment changes; (2) the organism senses that change; (3) the organism alters gene expression; and (4),usually, the altered gene expression yields additional observable phenotypes[e.g. see fig. 8 in Flück's paper in this issue(Flück, 2006)]. Several aspects of this scenario require amplification. With respect to (1), we may attempt to draw a distinction between environmental factors that are external or internal to an organism. Changes in ambient temperature, humidity or oxygen concentration would constitute external environmental factors, and many organisms respond to these with phenotypic plasticity that involves multiple organ systems and multiple levels of biological organization. Mechanical overload of the heart is an example of an environmental change that occurs within an organism, and it leads mainly to organ-specific changes that necessarily involve fewer levels of biological organization(Swynghedauw, 2006). Of course, external environmental `stresses' can also lead to tissue-specific responses (e.g. Cossins et al.,2006). Nonetheless, we may predict that, in general, external environmental changes will lead to more and more pervasive plastic responses as compared with internal changes. With respect to (2), some changes may occur without any formal sensing by the organism, e.g. as a result of direct (and possibly differential) effects of temperature on the rates of ongoing biochemical and physiological processes. With respect to (3), it is important to note that some plastic responses need not involve changes in gene expression (transcription) but instead could occur viaphosphorylation of existing proteins, changes in protein levels caused by variation in protein ubiquitination, or stimulation of existing microRNAs(Nelson et al., 2003; Schratt et al., 2006). For point (4) we emphasize the word `usually' because it is possible that lower level traits might change in offsetting ways such that higher level traits could show little or no apparent change. For example, it would be theoretically possible (though perhaps unlikely) for exercise training to cause an increase in maximal heart rate but a reduction in stroke volume such that cardiac output was unchanged.
Acclimation and acclimatization(Wilson and Franklin, 2002),as well as learning and memory (e.g. Magistretti, 2006), are encompassed by the most inclusive definitions of phenotypic plasticity. Therefore, environmentally induced changes may or may not be reversible,depending on the organism, trait, and when in the lifecycle and for how long the environmental exposure occurs (Hatle,2004; Johnston,2006). If the capacity for change is more-or-less fully reversible, then it may be termed phenotypic flexibility(Piersma and Lindstrom,1997).
Whether reversible or not, it is generally assumed that environmentally induced modifications are adaptive in the sense that they improve organismal function and/or enhance Darwinian fitness of the individual organisms that exhibit such effects (Nunney and Cheung,1997). In fact, this may or may not be true, and the claim that such changes will aid the organism has been termed the beneficial acclimation hypothesis (Leroi et al.,1994; Huey and Berrigan,1996; Huey et al.,1999; Wilson and Franklin,2002). In some cases, behavioral plasticity can shield lower level traits from selection (Huey et al.,2003; Price et al.,2003). At the population level, phenotypic plasticity in behavior and other traits can facilitate invasions of new habitats(Price et al., 2003; Price, 2006; Pigliucci et al., 2006). As reviewed elsewhere in this issue (Fordyce,2006), many ecological (cross-species) interactions are mediated by the phenotypically plastic responses of one or more species involved in the interaction. Some of these ecological interactions can be quite complex and difficult to predict, as when an herbivore induces a plant phenotype that in turn affects the performance of other herbivores(Fordyce, 2006)
At this point it is worth remembering that the word adaptation has numerous meanings in biology (Garland and Carter,1994; Bennett,1997). Most generally, we should keep in mind the distinction between what is often called `physiological adaptation' (environmentally induced changes that occur within individual organisms during their lifetimes,including acclimation and acclimatization) and `evolutionary adaptation'(cross-generational changes in the genetic composition of a population in response to natural selection). Physiological adaptation is one type of phenotypic plasticity, but the ability to be plastic for any particular trait may also be an evolutionary adaptation whose details vary among organisms.
As noted above, although biologists have usually assumed that physiological adaptation is adaptive in the evolutionary sense, this is not always a safe assumption because some changes will be simply the result of activation of control systems designed to do something else, and they can even be maladaptive, including various human pathologies(Nesse, 2005; Swynghedauw, 2006). In general, non-adaptive plasticity might be expected to occur any time that an organism is exposed to environmental conditions with which it is `unfamiliar'in terms of its evolutionary history. This follows from the general evolutionary principle that organisms gradually lose abilities and traits that are no longer under positive selection, well-illustrated by things like blind cave fish or flightless birds on islands that lack predators(Diamond, 1986). Thus, imagine a species that has inhabited low-elevation environments for millions of years,adapting evolutionarily to function (reasonably) well in `normal' levels of atmospheric oxygen (∼21%). If one were to expose individuals of this species to high altitude, then they might be expected to exhibit inappropriate physiological responses to reduced atmospheric oxygen. The literature on human physiological responses to high altitude, both acute and chronic, is interesting in this context because it offers conflicting views on whether and to what extent various changes are adaptive versus maladaptive, and whether long-term, high-altitude native populations exhibit evolutionary adaptations to hypoxia (e.g. Winslow et al., 1989; Beall,2001; Brutsaert et al.,2005; Norcliffe et al.,2005; Wu et al.,2005). More generally, it is worth noting that the environment that many human beings experience (including aspects of nutrition, sanitation,medicine and the so-called built environment) has changed very rapidly relative to our generation time. Concomitantly, average lifespan has increased in many countries and diseases associated with aging have become much more common (e.g. Swynghedauw,2006). Therefore, it may be expected that at least some aspects of our phenotypic plasticity may not be adaptive.
To be or not to be: when should plasticity evolve?
Intuitively, plasticity might be good or bad, depending on the amount of spatial heterogeneity in the environment, the speed of temporal environmental changes, the predictability of spatial and temporal heterogeneity, and the size or duration of heterogeneity relative to an organism's mobility and lifespan. From a formal theoretical perspective, the evolution of plasticity has been studied with optimality models, quantitative genetic models, and gametic models (Scheiner,1993). Generally, all of these models suggest that adaptive plasticity will evolve when environmental heterogeneity exists, environmental cues about that heterogeneity are somewhat reliable, plastic responses confer a net fitness benefit, and the population contains some additive genetic variance for the plastic response (Berrigan and Scheiner, 2004). With regard to spatial variability,optimality, quantitative genetic and gametic models all predict further that plasticity is most favored when (1) inter-habitat variability is high, (2) all habitats are equally regular, (3) selection acts equally strongly across habitats, (4) the environmental cue-dependent phenotype is correlated with the environment of selection, (5) habitat selection is correlated with trait plasticity [for specific references, see elsewhere(Scheiner, 1993)].
Phenotypic plasticity is typically induced by environmental heterogeneity or environmental stress (Harshman et al.,1999; Wilson and Franklin,2002; Berrigan and Scheiner,2004; Gabriel,2005). In this context, `stress' is generally taken to mean anything that threatens physiological homeostasis (e.g. Sapolsky, 2005) and/or reduces Darwinian fitness. Environmental stress can be categorized into biotic(e.g. predator presence) versus abiotic (e.g. ambient temperature),and either type may cause changes in behavior, morphology, and/or physiology(Gabriel, 2005). If the mean fitness of individuals with plastic strategies exceeds the mean fitness of those with fixed strategies, then phenotypic plasticity or flexibility will tend to evolve (Scheiner,1993; Berrigan and Scheiner,2004; Gabriel,2005). Environment tolerance curves have been defined as `the response of a genotype's total fitness over an environmental gradient'(Lynch and Gabriel, 1987),distinguishing this as a special case of the norm of reaction, and using them to predict when irreversible plasticity will tend to evolve.
However, as noted elsewhere, `If stress periods are short compared to the life-time of an organism, then irreversible phenotypic plasticity is unlikely to be a favorable response'(Gabriel, 2005). Therefore,Gabriel proposed models predicting the selective advantage of reversible plasticity (phenotypic flexibility)(Gabriel, 1999; Gabriel, 2005). He concluded(Gabriel, 2005) that `... reversible phenotypic plasticity would be expected for all organisms under the following conditions: they are exposed to stress periods that last shorter than life span; stress appears in the long run with some regularity so that natural selection can shape non-induced and induced values of adaptive plastic traits.' In these models, he assumed that plasticity was not costly, with the rationale that `Plasticity costs would usually enter as constant factors that do not alter the optimal values of mode and breadth' [see p. 875 (Gabriel, 2005)]. He added the caveat that `if plasticity costs depend significantly on the amount of performed phenotypic change, then costs might become a function of the environmental state during stress in a way that the optimal values of mode and breadth are affected' (p. 875), but concluded by arguing that `given the predicted huge fitness advantages, the cost of plasticity would have to be unexpectedly high in order to counteract selection for reversible phenotypic plasticity' (pp. 880–881). Thus, it is important to remember Pigliucci's point on p. 483 (Pigliucci,2005) that `Research of costs of plasticity is still in its infancy, but is both theoretically important and empirically challenging, and should become a major area of future inquiry.'
Studying the evolution of plasticity
As discussed above, natural selection ought to affect plasticity, and organisms ought to vary in plasticity. How can we test such theoretical predictions? In general, the same way that we may seek to study adaptation in any sort of trait. Four general approaches to studying adaptation are commonly used by evolutionary biologists (e.g. see Huey and Kingsolver, 1993; Garland and Carter, 1994; Bennett, 1997; Futuyma, 1998; Schlichting and Pigliucci,1998; Feder et al.,2000; Orzack and Sober,2001; Pigliucci,2001; Freeman and Herron,2004; Stearns and Hoekstra,2005). First, as outlined in the previous section, real organisms can be compared with predictions of theoretical models, such as those based on optimality (e.g. Garland,1998; Orzack and Sober,2001). Second, examinations of the biology of natural populations can determine what sorts of traits vary, are heritable, and are currently under sexual or natural selection (e.g. Young et al., 2004). Experimental manipulations of putatively adaptive traits are often employed in such studies (e.g. Sinervo and Basolo,1996; Ketterson and Nolan, Jr,1999) and pp. 224-229 (Costa and Sinervo, 2004). Although several studies have attempted to quantify how natural selection acts on plasticity in the field (e.g. Trussell, 1997; Donohue et al., 2000; Nussey et al., 2005), this area of investigation will not be covered here. Third, one can compare species(or populations) that vary with respect to ecological factors that might cause variation in how selection `views' plasticity [overviews of studying adaptation via `the comparative method' and with a phylogenetic perspective have been published previously(Garland and Adolph, 1994; Garland et al., 2005)]. In the following subsection, we provide a brief summary of some comparative studies of plasticity. Finally, selection experiments(Bennett, 2003; Garland, 2003; Swallow and Garland, 2005)can be used to study adaptation, and this is our main focus, with emphasis on those that would qualify as `experimental evolution' (e.g. Rose et al., 1996; Rose et al., 2004; Ebert, 1998; Bennett, 2002; Bennett, 2003; Swallow and Garland, 2005)(http://en.wikipedia.org/wiki/Experimental_evolution).
Vertebrate morphology and physiology provide dramatic examples of both inter-specific and inter-trait variation in plasticity [plasticity of the water barrier in vertebrate integument is reviewed elsewhere(Lillywhite, 2004)]. With respect to variation among traits, vertebrate skeletal muscle(Flück, 2006) and the gastrointestinal tract (Secor,2005) are very responsive to use and disuse (`training' and`detraining' effects). Bone size, shape and architecture also change in response to variation in loading conditions, but to a much smaller extent than for muscle [for example, compare(Houle-Leroy et al., 2000)with (Kelly et al., 2006)]. Adult vertebrate lung also seems to have relatively low plasticity (e.g. Hoppeler et al., 1995; Weibel, 2000; Hsia, 2001; Henderson et al., 2002). In plants, one study shows that aspects of gas exchange may be more plastic than structural traits (Valladares et al.,2000).
Interspecific variation in plasticity has also been documented. In vertebrates, for example, attempts at aerobic exercise training (to improve cardiopulmonary and/or muscular function) of lizards have generally not been successful, even when patterned after those that cause large changes in mammals (Garland and Else,1987; Conley et al.,1995) (A. Szucsik, personal communication). In amphibians and squamates, species differences in gut plasticity seem to be related to their feeding ecology, in particular the frequency and/or regularity of feeding(Secor, 2005). In Burmese pythons, ventricular mass can increase 40% within 48 h after feeding, a change that is fully reversible (Andersen et al.,2005). Among species of fishes, carp and goldfish seem to be especially plastic (Cossins et al.,2006; Johnston,2006). In plants, a common-garden study of 16 shrubs in the genus Psychotria showed that species found in the understory, where light is less variable, showed less plasticity for a variety of traits as compared with species that generally occur in forest gaps, where light is more variable(Valladares et al., 2000). Population differences in plasticity have also received considerable attention in plants, with several studies suggesting that the they are indeed adaptive(e.g. Cook and Johnson, 1968; Donohue et al., 2000).
Selection experiments and experimental evolution
Selection experiments have provided valuable insights into central questions surrounding the evolution of phenotypic plasticity (see Scheiner, 2002). At their most basic, they have demonstrated that the plasticity of a trait is often heritable, capable of responding rapidly to selection, and determined by multiple genetic loci. [A genetic basis for the response to aerobic exercise training has also been demonstrated in human twin studies (references in Koch et al., 2005).] In addition, selection experiments have shown that plasticity (environmental sensitivity) of a given trait can evolve independently of the population mean value for that trait. More specifically, experiments with plants and invertebrates have shown plasticity to evolve in response to selection (1)directly on the reaction norm, (2) on a single trait in one environment, and(3) on a single trait across multiple environments. With respect to vertebrates, although many selection experiments have been performed, very few have focused on phenotypic plasticity as a component of the response to selection (Falconer, 1990; Scheiner, 2002).
The reaction norm has been directly selected upon in Drosophila melanogaster, the butterfly Bicyclus anynana, and the tobacco plant Nicotina rustica(Scheiner, 2002). We will only highlight the experiment performed by Scheiner and Lyman(Scheiner and Lyman, 1991),as it appears to be the most comprehensive and has also been reviewed in detail (Scheiner, 2002).
The stated purpose in Scheiner and Lyman's experiment(Scheiner and Lyman, 1991)was to determine if plasticity could respond to selection that was imposed under controlled and reproducible conditions. They began by capturing 301 individual D. melanogaster from the wild. These flies founded a stock that was maintained in the laboratory by mass culture at 21°C for 2–3 months (several generations), thus establishing a genetically heterogeneous base population. They then used 50 randomly chosen pairs to establish each of 14 separate experimental lines, which comprised two replicates of each of six selection regimes (increased thorax size at 19°C, decreased thorax size at 19°C, increased thorax size at 25°C, decreased thorax size at 25°C, increased plasticity, decreased plasticity) plus a control line that was not intentionally selected. To impose selection, plasticity was defined as the difference in average thorax length for sets of full-sibs raised at 19°C and 25°C. Plasticity did indeed respond to selection, but with a rather low realized heritability of 0.088±0.027 (mean ± s.e.m.). The authors concluded that the plasticity was not the result of overdominance, but rather a genetic interaction among multiple loci.
As noted above, the evolution of phenotypic plasticity has also been examined as a correlated response to selection on a specific trait in a single environment. For example, Harshman et al.(Harshman et al., 1991)studied detoxification enzymes in D. melanogaster. After establishing a base population from wild-caught flies, three Control (C) lines were reared on standard medium and three Selected (S) lines on lemon for 20 generations. For the lemon-cultured lines, the selection process was as follows: (1) flies were placed in bottles with freshly cut lemon (10 g, pesticide free) at room temperature for 7–10 days; (2) approximately 50% mortality occurred; (3)survivors were placed into a new bottle of freshly cut lemon (30 g) and vermiculite to produce the next generation. According to Harshman et al., the 50% mortality (during the lemon selection episodes) may have been caused by natural insecticide activity in lemons, or by toxins produced by bacteria growing on the fruit (Harshman et al.,1991). Flies (35–70) were randomly mated to produce the subsequent generation in all six lines. In the Control lines, flies were transferred to fresh medium for mating.
After 20 generations, all flies to be tested were reared on ordinary medium for one generation to standardize environmental conditions. They were then transferred to either lemon (which may induce the expression of detoxification enzymes) or fresh medium (to allow determination of enzyme activities under baseline conditions) for 24 h prior to sacrifice. Activities of epoxide hydrolases and glutathione S-tranferase (GST) were then measured. For GST measured using trans-stilbene oxide (TSO) as a substrate, S and C lines showed no significant difference for the sample exposed to fresh medium for 24 h, but the S lines showed substantially higher enzyme activities than C lines when lemon-exposed for 24 h (Fig. 3). Harshman et al. concluded(Harshman et al., 1991):`After 20 generations on lemon there was a pronounced change in environment-dependent expression... The response appeared independently in all three lines on lemon.' They also noted that: `In the present study the culturing regime used was ostensibly continuous, unless the process of lemon rotting every generation constitutes temporal variation. Normally, one would anticipate selection for change in environment-dependent enzyme expression to occur in variable environments but the results of the present study suggest it can evolve in a relatively constant regime.' Harshman et al. give additional examples in which the plasticity of an enzyme activity seems to have evolved as a correlated response(Harshman et al., 1999).
In numerous other selection experiments where plasticity could potentially evolve (because the selection regime is more than instantaneous), plasticity of the selected trait or of potentially related or subordinate traits does not appear to have been examined. In one such example(Bubliy and Loeschcke, 2005),correlated responses to selection for stress resistance and longevity in a laboratory population of D. melanogaster were examined. Several selection regimes were imposed in this experiment: cold-shock resistance selection, heat-shock resistance selection, heat knockdown resistance selection, desiccation resistance selection, starvation resistance selection and longevity selection. Here we discuss the cold-shock resistance selection line, for which selection was imposed for 21 generations. The selection regime was as follows. Flies were maintained on standard medium for 5 days at 11°C for acclimation, then placed in empty vials and exposed to 0.5°C for 27–50 h with relative humidity near 100%. Surviving flies were allowed to recover for 24 h at 25°C in vials with standard medium, then allowed to reproduce. We would argue that plasticity of traits that may support cold resistance in D. melanogaster could potentially be altered during the acclimation phase of this selection protocol. Furthermore,plasticity of traits that may support cold resistance may be evolving across generations. Although phenotypic plasticity apparently has not been investigated in these flies, we contend that selection experiments of this type should explore whether plasticity has increased as a component of the response to selection. In the following section, we discuss an ongoing experiment with house mice that has begun to examine plasticity in various traits as a potential correlated response to artificial selection for high voluntary wheel running, as expressed during days 5 and 6 of a 6-day exposure to wheels.
Selective breeding for high voluntary wheel running in house mice
Since 1993 our laboratory has been conducting a replicated selection experiment for high voluntary wheel-running behavior on days 5+6 of a 6-day wheel exposure. By housing mice from each of the four replicate S lines and from each of the four replicate C lines with or without wheel access for several days or weeks, we can test for differences in plasticity (training effects) in various traits. As outlined in the remainder of this section, we have found several traits that show greater differences between S and C lines when they are housed with wheel access than when they are housed without wheel access (or, in some cases, housed with access to wheels that are locked to prevent rotation). For some of these traits, the greater differences can be explained statistically by the greater wheel running exhibited by mice from S lines. For others, however, the differences seem to reflect greater plasticity in the S lines [i.e. for a given amount of stimulus (wheel running/day),individuals in the S lines show a greater response than in the C lines].
Animals and experimental protocol
The original progenitors (founding population) were outbred, genetically variable house mice (Mus domesticus) of the Hsd:ICR strain(Harlan-Sprague-Dawley, Indianapolis, IN, USA). After purchase from HSD, mice were randomly mated for two generations, paired, and then assigned randomly to eight closed lines (10 pairs in each). Four of these lines have been designated to experience selective breeding for high voluntary activity (lab designation, lines 3, 6, 7, 8) and four serve as controls (lines 1, 2, 4,5).
The selection protocol has been described in detail elsewhere(Swallow et al., 1998a). In brief, when each generation of mice are 6–8 weeks old, they are housed individually with access to running wheels (circumference=1.12 m) for 6 days. Daily wheel-running activity is monitored with photocell counters linked to a computer-automated system. Wheel activity is recorded in 1-min bins for 23–24 h of each of the 6 days of wheel access. For purposes of selection, wheel running is quantified as the total number of revolutions on days 5 and 6 of the 6-day test. After accounting statistically for any variation related to measurement block, age, wheel resistance, and sex,breeders are chosen. In the four S lines, the highest running male and female are chosen from each family as breeders to propagate the lines of the next generation. Within-family selection is performed to increase the effective population size (Ne), while reducing maternal and environmental variances, including effects of genotype-environment interactions (Henderson,1989). In the four C lines, breeders are randomly chosen from each family. Within all lines, sibling matings are disallowed.
By generation 16, the high-activity lines exhibited a 170% increase in total revolutions/day as compared with the C lines. This was caused primarily by S mice running faster rather than for more minutes each day, but the relative importance of the two components differs between the sexes, with females from the S lines typically showing little or no increase in amount of time running whereas males do show an increase in time running(Swallow et al., 1998a; Koteja et al., 1999a; Koteja et al., 1999b; Rhodes et al., 2000; Girard et al., 2001). This increase in wheel running greatly exceeds that of wild house mice born and raised under the same conditions (Dohm et al., 1994), and comes close to spanning the range of variation that has been reported among 13 species of wild murid rodents(Garland, 2003). Therefore, it seems that we have an evolutionarily `important' amount of divergence in wheel running between the S and C mice. Additionally, based on high-speed video analyses, estimates of instantaneous running speeds have shown that S line females run twice as fast as C line females, as well as more intermittently(Girard et al., 2001). However, since approximately generation 16, the differential in wheel-running distances has remained relatively constant, indicating that a selection limit or plateau may have been attained.
Plasticity of wheel running
Because our wheel-testing protocol is prolonged (6 days) rather than instantaneous (e.g. a few minutes), it is possible that the S lines may have evolved greater plasticity in this behavior. In other words, as compared with the C lines, mice from S lines might now exhibit a greater increase in wheel running across the 6 days of wheel access, given that only their performance on days 5+6 affects their probability of reproducing. Fig. 4 shows some hypothetical examples of how plasticity might be greater in the S lines. In Fig. 4A, Control lines show constancy of wheel running, whereas S lines increase monotonically across days 1–6; clearly, plasticity is greater in S lines. In Fig. 4B, wheel running is identical and increases gradually in both S and C lines over the first four days of testing. Selected lines then show a much greater increase between days 4 and 5, thus indicating greater plasticity during this time period. In these two cases, the greater plasticity of S as compared with C lines is reflected in the ratio of S/C (see Fig. 4A,B).
In the case of Fig. 4C, the interpretation is more complicated. Both S and C lines increase monotonically across days 1–6. On an absolute basis, S lines increase more (2000) than C lines (667) on each day. Relative to their own starting values, however, S and C lines increase by the same percentage each day, although this increase becomes smaller each day (40, 29, 22, 18 and 15%, respectively). As a result,the S/C ratio is a constant. Thus, whether one considers the S and C lines to differ in plasticity depends on whether absolute or relative values are considered.
Fig. 5 shows example data from our selection experiment, and the pattern resembles the one shown in Fig. 4C. As reported elsewhere(Belter et al., 2004), 48 female mice from generation 23 were studied. As shown in Table 1, S lines ran significantly more than C on every day. A repeated-measures ANOVA (SAS Proc Mixed with autoregressive error structure) indicated highly significant effects of day (P<0.0001) and line type (P=0.0001), but no significant day-by-line type interaction (P=0.7184). The foregoing results suggest that S lines do not exhibit a greater plasticity in wheel running.
|.||Total no. of revolutions|
|Trait .||Control .||Selected .||S/C Ratio .||P .|
|Day 6—Day 1||2112±1506||5658±1365||2.68||0.1047*|
|log10 (Day 6/Day 1)||0.2525±0.0723||0.2236±0.0649||0.7541|
|.||Total no. of revolutions|
|Trait .||Control .||Selected .||S/C Ratio .||P .|
|Day 6—Day 1||2112±1506||5658±1365||2.68||0.1047*|
|log10 (Day 6/Day 1)||0.2525±0.0723||0.2236±0.0649||0.7541|
Values are least-squares (adjusted) means ± s.e.m.(N=48).
P-values are for two-tailed tests comparing Control and Selected lines.
Data were analyzed by mixed-model nested ANCOVA in SAS Proc Mixed version 8. The effect of line type (selected vs control) was tested over replicate lines nested within line types (d.f.=1,6), and replicate lines were considered a random factor nested within line type. Age was entered as a covariate and `mini-muscle' was entered as an additional factor in all analyses; neither was ever statistically significant (all P>0.05).
If one statistical outlier is removed, the line type effect becomes significant (P=0.0325).
On the other hand, the difference between total revolutions on day 6 and day 1 was considerably higher, on average, for S lines (5658) than for C lines(2112). This greater absolute increase in wheel running across 6 days is not statistically significant (P=0.1047), but becomes significant(P=0.0325) when one outlier is removed. This was an S individual whose wheel running declined anomalously from 14 375 on day 1 to 7603 on day 6, the greatest decline for any mouse in the sample of 48. This may represent a real phenomenon, or it might indicate a problem with the wheel on day 6. We intend to explore the plasticity of wheel running more in future studies, with larger sample sizes. In any case, we believe that the greater increase in wheel running across the 6-day trial may well have biological significance,and may well have required coadaptational changes in one or more subordinate traits that support wheel running.
Apparent exercise adaptations in the high-activity lines
A main goal of the selection experiment was to identify traits that have evolved in concert with increased activity levels and that may be necessary for them, i.e. evolutionary adaptations for the high wheel running. Several considerations make this goal non-trivial. First, exercise physiology is complicated, and we have not examined all possible subordinate traits that could be key in terms of allowing high wheel running. Second, of those traits that have been examined, not all have been examined in the same generation. Some adaptations may have occurred in earlier generations and others in later ones, and indeed those occurring in later generations may even have supplanted some that occurred earlier. Third, adaptations may only exist, or at least be more developed, around the age at which wheel testing normally occurs, which is 6–8 weeks of age. Fourth, adaptations may only exist on days 5 and 6 of wheel testing, i.e. they require some days of wheel access to develop. Fifth, adaptations may to some extent be sex-specific, especially given that females in the S lines have increased total activity almost entirely by running faster, whereas males also show an increase in amount of time spent running. Given that we have not studied both sexes, at all ages, under all possible housing conditions (e.g. with or without wheel access), let alone in every generation, we may well have missed some key adaptations. With those cautions in mind, we have discovered a number of traits that seem to represent adaptations for high wheel running in the S lines. We review the motivational basis for high wheel running elsewhere(Rhodes et al., 2005).
Mice from the selected lines have higher maximal oxygen consumption during forced treadmill exercise (VO2max), especially in males (Swallow et al.,1998b; Rezende et al.,2006a; Rezende et al.,2006b) and higher insulin-stimulated glucose uptake in the extensor digitorum longus muscle [located in the hindlimb(Dumke et al., 2001)]. Mice from S lines have larger femoral heads and more symmetrical hindlimb bone lengths (Garland and Freeman,2005; Kelly et al.,2006). Interestingly, S lines exhibit reduced hindlimb muscle mass, especially in two lines that contain a Mendelian recessive allele that halves hindlimb muscle mass while increasing mass-specific aerobic capacity and having a variety of other pleiotropic effects(Garland et al., 2002; Houle-Leroy et al., 2003; Swallow et al., 2005; Syme et al., 2005; Kelly et al., 2006). The S and C lines differ with respect to many other traits as well, such as higher plasma corticosterone levels (Girard and Garland, 2002) and reduced body fat in S lines(Swallow et al., 2001; Dumke et al., 2001). We are currently attempting to determine which of these are adaptations that enhance wheel-running ability, as opposed to non-adaptive (and possibly even maladaptive) correlated responses.
Plasticity of exercise-related traits
Many traits (e.g. heart mass, VO2max) that one might expect to evolve as a correlated response to selection for high activity levels are also known to respond to the amount of exercise that an individual organism exhibits. Indeed, the literature on mammalian training effects is immense, in part because of our interest in competitive athletics but also because many exercise-related traits are known or thought to be important in promoting physical and/or psychological health(Booth et al., 2002; Castaneda et al., 2005)(Health Activity Center: www.cvm.missouri.edu/hac/index.html). Given that mice from the S lines run more than C when given wheel access, they might also be expected to exhibit greater training responses (physical conditioning) over a given period of time, such as several weeks.
Imagine that groups of both S and C mice were housed either without(sedentary group) or with (active group) access to a running wheel for 8 weeks(e.g. Swallow et al., 2005). Imagine further that for mice housed without wheels, we observed no difference in some phenotype, such as hematocrit. For the mice housed with running wheels, consider a hypothetical phenotype for which values are higher in the S lines (Fig. 6A); this can be explained, statistically at least, by their higher wheel running: a single regression line adequately describes the relation. In this case, we would interpret the data as indicating that mice from S and C lines are equally plastic: it seems to be a simple case of `more pain, more gain.' One real example of this pattern involves the level of brain-derived neurotrophic factor (BDNF) in the hippocampus of S and C mice after one week of access to running wheels [see fig. 2 (Johnson et al., 2003)].
Fig. 6B shows a different situation. When housed with wheel access, mice from S lines again have higher values for the phenotype, but we see no relation with the amount of running within either group. If we imagine further that S and C mice housed without wheels showed no difference (or at least values similar to those of C mice housed with wheels), then the S mice seem to be more responsive to wheel exposure, i.e. they are more plastic. For a given amount of exercise (wheel running), S mice experience a greater training response. Remember also that phenotypic differences between genotypes (e.g. S versus C mice) that appear only in some environments are termed genotype-by-environment interactions.
As in the hypothetical scenarios just discussed, we have published several papers that involved groups of both S and C mice housed with or without access to functional wheels for several weeks. We have studied various traits,including body mass, VO2max, organ masses, bone properties and enzyme activities (Swallow et al., 1999; Houle-Leroy et al., 2000; Thomson et al.,2002; Belter et al.,2004; Swallow et al.,2005; Kelly et al.,2006). We found a variety of responses in these phenotypes,including some that differ between the sexes. Some traits do not differ between S and C mice regardless of housing conditions [e.g. tail length,adjusted for variation in body mass, in both sexes(Swallow et al., 2005)]. Some traits were found to differ between S and C mice regardless of housing conditions [e.g. S mice are smaller in body mass but have relatively larger kidneys (Swallow et al.,2001; Swallow et al.,2005; Kelly et al.,2006)]. Others showed a difference between S and C lines when housed with wheels but not when housed without [e.g. hematocrit and blood hemoglobin content (Swallow et al.,2005)].
For traits that differ more between S and C lines when they are housed with wheel access, we can examine statistically which of the competing patterns shown in Fig. 6A,B better describes the data (see also Swallow et al., 2005). The general strategy is as follows. First, we identify a trait that shows a statistical interaction between the effects of line type(S versus C lines) and wheel access (the environmental factor). For these analyses, we use SAS Proc Mixed to implement a mixed-model, nested ANOVA(or ANCOVA if such covariates as age or body mass are included in the model),in which replicate line is a random effect nested within line type (S or C). Degrees of freedom for testing the effect of line type, the effect of wheel access, and the line type × wheel access interaction are all 1 and 6. With this type of analysis, one trait that showed a statistically significant interaction is hematocrit in a sample of 81 female mice housed with or without wheels for 8 weeks (Swallow et al.,2005), as repeated here in Table 2.
|Trait (± wheels) .||In likelihood (larger is better) .||Likelihood ratio testa .||No. parametersb .||AIC (smaller is better)c .||N .||S vs Cd .||Wheel access or revs in final weekd .||Interactiond .||Mini-muscled .|
|Trait (± wheels) .||In likelihood (larger is better) .||Likelihood ratio testa .||No. parametersb .||AIC (smaller is better)c .||N .||S vs Cd .||Wheel access or revs in final weekd .||Interactiond .||Mini-muscled .|
Values represent nested ANCOVAs comparing all (female) mice (S vsC lines housed with vs without wheel access for 8 weeks). Selected(S) and Control (C) lines for mice housed only with wheel access are then compared to determine whether S lines show greater phenotypic plasticity; see text for further explanation.
AIC, Akaike Information Criterion.
P<0.05. For cytrate synthase, this statistical difference indicates that the higher values exhibited by mice from the S lines can be explained as being a simple linear function of their greater amount of wheel running during the final week of wheel access. In contrast, for the other four traits, mice from S lines seem to exhibit greater phenotypic plasticity(training effects).
Twice the difference in likelihood is distributed as a χ2with 1 d.f., i.e. 3.841 for P=0.05. Values larger than this indicate that the model including amount of running during final week as a covariate(full model) fits the data significantly better than a model that does not include this covariate (reduced model).
Including intercept, fixed effects, random effects (even if estimated as zero), residual variance, and all covariates. Other covariates (not shown) in all analyses included log body mass (except for analysis of log body mass itself), age, time of day and (z-transformed time of day)2.
AIC calculated using all parameters in model, as indicated in previous column.
Degrees of freedom for S vs C, for wheel access vssedentary, and for the interaction of those two factors were 1 and 6 in all models. For effect of mini-muscle phenotype, d.f. were approximately 1 and 26-28, depending on sample size. In the analyses of wheel-access mice only,d.f. for amount of running during final week were approximately 1 and 26,depending on sample size; see Swallow et al.(2005) and text for further details. Signs indicate direction of effect: + indicates S>C, wheel access>sedentary, or positive effect of body mass.
P values for S vs C, Wheel access vs sedentary,Interaction of those two effects, and Mini-muscle are for two-tailed tests;values <0.05 effects are in bold type. P values for amount of running in final week are also for two-tailed tests.
Second, we examine the mean values for the four subgroups. In the case of hematocrit in females, adjusted means (SAS Proc Mixed) were 48.51, 48.76,48.66, and 50.90 for Control Sedentary, Control Active, Selected Sedentary and Selected Active, respectively [see table 3(Swallow et al., 2005)]. Thus, the line type effect is greater when mice are housed with wheel access. Indeed, separate ANCOVAs reveal no significant effect of line type(P=0.8502) for sedentary mice but a significant effect(P=0.0472) for the active group. Third, within the active group(Fig. 7), we can ask whether the data are better fit by a model that does or does not include the amount of wheel running as an additional covariate. For hematocrit in females, Table 2 shows that the ln likelihood of the nested ANCOVA model without wheel running (–75.7) is larger (less negative, in this case) than for the model with wheel running(–83.7). As the latter model contains one additional parameter(estimating the effect of wheel running), twice the difference in ln likelihoods (16.0, in this case) can be compared with a χ2distribution with one degree of freedom, for which the critical value for P=0.05 is 3.841. Therefore, the model with wheel running as an additional covariate yields a significantly worse fit to the data, and we conclude that the difference in hematocrit between S and C mice when housed with wheel access is not best explained as a simple function of the greater running by S mice. Instead, the greater training effect experienced by S mice seems to indicate that they have greater plasticity for this trait when given wheel access. Results are similar for blood hemoglobin content.
We are currently in the process of examining or reexamining a variety of traits, from several different published and unpublished studies, by the approach outlined in the previous paragraph. Table 2 shows some additional examples for the female mice studied elsewhere(Houle-Leroy et al., 2000; Swallow et al., 2005). We are finding a number of traits that fit the pattern described for hematocrit in females, including hemoglobin content of the blood as well as cytochrome c oxidase and pyruvate dehydrogenase activity in mixed hindlimb muscle in females. Thus, several traits seem to show greater plasticity in response to wheel access in S lines as compared with C lines. In addition, the studies by Houle-Leroy et al. and Swallow et al.(Houle-Leroy et al., 2000; Swallow et al., 2005) housed the `sedentary' mice with access to locked wheels (unable to rotate), and a subsequent study revealed that mice from S lines climb more than those from C lines in locked wheels. Therefore, the relative magnitude of training effects in S and C lines might differ if `sedentary' mice were housed in ordinary cages with no access to even locked wheels.
The results shown in Table 2are for (female) mice housed with or without access to a functional wheel for 8 weeks. However, the selection regime involves only 6 days of wheel access,so it will be crucial in future studies to see if a similar pattern emerges for shorter periods of wheel access. In fact, we have already found that, for some traits, the effects of wheel access can be dramatic even over a matter of days. For example, the amount of GLUT-4 glucose transporter in gastrocnemius muscle did not differ between S and C females when they were housed without wheels (Gomes et al., 2004). After 5 days, both groups exhibited an increase in GLUT-4, but the increase was much greater in S mice, such that S and C showed no overlap in values. When the amount of GLUT-4 was plotted against the amount of wheel running on day 5, the relation was similar to that shown in Fig. 6B. Thus, the difference between S and C lines in gastrocnemius GLUT-4 expression is not a simple linear function of the amount of wheel running; rather, mice from the S lines seem to have greater plasticity for this trait, and this greater plasticity can have large effects even in as few as 5 days.
Some traits show altered plasticity in the S lines, but in a complicated way. For example, the amount of neurogenesis in the hippocampus [see fig. 2D(Rhodes et al., 2003)] shows a relation similar to that depicted in Fig. 8. Mice from C lines (gray squares) exhibit a positive and quantitative relation with the amount of wheel running exhibited over several weeks, but this relation is lost in the S lines (black circles). Finally, we have also observed some traits that show an actual reversal of the direction of plasticity in S lines. For instance, relative ovary mass was found to be larger in S mice than in C when both were housed without wheels, but the opposite was true for mice housed with wheels for 8 weeks(Swallow et al., 2005).
Evolutionary change versus phenotypic plasticity
As noted above, some traits do not show a significant interaction between line type and wheel access. For these traits, the magnitude of any S versus C difference is relatively constant, regardless of housing conditions, and the magnitude of any training effect is similar in both S and C. Therefore, we can compare these two effects in a straightforward way plotting one versus the other.
Kelly et al. report hindlimb bone properties for male mice from generation 21 that were given wheel access for 8 weeks and compared with counterparts housed in ordinary cages with no wheels(Kelly et al., 2006). As shown in Fig. 9, bone lengths were not affected by either selective breeding or chronic wheel access. Diameters,in contrast, tended to be increased by both factors, with the magnitude of the evolutionary effect being somewhat greater than the training effect. In spite of the fact that mice from S lines ran considerably more than C, the magnitude of training effects was similar in S and C lines, indicating no genotype-by-environment interactions (see Kelly et al., 2006). Similarly, plastic and evolved (5–6 generations) responses of larval tracheae were in the same direction in Drosophila melanogasterexposed to different atmospheric oxygen content(Henry and Harrison, 2004)[for other examples in Drosophila, see(Price et al., 2003) and references therein]. An important area for future research will be to determine how often the underlying mechanisms of change are similar versus different for plastic and evolved responses [e.g.(Hodin and Riddiford, 2000)and references therein].
Whenever a selective event is more than instantaneous, individuals exposed to it have the opportunity to exhibit phenotypic change (plasticity) even as that event is in progress. If the direction of plastic change is such that it increases the probability of survival or other components of Darwinian fitness, then those individuals will be at a selective advantage. Hence, we may expect that directional selection will often lead to the evolution of increased plasticity in the appropriate direction(Fig. 2). This prediction would apply to organisms in nature or in laboratory or other experimental systems. Hence, the tools of selective breeding (sensu lato) can be used to study the evolution of phenotypic plasticity, just as they can be used to study any other complex phenotype(Swallow and Garland, 2005). Even if plasticity per se is not the intentional target of a selection experiment, we suggest that any experiment in which the selective event is more than instantaneous should explore whether plasticity has increased as a component of the response to selection. Our own experiment that involves selective breeding for high voluntary activity levels in mice has uncovered some examples of increased and apparently adaptive plasticity in the selected lines, but not for all traits (e.g. not for hindlimb skeletal dimensions). We believe that experimental evolution approaches have much to offer with respect to clarifying the evolution of plasticity, including self-induced adaptive plasticity(Houle-Leroy et al., 2000; Swallow et al., 2005) and the related concept of genetic assimilation(Price et al., 2003; Pigliucci et al., 2006).
Although we have argued here that plasticity might be predicted to increase in the appropriate direction as a correlated response to directional selection(e.g. see Fig. 2), the opposite prediction has also been offered. For example, in our original paper on enzyme activities, we noted that `genetic selection for increased voluntary wheel running did not reduce the capability of muscle aerobic capacity to respond to training' [see pp. 1608 and 1613(Houle-Leroy et al., 2000)]because, a priori, we had thought that organisms innately higher for a physiological function might exhibit relativity less ability to increase further in that direction via physical conditioning, perhaps because of some sort of `ceiling effect'. Similarly, Koch et al. in a training study of rat exercise capacity (Koch et al.,2005) noted that `According to the principle of initial values, if capacity is low then the percentage gain in capacity in response to training will be high, and vice versa.' However, their results for two inbred strains that differed in intrinsic aerobic running capacity were not in accord with the principle of initial value.
We thank A. F. Bennett, T. D. Brutsaert, A. G. Gibbs, R. B. Huey and E. R. Weibel for helpful discussions. We also thank L. G. Harshman, M. Pigliucci, T. Price, M. Van Sant and an anonymous reviewer for comments on the manuscript. Supported by US National Science Foundation grant IBN-0212567 to T.G.