Phenotypic Plasticity, Costs of Phenotypes, and Costs of Plasticity

Phenotypic Plasticity, Costs of Phenotypes,and Costs of Plasticity

Toward an Integrative View

Hilary S. Callahan,a Heather Maughan,b and Ulrich K. Steinerc

aDepartment of Biological Sciences, Barnard College, Columbia University,New York, New York, USA

bDepartment of Zoology, University of British Columbia,Vancouver, British Columbia, Canada

cDepartment of Biology, Stanford University, Stanford, California, USA

Why are some traits constitutive and others inducible? The term costs often appearsin work addressing this issue but may be ambiguously defined. This review distin-guishes two conceptually distinct types of costs: phenotypic costs and plasticity costs.Phenotypic costs are assessed from patterns of covariation, typically between a focaltrait and a separate trait relevant to fitness. Plasticity costs, separable from phenotypiccosts, are gauged by comparing the fitness of genotypes with equivalent phenotypeswithin two environments but differing in plasticity and fitness. Subtleties associatedwith both types of costs are illustrated by a body of work addressing predator-inducedplasticity. Such subtleties, and potential interplay between the two types of costs, havealso been addressed, often in studies involving genetic model organisms. In some in-stances, investigators have pinpointed the mechanistic basis of plasticity. In this vein,microbial work is especially illuminating and has three additional strengths. First, infor-mation about the machinery underlying plasticity—such as structural and regulatorygenes, sensory proteins, and biochemical pathways—helps link population-level stud-ies with underlying physiological and genetic mechanisms. Second, microbial studiesinvolve many generations, large populations, and replication. Finally, empirical esti-mation of key parameters (e.g., mutation rates) is tractable. Together, these allow forrigorous investigation of gene interactions, drift, mutation, and selection—all potentialfactors influencing the maintenance or loss of inducible traits along with phenotypicand plasticity costs. Messages emerging from microbial work can guide future effortsto understand the evolution of plastic traits in diverse organisms.

Key words: experimental evolution; selection analysis; phenotypic evolution; tradeoffs;life history theory; environmental heterogeneity

In this age of the genome, phenotypic traitssuch as behavior, morphology, and physiologyremain compelling to many researchers. Ecol-ogists, for example, are interested in connect-ing variation in organismal traits with com-munity and ecosystem patterns and processes(Eviner 2004; Miner et al. 2005). Developmen-

Address for correspondence: Hilary S. Callahan, Department of Bio-logical Sciences, Barnard College, Columbia University, 3009 Broadway,New York, NY 10027. Voice: 212-854-5405. [email protected]

tal and molecular geneticists also examine vari-ation in organismal traits, connecting this vari-ation with underlying genetic mechanisms andbiochemical pathways. Ecological geneticistsand other evolutionary biologists are also inter-ested in connecting phenotypes and associatedgenes, as well as in how both phenotypes andgenotypes are altered by multiple evolutionaryprocesses—such as natural selection, migrationand gene flow, functional tradeoffs among mul-tiple traits, pleiotropy, mutation, and genetic

Ann. N.Y. Acad. Sci. 1133: 44–66 (2008). C© 2008 New York Academy of Sciences.doi: 10.1196/annals.1438.008 44

Callahan et al.: Phenotypic Plasticity, Costs of Phenotypes, and Costs of Plasticity 45

drift (Pigliucci 2001; Lee 2002; Schlichting &Smith 2002; West-Eberhard 2003; Weinig &Schmitt 2004; Pigliucci et al. 2006; Ghalamboret al. 2007; Masel et al. 2007).

To examine the interplay among phenotypictraits, genes, and such processes, a primary taskfor most phenotypic research is simply to exam-ine the scope and pattern of trait variation. Thiscritical task is often complicated by the phe-nomenon of phenotypic plasticity: variation inenvironmental conditions eliciting variation inthe traits expressed by a given genotype. Con-ceptually, studying the phenotypic plasticity oftraits requires recognizing an organism as aduality, as both a phenotype and a genotype.Doing so often involves the concept of a reac-tion norm: a genotype’s range of phenotypesexpressed as a function of the environment(Sarkar 1999; Pigliucci 2001). Operationally,studying plastic traits and reaction norms re-quires a biologist to meet a series of challenges.First, one must identify and specify groupsof individuals with similar genotypes—clones,half-sibs, artificial selection lines, conspecifics.Then, one must characterize the phenotypes ofthese “replicate genotypes” when grown in twoor more environments. Next, one must decidewhether to focus not only on the different traitsexpressed in those environments but perhapsalso on the traits’ plasticities. That is, plasticitycan be conceptualized as a complex trait in andof itself. Finally, one must decide whether oneneeds to obtain information about underlyingphysiological and genetic mechanisms regulat-ing traits and their plasticity (Schlichting 1986;Via et al. 1995; Pigliucci 1996).

Woltereck, studying genetically homoge-neous lines of the small crustacean Daphnia

during the early 20th century, was among thefirst to grapple with the challenge of genotype–phenotype mapping. He documented the ef-fects of many different environmental factorson variation in head height, a continuouslyvarying quantitative trait of Daphnia’s exoskele-ton (Sarkar 1999). He is credited with coiningthe term Reaktionsnorm (Schlichting & Pigliucci1998), an idea that languished for decades be-

fore interest in phenotypic plasticity was rekin-dled during the mid-1960s and 1970s. Sincethen, the concept of the reaction norm has of-ten been used alongside other ecological andquantitative genetic techniques. It has been ahelpful unifying concept for empiricists andtheoreticians studying traits that exhibit phe-notypic plasticity (Via & Lande 1985; Endler1986; De Jong 1995; Rose & Lauder 1996).

Often, research in phenotypic plasticity hasprogressed by zooming out, ignoring the de-tails of the genes underlying a reaction norm.Instead, such work typically focuses on a par-ticular plastic trait or traits, examining how se-lection acts on them or investigating their effecton ecological performance. With this perspec-tive, many decisions must be made in translat-ing the phenomenon of phenotypic plasticityinto quantifiable terms. Often, models treat theplasticity of a trait as itself a complex, quantifi-able trait. Then, variation in traits and in theplasticities of the traits can be conceptualizedinto discrete, hierarchical categories. The firstissue is whether the trait is absent or present.If the trait is present, it is necessary to examinewhether the trait’s expression is constitutive orinducible (i.e., expressed in some environmentsbut not others).

Beyond categorizing a plastic trait as con-stitutive or inducible, it is often desirable toquantify inducibility by scoring trait expressionquantitatively across two or more environ-ments, sometimes along a gradient. Quantifi-cation of trait expression in multiple environ-ments can be useful for translating a trait’splasticity into a trait in and of itself, a trait thatis necessarily quantitative. Quantifying plastic-ity as a continuous trait has been carried outby calculating measures of spread (e.g., vari-ance, coefficient of variation) (Schlichting 1986)or by using the raw or standardized differ-ence between contrasting environments (Fal-coner 1990; Ungerer et al. 2003). In some situ-ations, the absolute value of differences is used.This may be an appropriate choice because itis clear that the direction or pattern of plas-ticity is bidirectional (i.e., passive phenotypic

46 Annals of the New York Academy of Sciences

plasticity, for which the range of responses isof greater interest than the directionality). Orit may be used because of insufficient knowl-edge about the details of a given syndrome ofplasticity (Scheiner & Berrigan 1998; Dewitt &Scheiner 2004; van Kleunen & Fischer 2007).Researchers who already know that the mag-nitude of plasticity varies, and that it typicallyshifts in one direction (i.e., active phenotypicplasticity), tend to quantify a trait in two en-vironments and to use the difference in traitvalues between environments as a metric ofplasticity. In such systems, plasticity can alsobe examined across an environmental gradi-ent, followed by fitting a reaction norm functionand using the function’s parameters (e.g., slope,intercept, higher-order curvature) to quantifyplasticity (Finlay & Wilkinson 1963; Gibert et al.

1998; Stratton 1998; Stinchcombe et al. 2004;Kingsolver et al. 2007). Any quantification ofplasticity allows for ranking genotypes in termsof greater or lesser “magnitude” or “level” ofplasticity and for analyzing plasticity as itself aquantitative trait.

Both theorists and empiricists have had tomake decisions about these and other method-ological details, which sometimes assume sub-stantial differences in the underlying biology orecological function. This is an essential step re-quired before developing or applying modelsaddressing whether the evolution and mainte-nance of plasticity depends on the details of theenvironment. Intuitively, and consistent withmany theoretical models, loss of plasticity andgreater stability are generally favored in morestable environments, whereas plasticity is gen-erally favored by heterogeneous or fluctuatingenvironments—but outcomes can be complexdepending on the reliability of the cue and/orthe sensory detection mechanism that organ-isms use to detect environmental fluctuations,by time lags, by details of the plasticity-elicitingand selective environments, and by costs as-sociated with plastic phenotypes (van Tien-deren 1991; Moran 1992; Scheiner & Callahan1999; Sultan & Spencer 2002; Zhang 2006;Kingsolver et al. 2007).

Indeed, as work in this vein has progressed,it has often been noted that heterogeneity inthe environment is ubiquitous (e.g., Lechowicz& Bell 1991), yet plasticity is neither univer-sal nor infinite. Indeed, many traits are stableor canalized rather than plastic, and the exis-tence of substantial genetic diversity tells us thatno genotype has evolved plastic traits so flex-ible that it can dominate in all environments(Tollrian & Harvell 1999; Pigliucci 2001). The-orists and empiricists have therefore often em-phasized “benefits” and “costs” to account forwhy plasticity versus stability may be selectedin different ecological contexts. Unfortunately,definitions and usage of the term costs arefrustratingly idiosyncratic. Sifting through themany reports with “phenotypic plasticity” inthe title or keyword list, one will find some ar-guing that “costs” contribute to natural selec-tion favoring plasticity, others arguing the exactopposite, and some even arguing both. In thisreview, we aim to clarify some of this confusionwhile still following the lead of many theoret-ical and empirical researchers specializing inthe evolution of phenotypic plasticity. We willexamine two basic yet distinct types of costs.On the one hand are “costs of the phenotype”and on the other hand are distinct “costs ofplasticity” that may accrue beyond costs of thephenotype.

Understanding and quantifying costs of thephenotype requires examining the evolutionaryconsequences of having one phenotype ratherthan another phenotype. That is, in a certainenvironmental context, a comparison betweendistinct phenotypes reveals different patternsof covariation between one or more quanti-fied traits and some other distinct organismalfunction. Many good examples of phenotypiccosts are found in the literature discussing an-tipredator defense traits in prey organisms. Thepotential benefit gained by expressing a de-fense trait may be offset by a cost—a decreasein an organismal function unrelated to avoid-ing, escaping, or resisting predators. A centralquestion is whether plastic trait expression al-lows organisms to avoid “paying the price” of


an inappropriately expressed trait in an envi-ronment where that trait is not advantageous.Where this is the case, an evolutionary advan-tage can be gained by shifting from expressingdefense traits constitutively to expressing themplastically only when there is risk of predation.Such costs are referred to by researchers in-vestigating many other ecological interactions,using many other terms: costs of defense, costs of

resistance, costs of induction, ecological costs, or evendirect costs (Levins 1968; Lynch & Gabriel 1987;Tollrian & Harvell 1999; Agrawal 2001). Re-gardless of such details, in this review we cate-gorize such costs as “costs of the phenotype” or“phenotypic costs.”

Conceptually distinct from costs of the phe-notype are costs of plasticity. One way to under-stand costs of plasticity is to consider genotypesthat have the same phenotype within an envi-ronment yet differ in their plastic responses to avariable environmental factor and in fitness. Insuch a situation, there is no phenotypic varia-tion, precluding phenotypic costs. Because thisis rarely the case in nature, and because thiscannot always be accomplished experimentally,plasticity costs are quantified using statisticaltools. After taking into account covariation be-tween a focal trait and a fitness-related trait, itis possible to examine residual variation in thefitness trait, variation for plasticity itself (ratherthan the trait), and covariation between thetwo (van Tienderen 1991; DeWitt et al. 1998;Scheiner & Berrigan 1998). Quantifying plas-ticity costs involves thinking about plastic traitsin a complex manner: the trait itself (as ex-pressed in one or more environments) and theplasticity of a trait (quantified using a variety ofmethods; see earlier discussion).

Although both plasticity costs and pheno-typic costs are characterized one environmentat a time, both can be examined in morethan one environment across a set of envi-ronments. Within individual environments, akey consideration is the relationship betweena trait’s phenotypic cost and its plasticity costbecause plasticity costs can offset phenotypic

costs. Specifically, finding that a trait’s plastic-ity is associated with a plasticity cost can po-tentially explain why the trait fails to be plastic(or shows suboptimal plasticity.) In this regard,plasticity costs can and should be examined inmore than one environment, because simula-tion studies suggest that plasticity costs are notlikely to counter the evolution of adaptive plas-ticity if they occur only “locally” but can beimportant if they occur “globally” (i.e., withinonly one environment across a set of environ-ments rather than in all or most environmentsacross a set of environments) (Sultan & Spencer2002).

The theoretical work of Sultan and Spencer(2002) shows that it is possible to think aboutplasticity costs and their implications whetherone is considering a trait’s plasticity or a trait’sstability. Commonly, one asks whether plas-ticity costs counterbalance across-environmentselection favoring plasticity (i.e., by reducingphenotypic costs). Or if one is more inter-ested in stability, one might flip the conceptand examine whether the “cost of stability” or“cost of canalization” counterbalances across-environment selection favoring stability (Dornet al. 2000; van Kleunen & Fischer 2007). Inthis review we will refer to this general phe-nomenon as a “plasticity cost,” but the sameconcept has been described with different terms(e.g., “indirect cost”). It has also been parsedinto specific subcategories such as genetic costs,maintenance costs, energetic costs, and sens-ing costs (van Tienderen 1991; Moran 1992;Newman 1992; DeWitt et al. 1998; Kussell &Leibler 2005).

Whether an investigator is studying pheno-typic costs, plasticity costs, or both, the litera-ture tends to focus on certain types of traits (gen-erally, plastic traits) and on populations withcertain types of genetic architecture (generally,genetic variation for traits and for the plastic-ity of traits, i.e., genotype–environment inter-actions). In general, when delving into this liter-ature, one also needs to bear in mind that manystudies of phenotypic costs have not addressed


plasticity costs. And some studies have focusedon plasticity costs only, paying little attention tophenotypic costs.

As a result, we currently lack an integra-tive understanding of these two types of costs,but we have important evidence for guidingprogress in this direction. With this objective inmind, we will highlight some recent work ad-dressing phenotypic costs only, some addressingplasticity costs only, and some addressing bothtypes of costs. We begin with work emphasizingphenotypic costs, most of it involving the studyof traits in prey organisms—putatively adaptivepredator-induced plasticity. A review of stud-ies addressing this phenomenon is useful fordemonstrating the many different factors thatcan complicate the quantification and interpre-tation of phenotypic costs, even when researchinto the issue of plasticity costs is postponedor ruled out as unimportant. Many differentenvironmental factors can change simultane-ously, multiple traits can show plasticity to thismultifaceted variation, traits can show plastic-ity at different points in the life cycle, traitscan vary in the time lags necessary for theirplastic expression, and some traits can showreversible plasticity. By focusing narrowly onone type of plasticity research, we will illustratewhy a sophisticated approach is essential forunderstanding just one type of cost: phenotypiccosts.

After considering the many possible compli-cations involved in detecting phenotypic costs,we turn attention to whether such complica-tions can explain why plasticity costs have beendetected in a few studies but found to be negli-gible in many others. We will highlight exper-imental strategies that have been particularlysuccessful for detecting costs of plasticity. Aninteresting question is whether it is possible tolink phenotypic costs and plasticity costs de-tected at the population level to the molecularmechanisms underlying traits and their plastic-ities. This question is of general interest eventhough pinpointing the gene or genes con-tributing to focal phenotypic traits (and theirplasticity) is probably more tractable in model

genomic organisms (e.g., Drosophila melanogaster,

Arabidopsis thaliana).Having introduced integrative approaches

for quantifying phenotypic costs and plastic-ity costs, we will discuss some recent studiesfocusing on microbial systems. In well-studiedmicrobes, most traits are plastic and many areamenable to environmental manipulations us-ing laboratory culture techniques. Such traitsinclude the use of particular sugars, the abilityto synthesize amino acids de novo, the devel-opment of flagella for motility, the formationof dormant spores from vegetative cells, andthe growth of fruiting body structures. All ofthese are examples of traits not constitutivelyexpressed but induced in the appropriate en-vironment(s). As well, for many microbes true“experimental evolution” studies can be per-formed for many generations, with large pop-ulations, and with replication. The potential tocarry out such studies allows direct testing oftheoretical models, particularly the predictionthat variable versus static environments indeedselect for or against plasticity, respectively. It isalso possible to obtain highly detailed informa-tion about, for example, the genes, sensory pro-teins, enzymes, and biochemical pathways—inshort, the molecular machinery responsible forphenotypic plasticity. As in work with othermodel organisms, work carried out with mi-crobes allows direct insight into how a plas-tic trait and its environment-specific expressionmaps to specific genes, noncoding regulatoryregions, or gene products within one or morebiochemical pathways. Simultaneously, empir-ical data about population sizes or mutationrates make it possible to probe the effect ofselection, mutation, and genetic drift on themaintenance or loss of plasticity and the as-sociated genes regulating that plasticity. Theseexamples should be of broad interest to re-searchers interested in reconciling the mecha-nistic details of plastic traits with the concepts ofphenotypic costs and plasticity costs, conceptsthat originally emerged from more “black box”approaches to modeling the evolution of plastictraits.


Costly Antipredator Phenotypes:Past Success Stories, Future

Challenges

For decades, prey organisms have been a fa-vorite study system for investigators interestedin understanding plastic traits from ecologicaland evolutionary perspectives. Prey organismsencounter variable and complex environments,exhibiting many different forms of plasticityin response to this variability. They often in-vest available time, carbon, energy, or other re-sources to different degrees into morphologicaland behavioral traits. Responses can be fairlygeneral against certain types of predators (VanBuskirk 2001; Relyea 2004) or can be adaptiveagainst one type of predator but maladaptiveagainst others (Mikolajewski et al. 2006). In ei-ther case, certain patterns of resource alloca-tion may be beneficial because they increasethe ability to evade or resist predators. Be-havioral responses—seeking shelter or reduc-ing activity—reduce the probability of detec-tion, encounter rates, or possibly both (Werner& Anholt 1993; Lima 1998). Morphologicaldefenses—neck teeth, helmets, spines, bulgybodies, shell shape, and thickness—increase thechance of prey survival in the event of an at-tack (DeWitt et al. 1999; Tollrian & Harvell1999; Kishida & Nishimura 2004; Mikolajew-ski et al. 2006). Other morphological structures,such as larger tail fins, improve the ability to flee(McCollum & Van Buskirk 1996). Physiologi-cal defenses such as toxicity are well charac-terized in plant–herbivore systems but are lit-tle explored in animal systems, though somehave been shown (Benard & Fordyce 2003).Finally, developmental responses in organismswith complex life cycles, such as earlier meta-morphosis, may reduce the time in predator-vulnerable life stages (Tollrian & Harvell 1999).Indeed, there is an almost overwhelming diver-sity of induced defenses across taxa in manytraits (Lima 1998; Tollrian & Harvell 1999; Lass& Spaak 2003; Relyea 2007).

For any given defense trait, the potential ben-efit gained by expressing the trait may be off-

set by a cost—a decrease in another importantorganismal function. Such functions might in-clude efficient feeding, rapid growth rate, op-portunities for mating, or defense against an-other type of predator. A common, general,verbal hypothesis is that natural selection fa-vors antipredator traits that are inducible ratherthan constitutive because inducible traits ac-crue fewer costs in environments that vary spa-tially or temporally in predation risk. Formaloptimality models are often developed to ex-amine inducible defense traits, and often thesetheoretical models assume that defense traitsare subject to time, energy, or other resourceallocation tradeoffs (Abrams 1984; Werner &Anholt 1993; McNamara & Houston 1994;Lima 1998; Steiner & Pfeiffer 2007). When two(or a set of) induced responses are all continuoustraits, the magnitudes of the plastic responsesand various tradeoffs can be quantified, allow-ing estimation of the phenotypic cost of eachresponse. This approach requires investigatorsto appropriately and explicitly define the traitor traits that are construed as costs—the “cur-rency” of the cost (Steiner 2007a; Steiner &Pfeiffer 2007). Empirically, quantifying trait-specific costs, identifying how traits are inte-grated, and investigating which traits trade offagainst each other has been challenging (VanBuskirk 2000).

Research addressing phenotypic costs andtradeoffs has often progressed with nonlethalpredator–prey experiments, involving cagedpredators in tanks, or mesocosms. Such stud-ies provide insights that cannot be gained usingfree-ranging predators, in which the conse-quences of within-population variation in de-fense traits are often confounded with the con-sequences of competitive release (i.e., popula-tion density decreases and resource increasesas predator-vulnerable individuals are removedfrom the population) (Van Buskirk & Yurewicz1998; Relyea 2007). Similarly, in many plant–herbivore systems, it can be difficult to distin-guish between an adaptive plastic response toherbivory and the negative effect of herbivoryon growth, competitive ability, and fitness (but


see Agrawal et al. 2002). Freshwater predator–prey systems have been particularly useful be-cause chemical cues often induce predator-defense traits, and these chemicals can be usedto trick the prey organisms to express the full setof behavioral, life-historical, physiological, andmorphological responses. It is possible not onlyto manipulate traits to assess and even quan-tify the extent to which the traits are plasticbut also to remove the potential benefit of theplastic trait.

Such well-targeted experiments are essen-tial for linking inducible traits to their potentialcosts, particularly when multiple tradeoffs actat the same time. In many organisms, the ener-getic cost of a response (i.e., the underlying en-ergy or carbon allocated) cannot be measureddirectly, but instead each response is gauged rel-ative to associated traits construed as costs. Inorganisms with complex life histories, account-ing for costs can be particularly challenging be-cause they accrue during a life cycle spanningmultiple stages (Van Buskirk & Saxer 2001).Also, the time of initiation matters. For ex-ample, predator-induced responses occurringlater during ontogeny may bear greater costs(Hoverman & Relyea 2007). Also, responsesmight be induced once and then maintainedwithout much extra cost, or they might becheap to initiate but costly to maintain (Tollrian& Harvell 1999). For instance, some predator-induced morphological defenses in Daphnia areinitiated in the mother’s generation, and oncethe morphological defense structure is built, itis cheap to maintain (Tollrian & Harvell 1999;Lass & Spaak 2003). In contrast, predator-induced morphological responses in tadpolescan be highly reversible, which is sometimesconsidered evidence that their maintenance iscostly (Kishida & Nishimura 2004).

Many studies look for associations betweencontinuous predator-induced phenotypes andfor reductions in a known component of fit-ness or an assumed fitness proxy—reduced re-production (Tollrian & Harvell 1999), reducedgrowth (Van Buskirk 2000), reduced rate of de-velopment (Tollrian & Harvell 1999; Relyea

2007), reduced immunity or reduced invest-ment in fat storage (Stoks et al. 2006b), orincreased mortality not caused by predation(Steiner 2007a). Simplistically, selection is ex-pected to eliminate costs that markedly reducean important fitness component. Yet such costscan persist because multiple costs or benefits areintegrated at the whole-organism level (Stearns1992).

It can be more challenging to explain whystudies sometimes fail to detect cost of defensetraits. Induced traits are expected to becomefixed if they do not involve phenotypic costs(Via & Lande 1985), but many defense traitsare induced by predators (or predator-relatedcues). However, costs might be eroded by se-lection over time, and studies may fail to de-tect them because they are infrequent or sub-tle. Also, empirical studies might overlook costsbecause they are not a major focus of the study.We use a review by Relyea (2007), who surveyed41 studies constituting 29 amphibian species ofpredator-induced responses involving shifts inthe time or size at metamorphosis in amphib-ian larvae, to illustrate these possibilities. Costswould be revealed by amphibians’ metamor-phosis being later and at a smaller size. Therewas no consistent evidence for costs. Whereasnine studies showed a phenotypic cost of re-sponding to predators, three studies revealedunexpected positive fitness effects, and 10 stud-ies showed combinations of costs and positivefitness effects, by being earlier and smaller (onestudy) or later and larger (nine studies). Despitecomparable experimental setups and manipu-lative environments, half the studies revealed noevidence for phenotypic costs associated witheither time to metamorphosis or size at meta-morphosis.

These findings make it difficult to judgeoverall phenotypic costs. In amphibians, sizeat metamorphosis is mostly linearly corre-lated with subsequent survival, whereas thecost of delayed metamorphosis increases ex-ponentially (Altwegg & Reyer 2003). Yet in halfthe studies that found later and larger meta-morphs and measured early growth, there was


reduced growth early during ontogeny followedby increased growth later. Although such catch-up growth has fitness costs in fish (Metcalfe &Monaghan 2001) and damselflies (Stoks et al.

2006a), evidence for costs of catch-up growthor delayed costs in postmetamorphic stagesof predator exposure during the larval phasehas not (yet) been found in amphibians (VanBuskirk & Saxer 2001; Altwegg 2002; Niciezaet al. 2006).

Despite the ambiguity of costs between stud-ies, results from studies using the same preyspecies and same predator species were con-sistent (Relyea 2007), indicating that variationsin costs are unlikely to be experimental arti-facts. Rather, there may be differences in costsbetween prey species and costs specific to thephenotypes induced by particular predators,which has been shown in other studies (VanBuskirk 2000, 2001; Relyea 2004). Variance inpredator- and prey-specific costs might arisefrom differences in selection strength, whichcould lead to variance in erosion of costs. Also,different predators pose different threats to dif-ferent prey organisms (Van Buskirk 2000, 2001;Relyea 2004). Also, between different prey andpredator species, induced responses differ intheir associated benefits or costs (Mikolajewskiet al 2006). Some of the equivocal evidencefor costs may be experimentally driven becausecosts are not necessarily linearly correlatedacross environmental factors (Steiner 2007a).Relyea (2007) reviewed experiments that wereconducted with many different species and un-der a range of conditions, and there might havebeen considerable variation among the stud-ies in how thoroughly the investigators quan-tified predator- and prey-specific relationshipsbetween environment-specific trait expressionand the trait or traits construed as costs (e.g.,growth, time, size at metamorphosis).

Evidence for costs during early ontogenycomes from one of the few studies that ex-plicitly investigated cost of induced defensesby comparing 15 anuran species in their re-sponse to exposure to chemical cues from inver-tebrate odonate predators (Van Buskirk 2000).

Growth costs were found in 13 species, and oneof the species that showed no growth cost (Hyla

chrysoscelis) showed high survival costs. This pro-vides one possible answer to the question of whysome studies focusing on only survival or onlygrowth would fail to detect costs. To quantifyand determine the origin of costs, Van Buskirk(2000) used an allocation tradeoff approach,with the expectation that the level of the (traitspecific) defense should be reflected in the levelof the cost. Contrary to the predictions, speciesthat showed increased costs of defense did notshow increased phenotypic responses in activ-ity, body length, or tail depth, although therewas a tendency for species that showed a strongreduction in activity when exposed to preda-tor cues to show some additional decrease insurvival. This lack of a correlation (betweenthe magnitudes of responses and costs) suggeststhat comparisons among species might not bean appropriate way to detect clear relationshipsbetween induced traits and the magnitude ofcosts, perhaps because costs of the responsediffer between species, as suggested by Relyea’s(2007) review.

Studies within one species that have relatedthe magnitude of response to the magnitudeof the phenotypic cost have also failed to re-veal clear relationships. For instance, a studyinvestigating induced defenses and their asso-ciated costs for Rana temporaria tadpoles alonga resource gradient revealed overall costs ofinduced defenses in survival and development(Steiner 2007a). Along the resource gradient,however, there was no simple relationship be-tween the induced defense and the traits con-strued as phenotypic costs of defense; theseshifted along the gradient. At low resourceavailability, costs resulted predominantly in re-duced survival, whereas at high resource avail-ability, costs yielded a reduced developmentrate. A study where prey density (competition)was manipulated instead of resources led to thesame conclusion: The level of the defense is notcorrelated with the level of the cost (Teplitskyet al. 2005). Despite clear evidence for costs ofdefense in these studies, defense traits and costs


were not linked in a simple, direct manner. Thisfinding returns us to the challenge of whole-organism phenotypes consisting of many dif-ferent traits.

The challenge of understanding costs andbenefits of plasticity in an integrated frame-work may be partially solved by building mod-els that investigate integrated trait responseswith multiple defense traits and their associatedphenotypic costs (i.e., Steiner & Pfeiffer 2007).Although such models are improvements com-pared with those examining one tradeoff, bothrely on assumptions about tradeoffs that mightnot be met. For instance, a reduction in forag-ing activity, shown by reduced growth rates, isone of the most effective defense mechanisms(Werner & Anholt 1993; Lima 1998) but is alsoassumed to be one of the highest costs of de-fense. However, experiments that used differenttime lags to disentangle behavioral, physiolog-ical, and morphological responses show thattadpoles that reduce their feeding activity un-der (nonlethal) predator exposure did not re-duce the amount of food ingested, evacuatedthe food from their guts at higher rates, anddid not show reduced growth rates (Steiner2007b). This finding shows that feeding ac-tivity is decoupled from ingestion and growth,potentially by physiological mechanisms suchas differences in conversion rates or metabolicrates. Similar results were found in a compar-ison of two damselfly species (McPeek 2004).Greater activity in one species did not trans-late into higher feeding rates, and both speciesingested the same amount of food, but highlevels of activity led to higher predation rates.The species differed in the conversion rate ofassimilated food under predation threats. In an-other experiment aimed at discovering underly-ing physiological mechanisms, it was confirmedthat predator-induced shell morphology in in-tertidal snails is caused by an active increasein calcification rate (Brookes & Rochette 2007).Improved understanding of underlying phys-iological mechanisms is critical, because thisknowledge can challenge common assumptionsfound in many models. It is therefore important

to recognize the tentative nature of interpreta-tions of single-species studies because they mayinvolve implicit or explicit assumptions aboutthe mechanisms of plasticity.

It may be time to move beyond simple cor-relation analyses between predator-inducibledefense traits and their costs to explorationof the mechanism(s) underlying tradeoffs, pos-sibly by taking advantage of established andemerging model organisms. The availability ofgood genetic information may permit explor-ing another potentially relevant issue: whetherthe same genes regulate both traits and plas-ticities, compared to separate genes regulatingthe plasticity of a trait but not the trait itself(e.g., Ungerer et al. 2003). This is an impor-tant issue because of its potential to influencewhether traits and associated plasticities canevolve independently (Scheiner 2002; Callahan& Pigliucci 2005). Indeed, preliminary resultsin the Arabidopsis model system are promising,and in time it should be possible to deepenour understanding of the links between ecolog-ically important traits (e.g., resisting or toler-ating predators) to the underlying physiologi-cal mechanisms (Banta & Pigliucci 2005) or toquantitative trait loci (QTLs), the chromosomalintervals harboring quantitative trait genes (e.g.,Weinig et al. 2003; also see review by Stinch-combe & Hoekstra 2008). Progress in attainingsuch longer-term goals may involve mappingof QTLs, followed by confirming that candi-date genes harbored within QTL intervals arein fact the genes involved in regulating plastictraits. Comparable progress in the arena of an-tipredator traits may eventually be feasible, andthe ease of crossing experiments in amphibianshas already been demonstrated (Laugen et al.

2005).

Assessing Phenotypic Costs andPlasticity Costs with Genetic

Model Organisms

Clearly, plasticity occurs not only in prey or-ganisms subject to predation but also in manyother species and in many other traits. Yet there


is no organism with a limitless ability to adjustits phenotype to match any and all environ-ments (i.e., perfect plasticity). The hypotheticalnature of such a phenotype, sometimes calleda “Darwinian monster” (Pigliucci 2001), hasoften led investigators to broaden their focusto examine not only phenotypic costs but alsoplasticity costs as a type of cost separable fromphenotypic costs. Both types of costs can pre-vent the evolution of perfect plasticity, and thetwo types of costs can reinforce or oppose eachother.

A quantitative genetics framework is oftenused to estimate and test whether there is selec-tion on plastic traits. In such analyses, one of thefirst and most basic goals is to test for genotype–environment interactions because this is evi-dence of genetic variation for plasticity anda prerequisite for its evolution (Via & Lande1985; Via et al. 1995). Because genetic correla-tions within and among environments theoret-ically constrain the evolution of plasticity (Via& Lande 1985), scoring and analyzing multipletraits and multiple plasticities is a feature foundin almost all published studies examining puta-tively adaptive phenotypic plasticity with thesemethods.

Investigators interested in plastic traits havemodified classical methods for analyzing se-lection (Lande & Arnold 1983; Endler 1986),using them to determine whether the direc-tion or strength of selection on a plastic traitdiffers among environments, an adaptive ex-planation for the evolution and maintenanceof phenotypic plasticity (e.g., Via & Lande1985; Dudley & Schmitt 1996). Two relatedmethods are commonly used, and both in-volve examining correlations (as estimated byselection differentials or selection gradients)between relative fitness and traits. One typeis an environment-specific analysis, which isperformed “locally”—within multiple environ-ments that evoke plasticity in at least one ofthe multiple traits. Another type is carried outacross environments, or “globally”—estimatesare obtained for each genotype’s mean fit-ness by appropriately averaging across environ-

ments, and the same procedure is used to es-timate each genotype’s trait mean. It can beargued that across-environment analyses areunrealistic unless they can somehow incorpo-rate information about the natural frequencyof alternative environments, but both within-environment and across-environment analysesare often found in the plasticity literature. Bothtypes of analysis address the issue of phenotypiccosts (i.e., whether traits expressed via plasticityare positively correlated with fitness), often bycomparing genotypes that vary widely in plas-ticity: Some may fail to strongly express a trait,some express the trait inducibly, and others ex-press it constitutively.

Because neither analysis separates out plas-ticity costs per se, a third and complementarytype of analysis is sometimes performed withplasticity cost analyses also conducted “locally”or within particular environments. This anal-ysis examines whether there is selection for oragainst the plasticity of a trait (or traits) afteraccounting for selection acting on directly thetrait(s) (van Tienderen 1991; DeWitt et al. 1998;Scheiner & Berrigan 1998).

Often, with data from one study, all threetypes of selection analysis can be carried out(DeWitt 1998; DeWitt et al. 1998), and it isfrustrating that relatively few studies combineall three types. One example is Stinchcombeet al.’s (2004) reanalysis of data for the modelplant species Arabidopsis thaliana. The data, orig-inally collected by Westerman and Lawrence(1970), and previously reanalyzed by Lacey et al.

(1983), focused on 21 inbred lines of naturalecotypes and for 12 mutant lines. Replicates ofeach genotypic line were grown in three differ-ent temperature environments.

After estimation of selection gradients withinthe temperature treatments, it was clear that thetraits expressed by more plastic genotypes wereassociated with lower genotypic mean fitnessthan the traits expressed by less plastic geno-types. Also, more plastic genotypes had lowerfitness averaged across environments than lessplastic genotypes; that is, there was across-environment selection against plasticity. Both


results are evidence that phenotypic costs werenot contributing to the maintenance of plastic-ity. Plasticity costs were found within two of thethree temperature environments. From thesethree perspectives in combination, it seems thatselection may favor reduction or elimination oftemperature-evoked plasticity in this species—because of a combination of phenotypic coststhat fail to maintain plasticity, combined withplasticity costs.

This issue brings us to a frequent and in-tuitively appealing argument: Selection oughtto eliminate costly plasticity (DeWitt 1998;DeWitt et al. 1998). Such an argument hasbeen upheld by simulation models (Sultan &Spencer 2002) and by many published studiesreporting their failure to detect significant plas-ticity costs (Scheiner & Berrigan 1998). Coun-terarguments are many, however, because somestudies have succeeded in detecting plasticitycosts. First, as argued by Stinchcombe et al.

(2004), selection may have a limited ability toeliminate maladaptive plasticity because it op-erates on multiple traits simultaneously. Thismay explain why their study found phenotypiccosts associated with a plastic trait (i.e., mal-adaptive plasticity), as well as plasticity costsseparated from phenotypic costs. Nonetheless,finding that a syndrome of plasticity is mal-adaptive may bias researchers against follow-up analyses to examine plasticity costs. Whileacknowledging this potential bias, van Kleunenand Fischer reviewed plasticity costs in plants(2005) and argued that costs have been de-tected often enough to merit continued investi-gation. These authors also discussed a secondbias that may prevent studies from finding plas-ticity costs: the use of insufficiently challeng-ing or realistic environments in experimentalstudies. Such treatments inflate environmentalvariance and therefore may obviate detectionof selection either on plastic traits (i.e., by di-minishing phenotypic benefits or costs) or onplasticity (i.e., by diminishing plasticity costs).

Although many authors have suggested thatplasticity costs may be larger or easier to detectin harsher environments (e.g., Steinger et al.

2003), few studies have been designed to di-rectly examine this idea. A notable exceptionis Steiner and Van Buskirk’s work (2008) withtadpoles. In addition to rearing tadpoles in con-ditions that elicit an antipredator phenotype,they also independently examined if plastic-ity costs varied across treatments with eitherhigh or low intraspecific competition. Plasticitycosts were not detected at the whole-organismlevel, and they were not consistently associ-ated with either plasticity or stability whenexamined at the level of individual plastictraits.

One impediment to progress in understand-ing plasticity costs is the challenge of com-paring studies. Each makes slightly differentchoices in quantifying plasticity and in decid-ing whether to focus only on active, adaptiveplasticity and plasticity costs or to be moreexpansive and to consider phenomena suchas passive plasticity, maladaptive plasticity, orcosts of canalization (van Kleunen & Fischer2007). Although reanalysis of past studies mightbe a productive project, doing so would re-quire cooperation in archiving and compil-ing raw data from previous and ongoing stud-ies. Another problem is that selection analysesare sometimes based on unrealistic or overlysimplistic assumptions about the heterogene-ity of plasticity-evoking environments and se-lective environments. Work by Stinchcombeet al. (2004), discussed earlier, involved across-environment selection analyses using simple,unweighted averages of genotypic trait meansand genotypic means for fitness. Yet severaltheoretical and simulation studies have demon-strated that the response to selection dependson the details of both the plasticity-elicitingand selective environments, which may not beidentical (Scheiner & Callahan 1999; Sultan& Spencer 2002; Zhang 2006). Unfortunately,few studies have attempted to characterize thespatiotemporal distribution of selective envi-ronments (Feder et al. 1997; Huber et al. 2004)or the environments known to trigger plastic re-sponses (Sultan et al. 1998; Scheiner & Callahan1999).


Even rarer are studies examining bothphenotypic costs and plasticity costs in fieldexperiments. However, field studies have de-tected plasticity costs associated with density-induced traits in annual mustards (Weinig et al.

2006; Dechaine et al. 2007). The findings ofWeinig et al. are probably directly tied to thegenetic material used rather than the organ-ism, traits, or ecological contexts examined.As the authors carefully explain, their studiesdid not use genotypes drawn from naturalpopulations. Instead, they used new experi-mental populations of segregating progenies—specifically, large populations of recombinantinbred lines. In natural populations, selectionmay have culled genotypes with unfavorablecombinations of plasticity loci and fitness loci,an argument echoing those of DeWitt (1998)in his influential work on plasticity costs. In anexperimental population of segregating proge-nies, some genotypes may have had fitness andplasticity loci in coupling phase (high plasticity,high fitness; low plasticity, low fitness), whereasothers may have these loci in repulsion phase(high plasticity, low fitness; low plasticity, highfitness). Beyond this argument, the many dis-tinct genotypes within such populations clearlyincrease statistical power for detecting plasticitycosts, and transgressive segregation enhancesgenetic variance for traits, for plasticities, andfor fitness (Callahan 2005; Weinig et al. 2006;Dechaine et al. 2007).

Studies with populations of recombinant in-bred lines are potentially amenable to QTLanalyses using traditional or array-based tech-nologies. QTL mapping of chromosomal re-gions can be performed for pairs of traits, mak-ing it possible to examine whether there areindividual loci (closely linked genes) that affecttraits and fitness, traits and plasticities, plastic-ities and fitness, or some combination of these(Callahan et al. 2005). This is an intermediatestep in pinpointing the specific QTL, as wellas the possible function of these genes withinregulatory pathways that detect environmen-tal inputs and that regulate developmental orphysiological responses to those inputs. With

this information in hand, it will be possibleto examine directly whether such genes con-tribute pleiotropically to fitness or perhaps in-teract with genes that contribute to fitness.

As previously argued by Agrawal (2001),knowledge of the genetic mechanisms of signaldetection and response is necessary for properlyinterpreting plasticity costs (or their absence).He discussed three hypothetical genotypes: oneplastic and two nonplastic. In one nonplasticgenotype, the sensory and physiological ma-chineries underlying plasticity are intact, exceptfor a defect in a small, downstream step. Despiteits static phenotype, it pays plasticity costs com-parable to those incurred by the plastic geno-type. The second nonplastic genotype derivesits static phenotype from a nearly completelack of plasticity machinery. Accordingly, it in-curs much lower costs. Distinguishing amongsuch alternatives requires integrative methodsfor characterizing plastic phenotypes, pheno-typic costs, and plasticity costs not only at thepopulation level but also at the level of geneticpathways operating within specific tissues andcells.

Interestingly, some of the earliest demonstra-tions of phenotypic costs and adaptive pheno-typic plasticity, carried out in plants, involvedcomparing wild types possessing a plastic traitwith mutant or transgenic knockout genotypeshaving a nonplastic phenotype—either lack-ing the trait or expressing the trait constitu-tively (Schmitt et al. 1995; Pigliucci & Schmitt1999, 2004). Similar demonstrations were per-formed with Drosophila melanogaster genotypesgenetically engineered to carry extra copies ofgenes for heat-shock protein 70 (HSP70) (Krebset al. 1998). A particularly innovative yet of-ten miscited and misinterpreted study (Krebs& Feder 1998) compared a D. melanogaster con-trol line to a transgenic line altered to synthe-size a “dummy” protein beyond the synthesis ofHSP70. This strategy essentially tried to super-impose an additional metabolic and energy ex-penditure (i.e., protein synthesis), a cost beyondthe normal HSP70 response. The researcherscould then compare the control and altered


genotypes in currencies more relevant to evo-lution and ecological function: survivorship andtime to metamorphosis. Their conclusion wasthat these energetic and metabolic costs werenegligible.

Plant researchers have also focused on well-characterized heat stress response systems toconduct integrative experiments addressingphenotypic costs and plasticity costs (Larkin-dale et al. 2005; Tonsor et al., 2008). Tonsor et al.

focused on HSP101, which responds rapidlyto temperature changes. Although there aremultiple heat-shock proteins in plants, loss ofHSP101 results in an inability to survive ex-treme heat stress. The study compared HSP101loss-of-function null mutants and wild types un-der both thermally benign and stressful condi-tions. In stressful conditions, protein contentvaried among wild types carrying a functionalgene, and this variation affected many pheno-typic traits—above- and below-ground, vege-tative and reproductive. Finding that the in-ducible HSP101 system can confer phenotypicbenefits rather than phenotypic costs was es-pecially intriguing because these phenotypicbenefits differed between two different geneticbackgrounds (the Columbia and Landsbergerecta lab strains), indicating strong epistasis. Inbenign conditions, loss of HSP101 functionalitysharply reduced reproductive output, indicat-ing pleiotropy.

In the same study, Tonsor et al. also surveyed10 wild-collected accessions of A. thaliana drawnfrom a latitudinal gradient. They found sig-nificant variation in temperature response ofHSP101 protein content, calling into questionthe notion that HSP101 is a consistently advan-tageous stress response system and raising thequestion of whether it is costly. Such a hypoth-esis has been previously advanced based on ar-guments about the energetic and metabolic de-mands required for synthesis of these proteins(e.g., Heckathorn et al. 1996). Having a costlyHSP rapid response system may be advanta-geous, but over time more frequent exposure toheat stress and selection to minimize such costsmay result in plants that cope with stress by

using less costly alternatives. Yet a functionalallele at the locus is maintained, perhaps be-cause the gene has such pervasive pleiotropiceffects.

Although it is informative to make compar-isons among multiple contemporary popula-tions, such studies entail implicit assumptionsabout the variable selection histories experi-enced by these populations. As already noted,the direct characterization of the plasticity-evoking potential and selective effect of hetero-geneity in contemporary environments is chal-lenging and has been attempted only rarely.Similarly characterizing the heterogeneity ofpast environments may be more problematic,if not impossible.

The limitations of such comparative and ret-rospective studies have led many researchers touse experimental evolution strategies, typicallywith rapid-cycling plants or animals (Scheiner2002; Callahan 2005; Garland & Kelly 2006).Such studies can be limited by their short-termnature, small population sizes, or lack of repli-cation. Accordingly, microbial organisms offera particularly attractive system for using experi-mental evolution to pursue questions about theevolutionary processes contributing to pheno-typic costs and plasticity costs.

Studying Costs with Microbes:Experimental Evolution and

Genomic Approaches

It has been known for many decades thatmicrobes cultivable in lab conditions are idealfor experimental evolution research strategiesgiven their short generation times (20 min)and large population sizes (>107). This poten-tial is also linked, for many different micro-bial species, to the availability of powerful ge-nomic and transcriptomic databases and tools.Moreover, microbial phenotypes are astonish-ing in their diversity and versatility, often in-cluding traits important for nutrient uptake ormetabolism, the ability to form spores in theface of nutrient scarcity, or motility and taxis


in response to a variety of stimuli. The pheno-types and genomes of microbes offer a powerfulframework in which phenotypic traits, plastic-ity, phenotypic costs, plasticity costs, and theirgenetic basis can be addressed in a direct andintegrative fashion. The remaining challenge isto understand the types of ecological contextsconfronted by microbial organisms and to de-vise methods for mimicking and manipulatingthese contexts in the laboratory.

In nature, some environments may supplyabundant macromolecules important for bac-terial viability (e.g., nucleotides, amino acids),but in other environments these are natu-rally scarce. Such contrasting conditions canbe mimicked in the laboratory by devising en-riched media that include abundant macro-molecules and contrasting them with medialacking one or more of these nutrients. Suchlaboratory methods can convincingly demon-strate that many bacteria can synthesize suchmacromolecules de novo from other, simplersources of carbon and nitrogen and that theseabilities are typically plastic rather than consti-tutive. When such macromolecules are presentin the environment, expression of the proteinsrequired for their de novo synthesis is typicallyrepressed. When these macromolecules are notavailable, maintaining fitness requires inducedexpression of these proteins. Many such re-sponses have been studied in microbes. In al-most every example studied, the phenotypicresponse is tightly regulated, requiring detec-tion and integration of external environmentalcues and, in turn, adjustments in the expres-sion of enzymes or other proteins involved inthe relevant metabolic pathways (for details,see Neidhardt et al. 1996; Sonenshein et al.

2002).Using a variety of microbial models, it is

therefore feasible to undertake the new researchstrategy of determining whether plastic pheno-types are maintained or lost over evolutionarytime (i.e., tens or thousands of generations) andto examine maintenance or loss of plasticity in avariety of environments. Such systems are idealfor examining whether there are plasticity costs

distinct from phenotypic costs. This examina-tion would not be carried out in an environmentlacking an essential macromolecule, where anonmutant parental genotype and a loss-of-function mutant genotype would differ pheno-typically. Rather, such studies are performedin permissive, stable environments where bothgenotypes have identical metabolic phenotypesbut might differ in fitness because of plasticitycosts. If significant, such plasticity costs wouldresult in mutant genotypes outcompeting andreplacing their parent genotypes, taking overthe population.

Such predictions about plasticity costs are infact long-standing ones, and several previousmicrobial studies have tested them. Zamenhofand Eichorn (1967) isolated Bacillus subtilis mu-tants that could synthesize neither histidine nortryptophan. Mutants were mixed and grownwith their wild-type parent strains, in an envi-ronment containing the amino acid they couldnot synthesize, and these mutants dominatedthe population within ∼50 generations. Thisfinding suggests that mutations that result inloss of macromolecular synthesis are advanta-geous when that macromolecule is constantlyavailable. However, the selective advantage as-sociated with the mutant is puzzling becausethe presence of histidine or tryptophan shouldrepress the expression of proteins required fortheir synthesis, so neither the mutant nor thewild-type strain is predicted to be expressingthem. Furthermore, the mutants were not reg-ulatory, so for leaky expression of unneededproteins, the mutant would be expressing a non-functional protein. Therefore, it is unlikely thatthe advantage is due to a difference in produc-tion costs. A lack of plasticity costs associatedwith amino acid production was also supportedby results from Dykhuizen (1978), who metic-ulously calculated the energy costs of makingthe amino acid tryptophan. He found that eventhough there were energy costs associated withmaking tryptophan, these were smaller than theselection differential observed between a mu-tant unable to make tryptophan and its wild-type parent.


Although both studies address whether thereis a cost associated with environmentallyregulated and induced macromolecular syn-thesis, we do not know if these lab-constructedmutants represent mutations that arise natu-rally. In a more realistic situation, where loss-of-function mutants arise spontaneously in apopulation, would selection favor an increasein their frequency? Would the loss of the abilityto synthesize macromolecules trade off for theoptimization of another phenotype? Maughanet al. (2006) addressed these questions in an evo-lution experiment where the bacterium Bacil-

lus subtilis evolved in nutrient-rich medium for6000 generations. Because this medium wasrich in macromolecules that would otherwiseneed to be synthesized, it was not surprisingthat they observed a decline or loss in the abil-ity of the population to synthesize all macro-molecules that were required for growth innutrient-poor medium.

Using experimentally measured populationsparameters such as mutation rates, competi-tive fitness, and fitness components (e.g., growthrate), Maughan et al. (2006) looked for an as-sociation between the decline in macromolec-ular biosynthesis and increases in fitness to de-termine whether selection was responsible forthe phenotypic loss. Although results from sta-tistical analyses suggest a role for selection inphenotypic loss, the fitnesses measured experi-mentally did not correlate with phenotypic loss.Perhaps the fitness assays performed did not ac-curately measure fitness, or perhaps selectionfor phenotypic loss was too subtle to detect inthe fitness.

Additional work using experimental popula-tions of bacteria has also shown that selectionfavors the loss of inducible phenotypes, suggest-ing that the maintenance of these phenotypes isindeed costly. The decay of metabolic breadthover evolutionary time was documented in pop-ulations of the bacterium Escherichia coli, whereit was found that the decay was attributableto tradeoffs, such that selection favored pheno-typic decay (Cooper & Lenski 2000). Loss ofthe genes whose products catabolize ribose was

also observed repeatedly in experimental pop-ulations of E. coli, suggesting that the loss of thisphenotype provides a fitness benefit (Cooperet al. 2001).

The results from the experiments describedsuggest that selection favors the loss of induciblemetabolic functions when these functions arenot important for fitness. This assertion in turnsuggests that these phenotypes are costly tomaintain. Whether it is more costly to moni-tor nutrients in the environment or to regulatethe expression of these phenotypes is not clearbecause usually the exact genes that are in-volved in phenotypic loss are not known. Forthe loss of the ability to catabolize ribose, thisloss was due to the deletion of genes encodingthe proteins for the transport and catabolismof ribose (Cooper et al. 2001). In most popu-lations, the gene whose protein product sensesribose in the environment, and consequentlyinduces expression of the proteins for ribosecatabolism, was deleted. However, 2 of 11 dele-tions did not include the gene encoding thesensing protein, leaving it unclear whether thesensing function was costly and/or whethera different component of the phenotype wascostly.

Most of the metabolic phenotypes discussedabove are relatively simple, requiring only afew proteins for both environmental sensingand metabolism. The evidence shows that themaintenance of these phenotypes is costly whenthey are not important for fitness. An impor-tant question is whether loss of phenotypes ofgreater complexity would be more strongly ad-vantageous. One of the most complex pheno-types in bacteria is the development of a vege-tative cell into a spore, and this phenotype mayhave high physiological costs. Many pathwaysin the cell are committed to sensing environ-mental change, and spore development is ini-tiated when the cell senses that nutrients arescarce. Environmental signals are integratedto increase the expression of early sporulationgenes. Once sporulation has been initiated itcannot be stopped; spores can become vege-tative again only if they complete the entire


sporulation process and then germinate. Bac-terial spores are metabolically dormant andcan withstand environmental assaults at a fre-quency much higher than that of vegetativecells.

Maughan et al. (2007) evolved 10 spore-forming populations of B. subtilis for 6000 gen-erations, five with and five without selectionfor spore development. For those five popula-tions without selection for sporulation, excessnutrients were added to the environment sothat sporulation would not even be initiated, al-though presumably environmental sensing wasstill occurring. Thus, any selection would bepredicted to remove plasticity costs associatedwith phenotype production, not sensing. Afterobserving that sporulation ability declined orwas lost in all experimental populations, it wasnecessary to determine whether mutation aloneresulted in loss of this complex trait or if selec-tion favored sporulation loss. To appropriatelyestimate corresponding selection coefficients,simulations were carried out to estimate rates ofneutral mutation accumulation, another expla-nation for loss of sporulation. Simulations wereconducted using an experimentally measuredrate of mutation. There was clear evidence thatselection favored the loss of sporulation (s =0.01) in only one population. In the remainingfour populations, simulations that incorporatedselection were no better at explaining the de-cline in sporulation than those that assumedneutral mutation accumulation. This findingsuggests that, even in a constant environment,sporulation was not a costly form of plasticityto maintain in most populations.

The lack of a cost associated with sporulationmaintenance seems contradictory to the selec-tive advantage that could potentially be associ-ated with the complex function. Sporulation is aphenotype that relies on the ordered expressionof hundreds of genes, and it might be expectedto incur a larger cost than a metabolic pathway,a phenotype usually encoded by fewer than 10genes. However, genes involved in the produc-tion of both sporulation and metabolism areexpressed only when sensory machinery recog-

nizes the appropriate environmental cue. Onlygenes whose products sense environmental cuesare constitutively expressed. In contrast, ex-pression of genes whose products are involvedin trait development occurs after the cue hasbeen sensed. Therefore, plasticity costs associ-ated with environmental sensing are likely tobe the same for all inducible traits, whetherencoded by hundreds or tens of genes.

Work with the B. subtilis system has not yetexplored the potential loss of sensory genes inbenign and static environments. However, aparallel issue was addressed in a study (Kussell& Leibler 2005) that addressed two differentphenomena often discussed as mechanisms forgenerating phenotypic variability within clonalpopulations of microbes: responsive switchingor stochastic switching. These strategies maponto what plasticity researchers sometimes re-fer to as “active” and “bet-hedging” plasticity(Kaplan & Cooper 1984). Responsive switchingis presumably more costly because it requiressensory mechanisms (i.e., a specific categoryof a plasticity cost) (see also Kussell et al. 2005).The work of Kussell and Leibler (2005) suggeststhat stochastic switching is likely to be favoredover responsive switching when environmentsfluctuate only rarely.

A main strength of these microbial studiesis their ability to focus narrowly on plasticitycosts by experimentally eliminating phenotypicvariation by using a benign and static environ-ment and then using this context to investigatefactors contributing to loss of plasticity-relatedgenes such as sensing genes. As a result,studies have tended to emphasize plasticitycosts per se, rather than phenotypic costs andplasticity costs occurring together. Yet phe-notypic costs associated with inducible traitscan be examined in microbes, as nicely illus-trated in a recent study examining the lossof flagellum-based motility in Pseudomonas flu-

orescens (Hall & Colegrave 2008). Mirroringwork with predator-induced traits in tadpoles,the study involved manipulating resource lev-els in the environment, combined with an ex-perimental evolution approach. This strategy


was successful in demonstrating that evolution-ary loss of motility occurred more rapidly inlow-nutrient environments, where flagella werepredicted to be more costly because of resourcelimitation, than in high-nutrient environments.This demonstration that flagellum develop-ment may be subject to costs that vary withresource availability is something that couldbe attempted with other traits such as sporu-lation ability or amino acid synthesis. A futurechallenge for work with microbes parallels thechallenges facing those working with plant oranimal systems: appropriately quantifying bothcost of phenotypes and costs of plasticities, mostlikely by doing so over multiple environments.

Over the long term and in natural environ-ments, there may be considerable ecologicalconsequences associated with loss of traits suchas sporulation or inducible metabolic pathways,because these traits are likely to be importantdeterminants of niche breadth. That is, thecomplete loss of a plastic trait can result in atransition from being a generalist to being muchmore of a specialist. This was shown to be thecase in experimental populations of E. coli thatevolved for 10,000 generations on one carbonsource (Cooper 2002). When each populationwas tested for its ability to use other carbonsources, growth on alternative carbon sourceswas on average lower. Furthermore, compet-itive fitness in four alternative environmentswas poor. These same E. coli populations alsohad reduced survival at different temperatures(Cooper et al. 2001). As in other organisms,however, interpretations of these results is con-tingent on how well the range of environmentsused in the laboratory reflect the actual rangeand frequency of environments encountered innature (e.g., alternative carbon sources tested inthe E. coli example or the nutrient availabilitiesused in studies of P. fluorescens).

All these examples have focused on a typicalexperimental evolution strategy in microbes:comparing genotypes in which an importantfunctional trait is either inducible or lost alto-gether. Such a comparison is extremely use-ful for examining plasticity costs, allowing di-

rect evaluation of how mutation and selectionbalance and contribute to either the mainte-nance or decay of a plastic trait. Sometimescomparisons can be made to examine costs ofthe phenotype. Studies carried out with traitsthat are either inducible or altogether absentwould be well complemented by comparisonsbetween genotypes unable to express a trait,able to express it only in certain environments,or expressing it constitutively. Although geneticmanipulations in the lab could probably cre-ate microbial genotypes with constitutive ex-pression, an important and interesting questionis whether such genotypes could arise sponta-neously. In either case, such genotypes couldbe useful for integrative microbial studies thatexpand beyond examination of plasticity costsin isolation to the study of plasticity costs com-bined with phenotypic costs.

Future Prospects

Whether working with animals, plants, ormicrobes, researchers investigating plastic traitswill continue to require familiarity with the phe-nomena of phenotypic costs and plasticity costsand need to be guided by previous theoreticaland empirical work addressing costs. We havetried to highlight the diversity of this vast andburgeoning literature, aiming to aid other re-searchers in discerning and developing creativeand promising directions for future research.

For example, studies are increasingly ask-ing questions about the mechanisms underlyingplastic traits. This type of work will probablybecome increasingly important and feasible asmore efficient and less expensive technologiesare developed for examining multiple genesor proteins, including those involved in sens-ing and responding to environmental signals.Currently, such work tends to be confined towell-established model organisms, but this nar-row focus is likely to broaden as the universe ofgenomic models rapidly expands. It is rapidlybecoming feasible to apply these tools as well toecological favorites such as black cottonwood


trees (Tuskan et al. 2006), the aquatic crus-taceans Daphnia (McClintock & Derby 2006),and emerging models closely related to existinggenomic models (Clauss & Koch 2006; Schranzet al. 2007).

Were Woltereck alive today, he could followup his careful and ultimately frustrating obser-vations of Daphnia reaction norms by zoom-ing in, focusing on physiological or geneticmechanisms that account for the patterns ofvariation. For those interested in such an ap-proach, there are clear steps to be taken. Whenstudying Daphnia and its predators, it may befeasible to pursue these types of questions byusing microarray-based studies of genetic vari-ation or gene expression. At the same time,it is a globally widespread organism that canbe investigated for basic ecological questions,as well as for more applied studies in pheno-typic plasticity (e.g., ecotoxicology). Finally, al-though scarcely comparable to bacteria, Daph-

nia can nonetheless be grown in lab cultureconditions in large populations with short gen-eration times.

Another emphasis in our review, and some-thing highlighted by other commentators, isthat it may be critical to understand not just themolecular genetics but also the population ge-netics underlying plasticity costs. This notion,emerging from studies of experimental popula-tions (e.g., recombinant inbred lines), is some-thing that can also be explored by working withexperimentally manipulated hybrid genotypesor with naturally occurring hybrids, such as hy-brid species and their progenitors or genotypesin hybrid zones. Such approaches will allow ex-ploring the importance of phenotypic plasticityin maintaining species boundaries or makingthese boundaries permeable to gene flow. Theywill also be generally useful for investigatinghow phenotypic costs and plasticity costs con-tribute to the processes that affect the evolutionof specialists and generalists (Lexer & Fay 2005;Picotte et al. 2007; Pinkhaus et al. 2007).

The genomics revolution notwithstanding,ecologists are continuing to study plasticityby using fairly traditional methods. In the

coming decade we expect to see many stud-ies of plasticity motivated by the importanceof understanding invasive species (Lee 2002;Richards et al. 2006) and how natural popula-tions will respond to climate change (Bradshaw& Holzapfel 2006). Such studies may serve asopportunities to refine our understanding ofhow plastic traits function and evolve, includ-ing the roles of phenotypic costs and plasticitycosts.

There will also be some work examiningplasticity in a wider diversity of organisms andin a phylogenetic framework. We are talkingnot so much about comparative studies focus-ing on a few species (Schlichting & Levin 1984;Pigliucci et al. 1999) but on much more ambi-tious surveys that are phylogenetically informed(e.g., Nicotra et al. 2008). Phylogenetically in-formed analyses may also be required when at-tempting to synthesize existing studies of plastictraits (e.g., Kembel & Cahill 2005), somethingthat has not yet been done in published meta-analyses of phenotypic costs and plasticity costs(Relyea 2007). Nonetheless, two main messagesare clear from early meta-analyses: There is ev-idence for phenotypic costs, even though theyare not universal, but evidence for plasticitycosts is equivocal. In part, this may reflect thefewer studies examining plasticity costs or a biasagainst publishing negative evidence. It mayalso stem from the difficulty of sifting throughreports that have made arbitrary or inconsis-tent (or unstated) decisions about how to de-fine, quantify, and analyze phenotypic costs andplasticity costs. It may also reflect a failure torecognize the difficulty of interpreting plasticitycosts without information about the phenotypiccosts associated with a trait (i.e., whether selec-tion favors loss of the trait, plasticity of the trait,or constitutive expression) (also see van Kle-unen & Fischer 2007). Finally, more studies ofplasticity in microbial organisms are forthcom-ing and should not be overlooked. This workis stimulating because it reminds us that plastictraits are affected not just by selection but alsoby mutation, drift, or migration. It will be in-teresting to see if those working with plants or


animals can devise and execute research that di-rectly examines how these different processes,together with phenotypic costs and plasticitycosts, jointly contribute to the evolution of plas-tic traits.

Acknowledgments

We thank Joanna Masel, Josh Van Buskirk,and an anonymous reviewer for constructivecomments. H.C. received financial supportfrom the National Science Foundation (IBN0344518). H.M. was supported by postdoctoralfellowships from the Killam Trust at UBC andthe National Institutes of Health. U.S. was sup-ported by NIA P01-AG0225000-01, MorrisonInstitute for Population and Resource Studies,and Swiss NSF PBZHA-110325.

Conflicts of Interest

The authors declare no conflicts of interest.

References

Abrams, P.A. 1984. Foraging time optimization and in-teractions in food webs. Am. Nat. 124: 80–96.

Agrawal, A.A. 2001. Phenotypic plasticity in the interac-tions and evolution of species. Science 294: 321–326.

Agrawal, A.A., J.K. Conner, M.T.J. Johnson & R. Walls-grove. 2002. Ecological genetics of an induced plantdefense against herbivores: additive genetic varianceand costs of phenotypic plasticity. Evolution 56: 2206–2213.

Altwegg, R. 2002. Trait-mediated indirect effects andcomplex life-cycles in two European frogs. Evol. Ecol.

Res. 4: 519–536.Altwegg, R. & H.U. Reyer. 2003. Patterns of natural selec-

tion on size at metamorphosis in water frogs. Evolution

57: 872–882.Banta, J.A. & M. Pigliucci. 2005. Effects of gibberellin

mutations on tolerance to apical meristem damagein Arabidopsis thaliana. Heredity 94: 229–236.

Benard, M.F. & J.A. Fordyce. 2003. Are induced defensescostly? Consequences of predator-induced defensesin western toads, Bufo boreas. Ecology 84: 68–78.

Bradshaw, W.E. & C.M. Holzapfel. 2006. Climatechange—evolutionary response to rapid climatechange. Science 312: 1477–1478.

Brookes, J.I. & R. Rochette. 2007. Mechanism of a plasticphenotypic response: predator-induced shell thick-ening in the intertidal gastropod Littorina obtusata. J.

Evol. Biol. 20: 1015–1027.Callahan, H.S. 2005. Using artificial selection to under-

stand plastic plant phenotypes. Integr. Comp. Biol. 45:475–485.

Callahan, H.S., N. Dhanoolal & M.C. Ungerer. 2005.Plasticity genes and plasticity costs: a new approachusing an Arabidopsis recombinant inbred population.New Phytol. 166: 129–139.

Callahan, H.S. & M. Pigliucci. 2005. Indirect conse-quences of artificial selection on plasticity to lightquality in Arabidopsis thaliana. J. Evol. Biol. 18: 1403–1415.

Clauss, M.J. & M.A. Koch. 2006. Poorly known relativesof Arabidopsis thaliana. Trends Plant Sci. 11: 449–459.

Cooper, V.S. 2002. Long-term experimental evolution inEscherichia coli. X. Quantifying the fundamental andrealized niche. BMC Evol. Biol. 2: 12.

Cooper, V.S. & R.E. Lenski. 2000. The population genet-ics of ecological specialization in evolving Escherichia

coli populations. Nature 407: 736–739.Cooper, V.S., D. Schneider, M. Blot & R.E. Lenski. 2001.

Mechanisms causing rapid and parallel losses of ri-bose catabolism in evolving populations of Escherichia

coli B. J. Bacteriol. 183: 2834–2841.De Jong, G. 1995. Phenotypic plasticity as a product of

selection in a variable environment. Am. Nat. 145:493–512.

Dechaine, J.M., J.A. Johnston, M.T. Brock & C. Weinig.2007. Constraints on the evolution of adaptive plas-ticity: costs of plasticity to density are expressed insegregating progenies. New Phytol. 176: 874–882.

Dewitt, T.J. & S.E. Scheiner. 2004. Modeling the evolu-tion of phenotypic plasticity. In Phenotypic Plasticity.T.J. Dewitt & S.E. Scheiner, Eds.: 82–97. OxfordUniversity Press. New York.

DeWitt, T.J. 1998. Costs and limits of phenotypic plas-ticity: tests with predator-induced morphology andlife history in a freshwater snail. J. Evol. Biol. 11:465–480.

DeWitt, T.J., A. Sih & J.A. Hucko. 1999. Trait compensa-tion and cospecialization in a freshwater snail: size,shape and antipredator behaviour. Anim. Behav. 58:397–407.

DeWitt, T.J., A. Sih & D.S. Wilson. 1998. Costs and limitsof phenotypic plasticity. Trends Ecol. Evol. 13: 77–81.

Dorn, L.A., E.H. Pyle & J. Schmitt. 2000. Plasticity to lightcues and resources in Arabidopsis thaliana: testing foradaptive value and costs. Evolution 54: 1982–1994.

Dudley, S.A. & J. Schmitt. 1996. Testing the adaptiveplasticity hypothesis: density-dependent selection onmanipulated stem length in Impatiens capensis. Am. Nat.

147: 445–465.


Dykhuizen, D. 1978. Selection for tryptophan auxotrophsof Escherichia coli in glucose limited chemostats as atest of energy conservation hypothesis of evolution.Evolution 32: 125–150.

Endler, J. 1986. Natural Selection in the Wild. Princeton Uni-versity Press. Princeton, NJ.

Eviner, V.T. 2004. Plant traits that influence ecosystemprocesses vary independently among species. Ecology

85: 2215–2229.Falconer, D.S. 1990. Selection in different environments:

effects on environmental sensitivity (reaction norm)and on mean performance. Genet. Res. 56: 57–70.

Feder, M.E., N. Blair & H. Figueras. 1997. Naturalthermal stress and heat-shock protein expression inDrosophila larvae and pupae. Funct. Ecol. 11: 90–100.

Finlay, K.W. & G.N. Wilkinson. 1963. Analysis of adap-tation in a plant breeding programme. Aust. J. Agr.

Res. 14: 742–&.Garland, T. & S.A. Kelly. 2006. Phenotypic plasticity

and experimental evolution. J. Exp. Biol. 209: 2344–2361.

Ghalambor, C.K., J.K. McKay, S.P. Carroll & D.N.Reznick. 2007. Adaptive versus non-adaptive phe-notypic plasticity and the potential for contempo-rary adaptation in new environments. Funct. Ecol.

21: 394–407.Gibert, P., B. Moreteau, J.R. David & S.M. Scheiner.

1998. Describing the evolution of reaction normshape: body pigmentation in Drosophila. Evolution 52:1501–1506.

Hall, A.R. & N. Colegrave. 2008. Decay of unused charac-ters by selection and drift. J. Evol. Biol. 21: 610–617.

Heckathorn, S.A., G.J. Poeller, J.S. Coleman & R.L.Hallberg. 1996. Nitrogen availability and vegeta-tive development influence the response of ribulose1,5-bisphosphate carboxylase/oxygenase, phospho-enolpyruvate carboxylase, and heat-shock proteincontent to heat stress in Zea mays L. Int. J. Plant Sci.

157: 546–553.Hoverman, J.T. & R.A. Relyea. 2007. How flexible is

phenotypic plasticity? Developmental windows fortrait induction and reversal. Ecology 88: 693–705.

Huber, H., N.C. Kane, M.S. Heschel, et al. 2004. Fre-quency and microenvironmental pattern of selectionon plastic shade-avoidance traits in a natural popu-lation of Impatiens capensis. Am. Nat. 163: 548–563.

Kaplan, R.H. & W.S. Cooper. 1984. The evolu-tion of developmental plasticity in reproductivecharacteristics—an application of the adaptive coin-flipping principle. Am. Nat. 123: 393–410.

Kembel, S.W. & J.F. Cahill. 2005. Plant phenotypic plas-ticity belowground: a phylogenetic perspective onroot foraging trade-offs. Am. Nat. 166: 216–230.

Kingsolver, J.G., K.R. Massie, J.G. Shlichta, et al. 2007.Relating environmental variation to selection on re-

action norms: an experimental test. Am. Nat. 169:163–174.

Kishida, O. & K. Nishimura. 2004. Bulgy tadpoles: in-ducible defense morph. Oecologia 140: 414–421.

Krebs, R.A. & M.E. Feder. 1998. Experimental ma-nipulation of the cost of thermal acclimation inDrosophila melanogaster. Biol. J. Linn. Soc. 63: 593–601.

Krebs, R.A., M.E. Feder & J. Lee. 1998. Heritability of ex-pression of the 70KD heat-shock protein in Drosophila

melanogaster and its relevance to the evolution of ther-motolerance. Evolution 52: 841–847.

Kussell, E., R. Kishony, N.Q. Balaban & S. Leibler. 2005.Bacterial persistence: a model of survival in changingenvironments. Genetics 169: 1807–1814.

Kussell, E. & S. Leibler. 2005. Phenotypic diversity, pop-ulation growth, and information in fluctuating envi-ronments. Science 309: 2075–2078.

Lacey, E.P., L. Real, J. Antonovics & D.G. Heckel. 1983.Variance models in the study of life histories. Am. Nat.

122: 114–131.Lande, R. & S.J. Arnold. 1983. The measurement of se-

lection on correlated characters. Evolution 37: 1210–1226.

Larkindale, J., M. Mishkind & E. Vierling. 2005. Plantresponses to high temperature. In Plant Abiotic Stress.M. Jenks & Y. Hasegawa, Eds. Blackwell. Oxford.

Lass, S. & P. Spaak. 2003. Chemically induced anti-predator defences in plankton: a review. Hydrobiologia

491: 221–239.Laugen, A.T., L.E.B. Kruuk, A. Laurila, et al. 2005. Quan-

titative genetics of larval life-history traits in Rana tem-

poraria in different environmental conditions. Genet.

Res. 86: 161–170.Lechowicz, M.J. & G. Bell. 1991. The ecology and ge-

netics of fitness in forest plants. 2. Microspatial het-erogeneity of the edaphic environment. J. Ecol. 79:687–696.

Lee, C.E. 2002. Evolutionary genetics of invasive species.Trends Ecol. Evol. 17: 386–391.

Levins, R. 1968. Evolution in Changing Environments; Some

Theoretical Explorations. Princeton University Press.Princeton, NJ.

Lexer, C. & M.F. Fay. 2005. Adaptation to environmentalstress: a rare or frequent driver of speciation? J. Evol.

Biol. 18: 893–900.Lima, S.L. 1998. Stress and decision making under the

risk of predation: recent developments from behav-ioral, reproductive, and ecological perspectives. Adv.

Stud. Behav. 27: 215–290.Lynch, M. & W. Gabriel. 1987. Environmental tolerance.

Am. Nat. 129: 283–303.Masel, J., O.D. King & H. Maughan. 2007. The loss of

adaptive plasticity during long periods of environ-mental stasis. Am. Nat. 169: 38–46.


Maughan, H., V. Callicotte, A. Hancock, et al. 2006. Thepopulation genetics of phenotypic deterioration inexperimental populations of Bacillus subtilis. Evolution

60: 686–695.Maughan, H., J. Masel, C.W. Birky & W.L. Nicholson.

2007. The roles of mutation accumulation and se-lection in loss of sporulation in experimental popu-lations of Bacillus subtilis. Genetics 177: 937–948.

McClintock, T.S. & C.D. Derby. 2006. Shelling out forgenomics. Genome Biol. 7: 312.

McCollum, S.A. & J. Van Buskirk. 1996. Costs and ben-efits of a predator-induced polyphenism in the graytreefrog Hyla chrysoscelis. Evolution 50: 583–593.

McNamara, J.M. & A.I. Houston. 1994. The effect of achange in foraging options on intake rate and pre-dation rate. Am. Nat. 144: 978–1000.

McPeek, M.A. 2004. The growth/predation risk trade-off: so what is the mechanism? Am. Nat. 163:E88–E111.

Metcalfe, N.B. & P. Monaghan. 2001. Compensation fora bad start: grow now, pay later? Trends Ecol. Evol. 16:254–260.

Mikolajewski, D.J., F. Johansson, B. Wohlfahrt & R. Stoks.2006. Invertebrate predation selects for the loss ofa morphological antipredator trait. Evolution 60:1306–1310.

Miner, B.G., S.E. Sultan, S.G. Morgan, et al. 2005. Eco-logical consequences of phenotypic plasticity. Trends

Ecol. Evol. 20: 685–692.Moran, N.A. 1992. The evolutionary maintenance of al-

ternative phenotypes. Am. Nat. 139: 971–989.Neidhardt, F.D., R. Curtiss II, J.L. Ingraham, et al., eds.

1996. Escherichia coli and Salmonella. ASM Press.Washington, DC.

Newman, R.A. 1992. Adaptive plasticity in amphibianmetamorphosis. Bioscience 42: 671–678.

Nicieza, A.G., D. Alvarez & E.M.S. Atienza. 2006. De-layed effects of larval predation risk and food qualityon anuran juvenile performance. J. Evol. Biol. 19:1092–1103.

Nicotra, A.B., M. Cosgrove, A. Cowling, et al. 2008. Leafshape linked to photosynthetic rates and temperatureoptima in South African Pelargonium species. Oecologia

154: 625–635.Picotte, J.J., D.M. Rosenthal, J.M. Rhode & M.B. Cruzan.

2007. Plastic responses to temporal variation in mois-ture availability: consequences for water use effi-ciency and plant performance. Oecologia 153: 821–832.

Pigliucci, M. 1996. How organisms respond to environ-mental changes: from phenotypes to molecules (andvice versa). Trends Ecol. Evol. 11: 168–173.

Pigliucci, M. 2001. Phenotypic Plasticity; Beyond Nature and

Nurture. The Johns Hopkins University Press. Balti-more.

Pigliucci, M., K. Cammell & J. Schmitt. 1999.Evolution of phenotypic plasticity a compara-tive approach in the phylogenetic neighbourhoodof Arabidopsis thaliana. J. Evol. Biol. 12: 779–791.

Pigliucci, M., C.J. Murren & C.D. Schlichting. 2006. Phe-notypic plasticity and evolution by genetic assimila-tion. J. Exp. Biol. 209: 2362–2367.

Pigliucci, M. & J. Schmitt. 1999. Genes affecting pheno-typic plasticity in Arabidopsis: pleiotropic effects andreproductive fitness of photomorphogenic mutants.J. Evol. Biol. 12: 551–562.

Pigliucci, M. & J. Schmitt. 2004. Phenotypic plasticity inresponse to foliar and neutral shade in gibberellinmutants of Arabidopsis thaliana. Evol. Ecol. Res. 6: 243–259.

Pinkhaus, O., S. Schwerin, R. Pirow, et al. 2007. Tempo-ral environmental change, clonal physiology and thegenetic structure of a Daphnia assemblage (D. galeata–

hyalina hybrid species complex). Freshwater Biol. 52:1537–1554.

Relyea, R.A. 2004. Fine-tuned phenotypes: tadpole plas-ticity under 16 combinations of predators and com-petitors. Ecology 85: 172–179.

Relyea, R.A. 2007. Getting out alive: how predators affectthe decision to metamorphose. Oecologia 152: 389–400.

Richards, C.L., O. Bossdorf, N.Z. Muth, et al. 2006. Jack ofall trades, master of some? On the role of phenotypicplasticity in plant invasions. Ecol Lett 9: 981–993.

Rose, M.R. & G.V. Lauder. 1996. Adaptation. AcademicPress. San Diego.

Sarkar, S. 1999. From the Reaktionsnorm to the adaptivenorm: the norm of reaction, 1909–1960. Biol. Philos.

14: 235–252.Scheiner, S.M. 2002. Selection experiments and the

study of phenotypic plasticity. J. Evol. Biol. 15: 889–898.

Scheiner, S.M. & D. Berrigan. 1998. The genetics of phe-notypic plasticity. VIII. The cost of plasticity in Daph-

nia pulex. Evolution 52: 368–378.Scheiner, S.M. & H.S. Callahan. 1999. Measuring nat-

ural selection on phenotypic plasticity. Evolution 53:1704–1713.

Schlichting, C.D. 1986. The evolution of phenotypic plas-ticity in plants. Annu. Rev. Ecol. Syst. 17: 667–693.

Schlichting, C.D. & D.A. Levin. 1984. Phenotypic plas-ticity of annual Phlox: tests of some hypotheses. Am.

J. Bot. 71: 252–260.Schlichting, C.D. & M. Pigliucci. 1998. Phenotypic Evolution:

A Reaction Norm Perspective. Sinauer Associates, Inc.Sunderland, MA.

Schlichting, C.D. & H. Smith. 2002. Phenotypic plastic-ity: linking molecular mechanisms with evolutionaryoutcomes. Evol. Ecol. 16: 189–211.


Schmitt, J., A.C. McCormac & H. Smith. 1995. A testof the adaptive plasticity hypothesis using trans-genic and mutant plants disabled in phytochrome-mediated elongation responses to neighbors. Am. Nat.

146: 937–953.Schranz, M.E., A.J. Windsor, B.H. Song, et al. 2007. Com-

parative genetic mapping in Boechera stricta, a closerelative of Arabidopsis. Plant Physiol. 144: 286–298.

Sonenshein, A.L., J.A. Hoch & R. Losick, eds. 2002. Bacil-lus Subtilis and Its Closest Relatives. ASM Press. Wash-ington, D.C.

Stearns, S.C. 1992. The Evolution of Life Histories. OxfordUniversity Press. New York.

Steiner, U.K. 2007a. Investment in defense and cost ofpredator-induced defense along a resource gradient.Oecologia 152: 201–210.

Steiner, U.K. 2007b. Linking antipredator behaviour,ingestion, gut evacuation and costs of predator-induced responses in tadpoles. Anim. Behav. 74:1473–1479.

Steiner, U.K. & T. Pfeiffer. 2007. Optimizing time and re-source allocation trade-offs for investment into mor-phological and behavioral defense. Am. Nat. 169:118–129.

Steiner, U.K. & J. Van Buskirk. 2008. Environmental stressand the costs of whole-organism phenotypic plastic-ity in tadpoles. J. Evol. Biol. 21: 97–103.

Steinger, T., B.A. Roy & M.L. Stanton. 2003. Evolutionin stressful environments II: adaptive value and costsof plasticity in response to low light in Sinapis arvensis.J. Evol. Biol. 16: 313–323.

Stinchcombe, J. & H.E. Hoekstra. 2008. Combining pop-ulation genomics and quantitative genetics: findingthe genes underlying ecologically important traits.Heredity 100: 158–170.

Stinchcombe, J.R., L.A. Dorn & J. Schmitt. 2004. Flower-ing time plasticity in Arabidopsis thaliana: a reanalysisof Westerman & Lawrence (1970). J. Evol. Biol. 17:197–207.

Stoks, R., M. De Block & M.A. McPeek. 2006a. Physio-logical costs of compensatory growth in a damselfly.Ecology 87: 1566–1574.

Stoks, R., M. De Block, S. Slos, et al. 2006b. Time con-straints mediate predator-induced plasticity in im-mune function, condition, and life history. Ecology

87: 809–815.Stratton, D. 1998. Reaction norm functions and QTL-

environment interactions for flowering time in Ara-

bidopsis thaliana. Heredity 81: 144–155.Sultan, S.E. & H.G. Spencer. 2002. Metapopulation

structure favors plasticity over local adaptation. Am.

Nat. 160: 271–283.Sultan, S.E., A.M. Wilczek, S.D. Hann & B.J. Brosi. 1998.

Contrasting ecological breadth of co-occurring an-nual Polygonum species. J. Ecol. 86: 363–383.

Teplitsky, C., S. Plenet, J.P. Lena, et al. 2005. Escape be-haviour and ultimate causes of specific induced de-fences in an anuran tadpole. J. Evol. Biol. 18: 180–190.

Tollrian, R. & C.D. Harvell. 1999. The Ecology and Evo-

lution of Inducible Defenses. Princeton University Press.Princeton, NJ.

Tonsor, S.J., C. Scott, I. Boumaza, et al. 2008. Heat shockprotein 101 effects in Arabidopsis thaliana: geneticvariation, fitness and phenotype under controlledtemperate growth conditions. Mol. Ecol. 17: 1614–1626.

Tuskan, G.A., S. DiFazio, S. Jansson, et al. 2006. Thegenome of black cottonwood, Populus trichocarpa (Torr.& Gray). Science 313: 1596–1604.

Ungerer, M.C., S.S. Halldorsdottir, M.A. Purugganan& T.F.C. Mackay. 2003. Genotype–environment in-teractions at quantitative trait loci affecting inflo-rescence development in Arabidopsis thaliana. Genetics

165: 353–365.Van Buskirk, J. 2000. The costs of an inducible defense in

anuran larvae. Ecology 81: 2813–2821.Van Buskirk, J. 2001. Specific induced responses to dif-

ferent predator species in anuran larvae. J. Evol. Biol.

14: 482–489.Van Buskirk, J. & G. Saxer. 2001. Delayed cost of an

induced defense in tadpoles? Morphology, hopping,and development rate at metamorphosis. Evolution

55: 821–829.Van Buskirk, J. & K.L. Yurewicz. 1998. Effects of preda-

tors on prey growth rate: relative contributions ofthinning and reduced activity. Oikos 82: 20–28.

van Kleunen, M. & M. Fischer. 2005. Constraints on theevolution of adaptive phenotypic plasticity in plants.New Phytol. 166: 49–60.

van Kleunen, M. & M. Fischer. 2007. Progress in thedetection of costs of phenotypic plasticity. New Phytol.

176: 727–730.van Tienderen, P.H. 1991. Evolution of generalists and

specialists in spatially heterogeneous environments.Evolution 45: 1317–1331.

Via, S., R. Gomulkiewicz, G. Dejong, et al. 1995. Adap-tive phenotypic plasticity: consensus and controversy.Trends Ecol. Evol. 10: 212–217.

Via, S. & R. Lande. 1985. Genotype–environment in-teraction and the evolution of phenotypic plasticity.Evolution 39: 505–522.

Weinig, C., J. Johnston, Z.M. German & L.M. Demink.2006. Local and global costs of adaptive plasticityto density in Arabidopsis thaliana. Am. Nat. 167: 826–836.

Weinig, C. & J. Schmitt. 2004. Environmental effects onthe expression of quantitative trait loci and impli-cations for phenotypic evolution. Bioscience 54: 627–635.


Weinig, C., J.R. Stinchcombe & J. Schmitt. 2003. Evolu-tionary genetics of resistance and tolerance to naturalherbivory in Arabidopsis thaliana. Evolution 57: 1270–1280.

Werner, E.E. & B.R. Anholt. 1993. Ecological conse-quences of the trade-off between growth and mor-tality rates mediated by foraging activity. Am. Nat.

142: 242–272.West-Eberhard, M.J. 2003. Developmental Plasticity and Evo-

lution. Oxford University Press. Oxford.

Westerman, J.M. & M.J. Lawrence. 1970. Genotype–environment interaction and developmental regula-tion in Arabidopsis thaliana. I. Inbred lines; description.Heredity 25: 609–627.

Zamenhof, S. & H.H. Eichhorn. 1967. Study of microbialevolution through loss of biosynthetic functions: es-tablishment of defective mutants. Nature 216: 456–&.

Zhang, X.S. 2006. The phenotypic variance within plastictraits under migration–mutation–selection balance.Evolution 60: 1125–1136.

Phenotypic Plasticity, Costs of Phenotypes, and Costs of Plasticity

Documents