the genetics of phenotypic plasticity. xiv. coevolution › rdholt › files › 297.pdfof natural...

The Genetics of Phenotypic Plasticity. XIV. Coevolution

Sa

1.W

Su

On

abeitfosimthabofphofevfittwouevplabmreanpo

Ketic

Atorociansoevitprsiv20evpo

*

Am00DO

vol . 1 8 5 , no . 5 the amer ican natural i st may 20 1 5

muel M. Scheiner,1,* Richard Gomulkiewicz,2 and Robert D. Holt3

Division of Environmental Biology, National Science Foundation, Arlington, Virginia 22230; 2. School of Biological Sciences,ashington State University, Pullman, Washington 99164; 3. Department of Biology, University of Florida, Gainesville, Florida 32611

bmitted May 30, 2014; Accepted December 23, 2014; Electronically published February 25, 2015

line enhancement: Fortran code.

stract: Plastic changes in organisms’ phenotypes can result from that plasticity is itself an evolving entity. In this article, we
ask two complementary questions: How does adaptationbyenisphexecseph
tospsotoWprWaninsoanligDcrpesothin20amhoalspstalornoity

her abiotic or biotic effectors. Biotic effectors create the potentialr a coevolutionary dynamic. Through the use of individual-basedulations, we examined the coevolutionary dynamic of two species

at are phenotypically plastic. We explored two modes of biotic andiotic interactions: ecological interactions that determine the formnatural selection and developmental interactions that determineenotypes. Overall, coevolution had a larger effect on the evolutionphenotypic plasticity than plasticity had on the outcome of co-olution. Effects on the evolution of plasticity were greater when theness-maximizing coevolutionary outcomes were antagonistic be-een the species pair (predator-prey interactions) than when thosetcomes were augmenting (competitive or mutualistic). Overall,olution in the context of biotic interactions reduced selection forasticity even when trait development was responding to just theiotic environment. Thus, the evolution of phenotypic plasticityust always be interpreted in the full context of a species’ ecology. Oursults show how the merging of two theory domains—coevolutiond phenotypic plasticity—can deepen our understanding of both andint to new empirical research.

ywords: coevolution, competition, mutualism, phenotypic plas-ity, predator-prey.

Introduction

daptation by natural selection often occurs in responseeffects of the environment on fitness. When those envi-nmental effects are caused by interactions with other spe-es, there is the potential for the species to coevolve, andextensive literature has explored this dynamic (Thomp-n 2005). The environment can play a second role in theolutionary dynamic beyond the determination of fitness;can shape the phenotype of the organism through theocess of phenotypic plasticity. Again, there is an exten-e literature about this process (DeWitt and Scheiner04). However, in all of the theoretical explorations of theolution of phenotypic plasticity, none has examined thessibility that the environmental factor responsible for

Corresponding author; e-mail: [email protected].

. Nat. 2015. Vol. 185, pp. 594–609. q 2015 by The University of Chicago.
20ou
03-0147/2015/18505-55516$15.00. All rights reserved.I: 10.1086/680552

This content downloaded from 128.227.186.1All use subject to JSTOR Te

plasticity versus genetic differentiation change when thevironment includes another, coevolving species? Howthe process of coevolution altered by the presence ofenotypic plasticity? We will address these questions byploring two modes of biotic and abiotic interactions:ological interactions that determine the form of naturallection and developmental interactions that determineenotypes.Coevolving systems that involve plasticity are ubiqui-us. For example, claw size in some crabs changes in re-onse to prey hardness (e.g., Smith and Palmer 1994), andme marine snails will alter shell morphology in responsethe presence of crabs (e.g., Appleton and Palmer 1988).hen some plants experience herbivory, they alter theocess of leaf development to produce tougher leaves (e.g.,alker et al. 1999) or other plant traits (Ohgushi 2012),d some insects developmentally alter their mouthpartsresponse to the toughness of their food (e.g., Thomp-n 1992). Plants can respond to changes in light quality,indicator of the presence of other plants competing forht, by plastically altering growth form and height (e.g.,udley and Schmitt 1996). Tadpoles can respond to de-eases in food availability caused by heterospecific com-tition (e.g., Pfennig 1992). Plants that formmutualistic as-ciations with root nodule–inhabiting bacteria can altere amount of resources directed to those bacteria, depend-g on the level of benefit being received (e.g., Simms et al.06). Indeed, phenotypic plasticity underlies many ex-ples of trait-mediated community interactions (Asche-ug and Callaway 2012; Schoener and Spiller 2012). Forl of these examples, the plasticity of only one of the twoecies in a given interaction was measured in a particularudy. Adding to this complexity, one or both species mayso have a plastic response to an abiotic variable as well asinstead of its coevolving partner. As far as we are aware,one has empirically examined the evolution of plastic-in any coevolutionary system (Robinson and Pfennig13), and we currently do not know what evolutionarytcomes to expect. Here we take a theoretical approach to

84 on Fri, 24 Apr 2015 07:08:31 AMrms and Conditions

http://www.jstor.org/page/info/about/policies/terms.jsp

generate expectations for a pair of interacting species dis-tributed along an environmental gradient. In doing so, weidsu

Inanfirjusespmspphacosplu

isinsibwofsptorespen

In all cases, an individual’s fitness is determined by itsphenotype, the abiotic environment, and the mean phe-nobithm

Fithitsofeit

Plasticity and Coevolution 595

entify which empirical systems might be most fruitful forch studies and what patterns are likely to be found.

Modeling Plasticity
coteotenExthexphw
totoprgrtyrempiesfrrabephdibe

our model of developmental plasticity in response toother species, the phenotype of one species is determinedst; that phenotype may be fixed or respond plastically tost the abiotic environment. Then the phenotype of thecond species is determined in response to that of the firstecies alone or also in response to the abiotic environ-ent. The phenotype of the first species may evolve in re-onse to the phenotype of the second species, while theenotype of the second species can both evolve and haveplastic phenotypic response to the first. Such plasticityuld directly influence the traits observed in the secondecies but also indirectly modulate the outcome of evo-tion in the first species.This asymmetric scenario for phenotypic determinationcaptured by the top half of the conceptual model shownfigure 1. This causal network encompasses the four pos-le ways that the phenotypes of species 1 and species 2ere determined in our model. In all cases, the phenotypesall individuals of species 1 are determined before those ofecies 2. Species 1 is either nonplastic or plastic in responsethe abiotic environment; species 2 is always plastic insponse to the biotic environment (the mean phenotype ofecies 1) and sometimes plastic in response to the abioticvironment.

Species 1


type of the other species. In our model, the effects of theotic environment—the biotic plastic cue for species 2 ande fitness effects for both species—are determined by theean phenotype of the other species experienced in a localmmunity. That is, we assume that a given individual in-racts with the aggregate effect of all individuals of theher species within the local deme rather than as a pairwisecounter between individuals (a mean field assumption).amples of such diffuse interactions include plants andeir pollinators, herbivory by wide-ranging herbivores, andploitative competition for pools of shared resources (e.g.,ytoplankton competing for nutrients in a well-mixedater body).An alternative approach to modeling plasticity would behave each species respond in a continuing plastic fashionthe other species. We decided against using that ap-oach for this study because it can lead to an infinite re-ess of individuals of each species altering their pheno-pes in response to changes by the other species. Such aciprocal response pattern could be modeled if develop-ent were treated as a multistage process with some stop-ng rule based on age or stage. Such an approach might bepecially appropriate if interactions among species emergeom repeated encounters between the same individualsther than on the basis of population means. It might alsoappropriate if trait plasticity was reversible so that theenotype of the individual changed as it interacted withfferent individuals (as would be appropriate for labilehavioral responses by individuals to other individuals).

Species 2

Genotype Genotype

Development

Selection

Fitness Fitness

Environment

Phenotype Phenotype

Environment

gure 1: Relationships between genotype, phenotype, and fitness in our two-species community. The fitness of each species is a function ofe abiotic environment and the phenotypes of both species. The trait value of each individual is determined by its genotype and possibly byplastic response to the abiotic or biotic environment. The phenotype of species 2 is always dependent to some degree on the phenotypespecies 1 (horizontal solid arrow). It may also depend on the abiotic environment (right dashed arrow). The phenotype of species 1 mayher be nonplastic or depend on the abiotic environment (left dashed arrow).



Lande (2014) presents a single-species model for plasticityevolution of a continually changing trait that might provideasuco

Fonafitencaparoty(spophatinarca(ede

aracottyroprfaim

enortoanlevaMprthfeonwtopesp

mop

the prey species maximizes its fitness by matching the abi-otic optimum while being as different as possible from theprpefoenof

596 The American Naturalist

template for developing such a model that fully encap-lates symmetric and time-varying plastic responses in aevolutionary context.

Fitness and Evolutionary Outcomes
mprfliabexmcaopneth
plwabvicuthtiotubephphtrphphplfatynom

auineninfosptophtedipoalof

ro

r the ecological interactions that determine the form oftural selection (bottom half of fig. 1), we assume that theness of a given phenotype at a given location along anvironmental gradient may be influenced both by the lo-l abiotic environment and by local interactions with thertner species. We model the effect of the abiotic envi-nment on fitness by a Gaussian function of the pheno-pe with an optimum that can vary from deme to demeee below for details). We assume that the biotic com-nent of an individual’s fitness depends on both its ownenotype and the mean phenotype of the other speciesa particular location. We further assume that the traitseach species that influence the interspecific interactione on a common scale if similar in type (e.g., body size) orn be relativized to a common scale if different in type.g., levels of a toxin in a prey species and the ability totoxify that chemical in a predator species).We explored selection in a metacommunity of demesrayed linearly along an environmental gradient. Selectionted via survival, which was determined by both the abi-ic and biotic environments and an individual’s pheno-pic state. Making survival a function of the abiotic envi-nment provides a stabilizing force on the coevolutionaryocess. It prevents runaway selection when coevolutionvors differences in trait values between species and min-izes drift when it favors matching trait values.Making survival a function of both the abiotic and bioticvironments creates the potential for either augmentingconflicting selection. Augmentation would be expectedoccur, for example, when the interaction is mutualisticd determined by phenotype matching. In this case, se-ction drives both species toward matching optimal traitlues based on the abiotic environment. For example, inüllerian mimicry, two poisonous species that share aedator will converge on the same warning signal, andat warning signal may differ among locations with dif-rent visual environments. Conversely, competition basedphenotype matching results in conflict, if individuals

ith similar phenotypes compete more strongly; selectionward the shared abiotic optimum is opposed by com-titive interactions driving phenotypes of the competingecies apart.For predator-prey interactions, we assume the predatoraximizes its fitness when it can match both the abiotictimum and the phenotype of the prey species, whereas


edator’s phenotype. For example, selection might be ex-cted to act on the times of peak activity during the dayr both a predator and its prey. Abiotic factors (e.g., ambi-t temperature) could influence the fitness consequencesactivity time. If the time that is best for avoiding ther-al stress is also when the prey is active, selection on theedator is augmentative (but not for the prey). This con-ct can be reversed if the optimal phenotypes due to theiotic environment differ for the predator and its prey. Forample, at one end of the abiotic gradient, the prey speciesay be selected to be large and the predator to be small be-use of selection from the abiotic environment, with theposite at the other end of the gradient, even though fit-ss of the predator is maximized when it is large wherevere prey is large.This interplay between augmentation and conflict alsoays out in selection on phenotypic plasticity, especiallyhen plasticity is in response to both the biotic and theiotic environments. Plasticity is favored when the en-ronment at the time of development provides a reliablee about the environment at the time of selection (theeory of the evolution of phenotypic plasticity proposi-n 4; see appendix in Scheiner 2013). In the case of mu-alism, the abiotic and biotic signals augment each othercause the fitness of both species is maximized when theirenotypes match each other and both match the optimalenotype determined by the abiotic environment. In con-ast, consider the case of competitive species where theenotype of the nonplastic species matches the optimalenotype determined by the abiotic environment. For theastic species, selection by the abiotic environment wouldvor plasticity that matches the abiotic optimal pheno-pe, whereas selection by the biotic environment (the phe-type of the other species) would favor plasticity thatoves the phenotype away from the abiotic optimum.Our article investigates how these potential patterns ofgmentation and conflict play out in a two-species coevolv-g community distributed along an environmental gradi-t. First, we consider all three types of species ecologicalteractions—competition,mutualism,andpredator-prey—r the case of a single type of phenotypic determination:ecies 1 nonplastic and species 2 plastic only in responsethe biotic environment. Second, we consider all types ofenotypic determination for a single type of ecological in-raction—predator-prey—because those revealed the mostsparate outcomes across the interaction types. A full ex-sition of all possible ecological interactions crossed withl types of phenotypic determination are beyond the scalewhat can be presented here.Adaptive evolution of plasticity requires a variable envi-nment. In this article, we examine an abiotic environ-



ment that varies over space but not through time. Wefocused on spatial abiotic variation for two reasons. First,wsutecibrcoonshatploftiositScinof(eattu

sppeoftivonrevoandidifulikloisulte20alinmpediinva

spmDeninby

dient with a spatially varying phenotypic optima, higherdispersal rates increasingly favor plasticity (Scheiner andHefineaerticdivaaan

minnodityon20heevaber

The Model

Otivnovitrsu

lovemexenatfegowarwticpl

dedipa


ith regard to coevolution, we wanted to compare our re-lts with those of Nuismer et al. (2010) concerning pat-rns of phenotypic correlation between interacting spe-es distributed across communities in a landscape. Moreoadly, we are interested in how plasticity might modulateevolution in a landscape mosaic (Thompson 2005). Sec-d, with regard to phenotypic plasticity, Scheiner (2013)owed that the combination of spatial and temporal vari-ion can sometimes lead to highly complex patterns ofasticity evolution. That article also compared the effectspure spatial variation with those of pure temporal varia-n and showed that they can have very different propen-ies to select for plasticity (for a detailed discussion, seeheiner 2013). Even in a constant environment, speciesteractions on their own can lead to fluctuating strengthsselection as a result of unstable population dynamics.g., predator-prey cycles). Because this article is an explor-ory study, such added complexities are best left to a fu-re manuscript.Dispersal is critical to the process of adaptation to aatially variable environment for several reasons. First, dis-rsal tends to increase the local heritable genetic variationquantitative traits (Barton 1999), thereby enabling rela-ely larger evolutionary responses to local selection. Sec-d, the local fitness of a dispersing individual is inverselylated to the distance between the optimal phenotype fa-red in its environment of origination and developmentd that favored in its new location. (In our model, speciessperse along an environmental gradient so that dispersalstance is related to the distance between these optima; therther individuals disperse, the more maladapted they areely to be.) In addition, immigrants impose a migrationalad by mating with residents. The net effect of movementthus to depress mean local fitness and to push local pop-ations away from their optimum; this increases the in-nsity of local directional selection (e.g., Polechová et al.09). In general, given density dependence, dispersal canso indirectly influence fitness and selection by perturb-g local population size (Gomulkiewicz et al. 1999); in theodels we use, however, we assume a form of density de-ndence where this does not happen. All of these effects ofspersal on adaptive evolution apply to any trait—includ-g phenotype plasticity per se—and to patterns of spatialriation in fitness determined by abiotic or biotic causes.Beyond these effects, dispersal plays an additional roleecifically with regard to the evolution of plasticity. In ourode, we assume that movement occurs before selection.ispersal then creates intergenerational variation experi-ced by a given lineage, because some descendants will bea range of environments broader than that experiencedtheir ancestor. When selection acts along an abiotic gra-


olt 2012; Scheiner 2013). In the context of this study, thatfect might differ, however, because both species are mov-g, possibly changing how that variation is perceived bych species. It is the combination of variation among gen-ations and certainty within a generation that favors plas-ity (Scheiner and Holt 2012). In our simulations, varyingspersal rate—and thus how differences in environmentalriation are experienced over many generations—providesuseful tool for revealing how different types of ecologicald developmental interactions respond to spatial variation.It is far from straightforward to intuit how dispersalight affect coevolution in response to spatial variabilityboth the biotic and the abiotic environments, given phe-typic plasticity. Previous studies of interacting speciesstributed over space show that in the absence of pheno-pic plasticity, dispersal can have important influencescoevolution in some cases (e.g., Gandon and Nuismer09) but not others (e.g., Nuismer et al. 2010). As we showre, the effects of dispersal and, indeed, the outcomes ofolution (including plasticity) are especially complex wheniotic and biotic sources of environmental variability gen-ate conflicting patterns of selection across space.

Model Structure

ur model is an individual-based simulation of quantita-e trait evolution for two interacting species with discrete,noverlapping generations distributed along a single en-ronmental gradient. The model was implemented in For-an 77 (see supplementary material, available online). Ammary of parameters is given in table 1.The genotype of an individual consisted of two types ofci: (1) genes with deterministic expression (i.e., no de-lopmental noise) that was independent of the environ-ent (nonplastic loci) and (2) genes with deterministicpression that was dependent on the abiotic and/or bioticvironment (plastic loci). As limiting cases, spatial vari-ion in adaptation can occur by two routes: genetic dif-rentiation in which the allelic values of the plastic locito 0 (i.e., are not expressed) or phenotypic plasticity in

hich the allelic values of the nonplastic loci go to 0 (i.e.,e not expressed). Intermediate outcomes are possible inhich individuals express the optimal phenotype in a par-ular environment through nonzero values of both theastic and the nonplastic loci.The metacommunity consisted of a linear array of 50mes (indexed by i from 1 to 50). The length of the gra-ent was chosen to minimize edge effects and allow com-risons with previous studies (Scheiner and Holt 2012;



Scheiner 20demes at or near the ends of the gradient were not includedincr(pfr1otcophplnospbiwjuensis

byphprsufiranthcufogrhi

(t

ibutedadditively to the trait. Allelic values at the plastic loci werembevieawsp(credothtim

wvenovaqureorticbeththvief

lora

Table 1: Summary of the model parameters

PaBiM5%Sa

Fi5.1502.1Se1010.11020


any calculations.) An abiotic environmental gradient waseated by varying the optimal value (vi) of a single traithenotype) in a linear increasing fashion along the arrayom 22.45 arbitrary units at one end of the gradient to2.45 arbitrary units at the other; that is, the optimal abi-ic phenotypes in adjacent demes differed by 0.1 units. Themmunity consisted of two interacting species. Either theenotype of species 1 was not plastic or it responded in aastic manner to the abiotic environment only. The phe-type of species 2 was always plastic, with the plastic re-onse due to the biotic environment only or to both theotic and abiotic environment. For comparative purposes,e also modeled cases where the community consisted ofst a single species that was plastic in response to the abioticvironment only and cases where the community con-ted of two species, neither of which was plastic.The life-history pattern was as follows: birth, followeddevelopment (i.e., the phase in the life cycle when theenotype is determined), then selection, dispersal, and re-oduction. Rates of dispersal and reproduction were as-med to be invariant across the gradient. This is the “selectst” life-history pattern of our previous models (Scheinerd Holt 2012; Scheiner 2013). For this life-history pattern,e environment of development provides a very reliablee to the environment of selection. Our previous modelingr this life history showed that in response to just an abioticadient, plasticity is strongly favored at intermediate togh dispersal rates (Scheiner and Holt 2012).For development of species 1, an individual’s phenotyperait value) was determined by 10 unlinked diploid loci:


ultiplied by an environment-dependent quantity beforeing summed over all allelic values. The effect of the en-ronment (Ei for deme i) on the phenotypic contribution ofch unit of plastic allelic value varied in a linear fashionith position along the array, with a slope of 0.01 units;ecifically, we used in our simulations Ei p 0.01(i2 25.5)hanging these particular parameters values in effect justscales dispersal relative to the slope of the gradient andes not alter our qualitative conclusions). For species 1,e phenotype of each individual was determined at thee of development as

T1ij p okp1, 10

N1ijk 1Ei okp1, 10

P1ijk, (1)

here T1ij is the phenotype of the jth individual that de-lops in the ith deme, N1ijk is the allelic value of the kthnplastic allele of that individual, and P1ijk is the alleliclue of the kth plastic allele. For a given genotype, theantity oN1ijk can also be thought of as the intercept of itsaction norm at the point along the gradient where Ei p 0the phenotype of the individual in the absence of plas-ity, and the slope of (Ei)oP1ijk calculated across demes canthought of as the slope of its reaction norm. Thus, bothe optimal phenotype due to the abiotic environment ande reaction norm due to plasticity were linear across en-ronments. If species 1 was not plastic, the environmentalfect was set to 0.When new offspring were generated, each allele at eachcus mutated with a probability of 10%. (Lower mutationtes changes the timescale over which evolution happens

Parameters Values

rameters explored:
Source of plasticity cue
84 on Fri, 24 Aprms and Condit

otic versus abiotic
Type of species interactionDispersal rate
utualism, competition, predator-prey

Direction of selection gradient for each species
–84%me versus opposite
xed parameters:
No. nonplastic and plastic loci each Steepness of gradient (change in optimum in adjacent demes) units Length of environmental gradient Strength of selection within demes (j)
demesunits

Strength of biotic interaction (a)
Life-history pattern lection before dispersal Population size 0 individuals/deme/species Mutation rate %/allele/generation Mutational effect (standard deviation) units No. generations ,000 No. runs per parameter combination
13). (Our conclusions would not change if the five nonplastic loci and five plastic loci. The loci contr

r 2015 07:08:31 AMions


rather than the eventual outcome for the models consideredhere [Scheiner andHolt 2012].)When amutation occurred,th(mprmBoquthju

decoreinof

wanspmthleoftr

sutwtimspcobithan

wTi

thofcrthThtyfo1

cies). (Eq. [3b] is the phenotypic-matching fitness functionof Nuismer et al. [2010]. For a visualization of its form, seefigleapsepafit

petiode(wananScityvamlikdimnaw

pamduthcaprsithefcoexneredeinditrbe

eainno2Evvaer10af


e allelic value was changed by adding a Gaussian deviateean of 0 and a standard deviation of 0.1 units) to theevious allelic value (i.e., this is a continuum of allelesodel; Kimura 1965). Allelic values were unconstrained.th the plastic and the nonplastic loci—and the subse-ent phenotypes—could take any value from2∞ to∞, andose phenotypes could be the result of just the plastic loci,st the nonplastic loci, or some combination of both.For development of species 2, the phenotype was againtermined by five nonplastic and five plastic loci, but inntrast to species 1, the phenotype of this species couldspond plastically to the mean phenotype of species 1 and,some cases, the abiotic environment as a linear functionthese plastic loci:

T2ij p okp1, 10

N2ijk 1 (Ei 1 z1i) okp1, 10

P2ijk, (2)

here z1i is the mean phenotype of species 1 in the ith deme,d the other quantities are as defined above for species 1. Ifecies 2 was plastic only in response to its biotic environ-ent, the abiotic environmental effect (Ei) was set to 0. Inis case, evolution in species 1 that changes its phenotypeads to corresponding variation in the realized phenotypesindividuals in species 2, responding developmentally toait values of species 1.Selection was based on survival, with the probability ofrviving being a Gaussian function of the difference be-een an individual’s phenotype and both the locally op-al phenotype and the mean phenotype of the otherecies. Fitness (a term that we use as shorthand for onemponent of fitness: the probability of surviving afterrth before dispersal and reproduction) was determined ase product of the effects of the fitness effects of the abioticd biotic environments: Wij p Waij # Wbij, where

Waij p exp

�212

�Tij 2

vi

j

�2�, ð3aÞ

Wbij p11 I exp ½2a(Tij 2 zsi)

2=(11 2azsi)�2

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi11 2azsi

p , (3b)

hereWij is the fitness of the jth individual in the ith deme,j is the phenotype of that individual, vi p 0.1(i 2 25.5) ise abiotic optimal phenotype in that deme, j is the strengthselection from abiotic factors (selection weakens as j in-eases), zsi is the mean phenotype of the other species ine ith deme, and a is the strength of the biotic interaction.e biotic effect is maximized when an individual’s pheno-pe matches the other species’ mean phenotype; I is 21r negative biotic effects (competition and prey species) orfor positive biotic effects (mutualism and predator spe-


. 1 in that article.) For all simulations, we set j p 2; thength of the spatial gradient across all demes was thenproximately 1.25 times the width (2j) of the within-demelection function, that is, weak abiotic selection. For therameters used here, the reaction norm that maximizedness with respect to the abiotic gradient had a slope of 0.1.After selection, surviving individuals could disperse. Dis-rsal occurred in a single bout of movement per genera-n. The dispersal probability and the distance moved weretermined using a zero-mean Gaussian random numberith the truncated magnitude determining the distance,d the sign the direction), so that the probability of movingd the average distancemoved were correlated (see fig. 1 ofheiner and Holt 2012). Increasing the dispersal probabil-(also called the dispersal rate) was done by increasing theriance of the Gaussian, so that at a higher dispersal rate,ore individuals were likely to move, and they were alsoely to move farther. Individuals that would otherwisesperse beyond the end of the gradient moved to the ter-inal demes and stayed there. Dispersal per se had no cost;mely, survival during dispersal was 100%. Unless other-ise noted, dispersal rates were the same for both species.Sexual reproduction was accomplished by assemblingirs of individuals within a deme at random with replace-ent (allowing for self-fertilization), with each parent pro-cing a haploid gamete of 10 unlinked alleles. Each pairen produced one offspring, repeating until the carryingpacity of that deme (100 for all demes) was reached. Thisocedure assumes soft selection, in that local populationze (after reproduction) was determined independently ofe outcome of selection. This implies that individuals infect compete to produce successful offspring. This proto-l also implicitly assumes that no local populations becometinct; with the strengths of selection used here, extinctionsver occurred. The model assumes that the spatial scale ofproduction and mating matches that of density depen-nce and the grain of the selective environment. Maintain-g constant deme size also means that species interactionsd not explicitly depend on density but only on mean localait values. Introducing such density dependence wouldan interesting extension for future work.Each simulation was initialized with 100 individuals ofch species being born in each deme. For each individualthe initial generation, allelic values for both plastic andnplastic loci were chosen independently from the values2, 21, 0, 1, and 2, with each value being equally likely.en though the alleles are integer-valued initially, theirlues could assume any real number in subsequent gen-ations due to mutation. All simulations were run for,000 generations to ensure that equilibrium (the pointter which all calculated quantities showed no further ob-



vious directional trend) was reached. Each parameter com-bination was replicated 20 times, and the results shown arethre

Wtwthtuenentesloxhaag

thdeea(1ticnuthveplby(5firnaaltonothdiwretioouevof

toototphdatheaoof

uals per deme, and Tijn is the phenotype of the jth individ-ual born in the final generation in the ith deme in the nthruticabde


e means of those replicates. Coefficients of variation ofported parameters were generally low (1%–5%).

Response Variables
thth
Eqco

1.2

A

Fiin2.grpltyOnore

e examined one measure of phenotypic plasticity ando measures of coevolution. For phenotypic plasticity,e reaction norm describes the phenotypes that are ac-ally or potentially expressed by a given genotype in allvironments. For a reaction norm that is linear over anvironmental gradient, its plasticity can be well charac-rized by its slope over this gradient. In our model, thepe of the reaction norm for an individual j of species(p1, 2) in deme i is the coefficient oP1ijk in the right-nd products of equation (1) and (2) when normalizedainst the rate of environmental change along the gradient.For these simulations, as the slope of Ei was constant,e final outcome was measured as the average across allmes of the sum of the values of the plasticity alleles forch individual born in the final generation. That is, Pi p=R)onp1, R½(1=N)ojp1,N Pijn�, where Pi is the mean plas-ity of the ith deme over all R runs (20), N p 100 is thember of individuals per deme, and Pijn is the sum ofe values of the plasticity alleles of the jth individual de-loping in the ith deme in the nth run. The overall meanasticity P is the average of Pi across deme, and is givenPp (1=D)oip1,D Pi, where D is the number of demes0). (The order of averaging—over runs within demesst or over demes within runs first—does not affect the fi-l average, because the number of demes was the same forl runs.) This average plasticity was standardized relativethe slope of the local abiotic optima vi so that a reactionrm with a relative plasticity of 1 has the same slope ase vi; a pure differentiation outcome (where all phenotypicfferentiation across space reflects genetic differentiation)ould have a relative plasticity of 0. A reaction norm withlative plasticity equal to 1 is not necessarily an evolu-nary optimum (Lande 2014); in fact, our results includetcomes with relative plasticities above and below 1 anden negative values (slopes in a direction opposite of thatthe vi).Coevolution was assessed by two measures of the extentwhich the phenotypes of the two species matched eachher or were otherwise affected by the presence of theher species. First, we measured the slope of the expressedenotype of each species (slope of the mean Tij), stan-rdized against the slope of the local abiotic optima vi ine absence of the other species. The mean phenotype inch deme for each species was calculated asTi p (1=R)#np1,R½(1=N)ojp1,N Tijn�, whereTi is the mean phenotypethe ith deme over all R runs, N is the number of individ-


n. This standardization is similar to that of relative plas-ity so that a perfect match ofmean phenotypes to the localiotic optima equals 1. Second, we measured the cross-me correlation coefficient of those phenotypes betweene two species, rp correlation(T1iT2i), whereT1i andT2i aree means of species 1 and 2 in the ith deme.

Results

Comparisons among Types of Ecological Interactions

ual Dispersal Rates between Species. We first focus onmparisons among different types of ecological interac-

0 10 20 30 40 50 60 70 80

Rel

ativ

e pl

astic

ity

-1.2

-0.8

-0.4

0.0

0.4

0.8

Dispersal rate (%)0 10 20 30 40 50 60 70 80

Rel

ativ

e pl

astic

ity

-1.2

-0.8

-0.4

0.0

0.4

0.8

1.2

CompetitionPredator plasticPrey plasticMutualism Single species

B

gure 2: Effect of dispersal rate (probability) and type of speciesteraction on selection for phenotypic plasticity (mean Pij) of speciesPlasticity is relative to the optimal reaction norm for the abioticadient. Species 1 was not phenotypically plastic. A, Phenotypicasticity due to the biotic environment only (pattern 1). B, Pheno-pic plasticity due to the biotic and abiotic environments (pattern 2).pen circles show the effect of dispersal rate on selection for phe-typic plasticity when there is only one species and plasticity is insponse to the abiotic environment only.



tions that determine the form of natural selection (bottomhalf of fig. 1), comparing competition, mutualism, andprwphnoabplthpetr

inspotthredisintodipeciopba

oncithwththgranthastywmvotrtr

ofcobepeinanretojuab

Different Dispersal Rates. In the previous comparison,both species dispersed at the same rate. We next comparedthre

n 1.0

Fiph1bibono


edator-prey interactions. We consider these interactionsith regard to two forms of interaction that determine theenotype: trait development depending on just the phe-type of the other species or also jointly depending on theiotic environment.When only one species (species 2) wasastic, the extent to which spatial variation in a trait evolvesrough plasticity rather than genetic differentiation de-nds on the type of species interaction, the determinants ofait development, and the rate of dispersal (fig. 2).For competitive and mutualistic interactions, increas-g dispersal rates resulted in greater trait plasticity ofecies 2 (mean Pij). Comparing plasticity due to the bi-ic environment only (fig. 2A) with plasticity due to bothe biotic and abiotic environments (fig. 2B), the lattersulted in greater plasticity at low dispersal rates. At highspersal rates, the amount of plasticity was similar to thegle-species, abiotic-only response. That plasticity duejust the biotic environment is favored most at highspersal rates implies, for example, that for tadpoles com-ting for food or for plant-bacterial mutualistic asso-ations, plasticity would be favored by decreased phil-atry of the tadpoles or by high dispersal rates of thecteria.For predator-prey interactions, the effect of dispersal ratethe amount of plasticity was greatest when the prey spe-

es was plastic, and least when the predator was plastic. Fore prey species, at low dispersal rates, its relative plasticityas negative, indicating that the reaction norm sloped ine opposite direction from what would be expected frome optimum of the abiotic gradient. Because the abioticadient selects for the same phenotype in the predatord the prey species, at low dispersal rates, the fitness ofe prey species is maximized whenever its phenotype isdifferent from the predator as possible. As with the otherpes of interactions, plasticity was greater in magnitudehen it was due to both the abiotic and the biotic environ-ents. These results imply, for example, that for traits in-lving antagonistic plant-herbivore interactions, the plantaits are more likely to be plastic than are the herbivoreaits.In contrast to the effect of coevolution on the evolutiontrait plasticity, trait plasticity had only modest effects onevolution, as measured by the among-deme correlationstween the two species (fig. 3). As expected, at low dis-rsal rates, the correlation was near unity for mutualisticteractions, zero to negative for competitive interactions,d intermediate for predator-prey interactions. All cor-lations tended to increase with dispersal rate and tendedbe similar whether trait plasticity was in response tost the other species or to both the other species and theiotic environment, or there was no plasticity.


e effect of different dispersal rates between the speciessponding in a plastic fashion to the other species but

1.2

0 10 20 30 40 50 60 70 80

Amon

g de

me

corr

elat

io

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

Dispersal rate (%)

0 10 20 30 40 50 60 70 80

Amon

g de

me

corr

elat

ion

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

CompetitionPredator plasticPrey plasticMutualism

0 10 20 30 40 50 60 70 80

Amon

g de

me

corr

elat

ion

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

A

B

C

gure 3: Effect of dispersal rate and type of species interaction on theenotypic correlation between the two species among demes. Specieswas not phenotypically plastic. A, Phenotypic plasticity due to theotic environment only (pattern 1). B, Phenotypic plasticity due toth the biotic and the abiotic environment (pattern 2). C, No phe-typic plasticity.



not the abiotic environment (species 2 in fig. 1, which re-sponded plastically only to species 1) and a nonplasticsparilyamlivva

pr

plastic species had no effect on the evolution of trait plas-ticity in the other species (fig. 4B, 4C). By contrast, when theinticnothticwfopr

Fi2.pl


ecies (species 1 in fig. 1). Such asymmetric dispersal ratese quite likely; for example, plants may disperse less read-than do their herbivores or bacterial mutualists, whileong competing marine invertebrates, some have long-ed, highly dispersing pelagic larvae and others have lar-e that settle quickly.In our model, if the interaction was mutualistic or theedator species was plastic, the dispersal rate of the non-


teraction was competitive or the prey species was plas-, trait plasticity increased with the dispersal rate of thenplastic species (fig. 4A, 4D). That is, the dispersal rate ofe nonplastic species mattered for the evolution of plas-ity only when the abiotic selective optimum conflictedith the biotic selective optimum. This effect was greatestr low dispersal rates of the plastic species, especially foredator-prey interactions. In all cases, plasticity tended to

0.0

0.2

0.4

0.6

0.8

1.0

0

2040

6080

020

4060

Rel

ativ

e pl

astic

ity

Nonpla

stic s

pecie

s

Plastic species

0.0

0.2

0.4

0.6

0.8

1.0

0

2040

6080

020

4060

Rel

ativ

e pl

astic

ity

Nonpla

stic s

pecie

s

Plastic species

-0.8

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

0

2040

6080

020

4060

Rel

ativ

e pl

astic

ity

Nonpla

stic p

reda

tor

Plastic prey

0.0

0.2

0.4

0.6

0.8

1.0

0

2040

6080

020

4060

Rel

ativ

e pl

astic

ity

Nonpla

stic p

rey

Plastic predator

A B

C D

gure 4: Effect of dispersal rates of the interacting species (axes on bottom plane) on selection for phenotypic plasticity (mean Pij) of speciesPlasticity is relative to the optimal reaction norm for the abiotic gradient. Species 1 was not phenotypically plastic, and phenotypicasticity was due to the biotic environment only (pattern 1). A, Competition. B, Mutualism. C, Predator plastic. D, Prey plastic.



increase with increasing dispersal rate of the plastic species,although this effect was weak when the predator was plastic.Thpllik

C

Acotefoplanvian(4

species 2 responding to both its biotic and abiotic environ-ment. The first two patterns are those explored above; inththtiothm

Firethre


ese results imply, for example, that for traits involvingant-herbivore interactions, the traits of the plant are moreely to be plastic than are those of the herbivore.

omparisons among Types of Phenotypic Determinations
nibathofpeprthhemco
biotic and Biotic Selection in Agreement. Next we focus onmparisons among different types of interactions that de-rmine the phenotype (top half of fig. 1). We compared allur possible patterns: (1) species 1 nonplastic and species 2astic to just its biotic environment; (2) species 1 nonplasticd species 2 responding to both its biotic and abiotic en-ronment; (3) species 1 plastic to the abiotic environmentd species 2 responding to just its biotic environment; and) species 1 responding to the abiotic environment and

1.0

A


ese comparisons, species 1 can now also be plastic. Forese comparisons, we focused on predator-prey interac-ns, since they were found to be the most disparate ine previous comparisons. Examples of prey traits thatight depend on both biotic and abiotic conditions includetrogen-based chemical defenses of plants and calcium-sed shell hardness of mollusks. In both cases, selection one prey could be influenced by environmental availabilitythese limiting resources, and plasticity might jointly de-nd on both the abiotic and the biotic environment. Ifedator-prey interactions depend on the timing of events,en both speciesmight use cues—such as day length or totalat days—to determine life-history events. Again, fitnessight be influenced by both species’ interactions and abioticnditions.

1.0 B

0 10 20 30 40 50 60 70 80

Rel

ativ

e pl

astic

ity

0.0

0.2

0.4

0.6

0.8

Sp 2 bioticSp 2 biotic + abioticSp 1, Sp 2 bioticSp 1, Sp 2 biotic + abiotic

0 10 20 30 40 50 60 70 80-1.5

-1.0

-0.5

0.0

0.5

Dispersal rate (%)0 10 20 30 40 50 60 70 80

-2.5

-2.0

-1.5

-1.0

-0.5

0.0

0.5

1.0

1.5

Dispersal rate (%)0 10 20 30 40 50 60 70 80

Rel

ativ

e pl

astic

ity

0.0

0.2

0.4

0.6

0.8

1.0

1.2

Single speciesC D

gure 5: Effect of dispersal rate and pattern of phenotypic determination on trait plasticity (mean Pij). Plasticity is relative to the optimalaction norm for the abiotic gradient. A, B, Species 2 p predator. C, D, Species 2 p prey. A, C, Plasticity of the predator. B, D, Plasticity ofe prey. Open circles show the effect of dispersal rate on selection for phenotypic plasticity when there is only one species and plasticity is insponse to the abiotic environment only.



When the predator was species 2 and responded in aplastic manner to the prey species (with or without abioticplpebetyitsaspnedirothbociaevon

When the prey was species 2 and responded in a plasticmanner to traits of the predator species, a plastic preda-tocolodiTwthmwthto

lethexat

FidiOp


asticity), the evolution of predator plasticity was inde-ndent of the plasticity of the prey species (fig. 5A). Asfore, the predator tended to evolve low levels of pheno-pic plasticity, especially when it was responding to justbiotic environment; moreover, dispersal rate had onlymodest effect on plasticity. In these scenarios, the preyecies (when plastic) responded directly in a plastic man-r only to the abiotic environment. At low to intermediatespersal rates, its plasticity depended on the type of envi-nmental response of the predator species (fig. 5B). Whene predator species was more plastic (i.e., in response toth the biotic and the abiotic environment), the prey spe-es had a steeply negative reaction norm. Thus, plasticity ingiven species can indirectly reflect plasticity in a co-olving species, even if there is no direct effect of the latterthe plastic response of the other.

1.2 A


r increased the plasticity of the prey species (fig. 5D), inntrast to the previous scenario.When dispersal rates werew, the prey species had a very steep reaction norm in therection opposite of that favored by the abiotic gradient.hus, coevolution combined with plasticity can lead tohat looks like countergradient selection. The plasticity ofe predator, which depended on just the abiotic environ-ent, was very similar to that of a single species communityith no coevolution (fig. 5C). Again, the type of plasticity ofe prey species (biotic only or biotic and abiotic) had littleno effect of the plasticity of the predator species.These asymmetric differences in plasticity by trophicvel resulted in differences in the expressed phenotypes ofe predator and prey species. In figure 6, the slope of thepressed phenotype (slope of the mean Tij) is shown rel-ive to the slope that would match the locally optimal

2.0

B

0 10 20 30 40 50 60 70 80

Rel

ativ

e ph

enot

ype

0.8

0.9

1.0

1.1

0 10 20 30 40 50 60 70 80

0.0

0.5

1.0

1.5

Dispersal rate (%)0 10 20 30 40 50 60 70 80

0.6

0.8

1.0

1.2

1.4

1.6

1.8

Dispersal rate (%)0 10 20 30 40 50 60 70 80

Rel

ativ

e ph

enot

ype

0.80

0.85

0.90

0.95

1.00

1.05

Sp 2 bioticSp 2 biotic + abiotic

Sp 1 abiotic, Sp 2 bioticSp 1 abiotic, Sp 2 biotic + abioticNo plasticityC D

Sp 1, Sp 2 biotic

gure 6: Effect of dispersal rate and pattern of phenotypic determination on the slope of the expressed phenotype (slope of the mean Tij

vided by slope of vi). A, B, Species 2 p predator. C, D, Species 2 p prey. A, C, Phenotype of the predator. B, D, Phenotype of the prey.en circles show the effect of dispersal rate on the expressed phenotype when there is no trait plasticity.



abiotic phenotype in all demes (the value 1 means that thetwo slopes are the same). When the predator’s phenotypedetyplthvedianspnofophthan

prhaatdiilaspatprThtoanpe

cophAthmhoplofbe

Aintioreefrete

aofprbetyto

abresponding to just its biotic environment, or at low dis-peHininlowvi

thhacoraplesthw

n

1.0

A

Fimamprra


pended developmentally on that of the prey, its pheno-pes closely tracked the abiotic optima, similar to whenasticity was absent (fig. 6A). The slope was shalloweran for the local abiotic optimum at low dispersal rates andry close to or steeper than the abiotic optimum at highspersal rates. The phenotype of the prey species showedeffect of dispersal rate similar to that of the predatorecies, although the effect magnitude was greater (fig. 6B;te difference in scale with fig. 6A). These results imply,r example, that sedentary prey species are likely to showenotypes that appear to be nonadaptive with respect toe abiotic environment when predator traits are plasticd the prey are coevolving with the predator.When the prey’s phenotype depended on that of theedator (i.e., the prey is species 2 in fig. 1), its phenotyped a slope that again was shallower than the optimumlow dispersal rates and steeper than the optimum at highspersal rates if the predator was not plastic (fig. 6D; sim-r to the previous pattern). The phenotype of the preyecies was closest to the abiotic optimum when the pred-or was also phenotypically plastic. In contrast, for theedator, there was little effect of dispersal rate (fig. 6C).e slope was very close to that of the optimum, similarwhen plasticity was absent. There thus appears to beemergent asymmetry by trophic level of the effect of dis-rsal on plasticity.These patterns of phenotypic expression affected theevolutionary pattern as measured by the among-demeenotypic correlation between the two species (fig. 7).gain, as expected, the correlation was close to 1.0 whene dispersal rate was high and generally declined tooderate values as the dispersal rate decreased. In general,wever, this pattern did not depend on the pattern ofasticity. These results imply that in general, assessmentscoevolution based on among-deme correlations will notaffected by trait plasticity.

biotic and Biotic Selection Opposite. For predator-preyteractions, we also examined the case where augmenta-n and conflict between abiotic and biotic selection wereversed. For these simulations, both the environmentalfector (Ei) and selection (vi) were reversed for one specieslative to the other species, leading to complex and in-restingly different patterns.In this instance, when the predator species responded inplastic manner to the prey species (fig. 8A), the evolutionits plasticity was independent of the plasticity of theey species, as with the previous scenario (fig. 5A). Asfore, the predator tended to have low levels of pheno-pic plasticity, although now the reaction norms tendedbe in the opposite direction of that favored by the


rsal when also responding to the abiotic environment.owever, the prey species responded very differently thanthe previous scenario (compare fig. 8B and fig. 5B), withtermediate levels of plasticity at low dispersal rates andw levels of plasticity at high dispersal rates, especiallyhen the predator was responding to just the biotic en-ronment.When the prey species responded in a plastic manner toe predator species (fig. 8D), the plasticity of the predatord little effect on the plasticity of the prey species, inntrast to the previous scenario (fig. 5D). When dispersaltes were low, the prey species had an intermediate level ofasticity that usually declined as dispersal rates increased,pecially when the prey species was responding to juste biotic environment. The exception to these patterns washen the prey species’ phenotype was determined by a

iotic environment, especially when the predator was

0 10 20 30 40 50 60 70 80

Am

ong-

dem

e co

rrel

atio

0.0

0.2

0.4

0.6

0.8

Dispersal rate (%)0 10 20 30 40 50 60 70 80

Am

ong-

dem

e co

rrel

atio

n

0.0

0.2

0.4

0.6

0.8

1.0

Sp 2 bioticSp 2 biotic + abioticSp 1 abiotic, Sp 2 bioticSp 1 abiotic, Sp2 biotic + abioticNo plasticity

B

gure 7: Effect of dispersal rate and pattern of phenotypic deter-ination on the phenotypic correlation between the two speciesong demes when they interact as predator and prey.A, Species 2pedator. B, Species 2p prey. Open circles show the effect of dispersalte on the among-deme correlation when there is no trait plasticity.



nocase, the prey species’ plasticity increased as the dispersalrawthanpr

spthap1.eirecrcospw

lection. However, assessment of phenotypes relative to justthvanoofgr

Ophofgrcoprmbi

1.0

Sp 2 biotic 1.2

1.4

A B

FiopA,wh


te increased. The plasticity of the predator (fig. 8C), whichas a function of just the abiotic environment, was loweran that of a single species in the absence of coevolution,d there was little to no effect of the type of plasticity of theey species.The expressed phenotypes of both the predator and preyecies were quite different from the previous scenario. Fore predator, the slope of the expressed phenotype wasproximately 0.7, while that of the prey was approximately6. These values did not depend on the type of plasticity ofther species or the dispersal rate. The among-deme cor-lation ranged from approximately 20.75 to 20.85, in-easing in magnitude with the dispersal rate. Again, therrelation did not depend on the type of plasticity of eitherecies. All of these values were identical to those that arosehen plasticity was absent. Thus, assessments of coevolu-


e abiotic environment will appear to imply that spatialriation in traits are nonadaptive. This suggests that ig-ring coevolution can lead to misleading interpretationshow adaptation is playing out across environmentaladients.

Discussion

verall, coevolution had a larger effect on the evolution ofenotypic plasticity than plasticity had on the outcomecoevolution. Effects on the evolution of plasticity wereeater when the fitness-maximizing coevolutionary out-meswere antagonistic between the species pair (predator-ey interactions) than when those outcomes were aug-enting (competitive or mutualistic interactions). Overall,otic interactions reduced selection for plasticity. If this

nplastic predator and the abiotic environment; in this tion are not affected by discordant biotic and abiotic se-

0 10 20 30 40 50 60 70 80

Rel

ativ

e pl

astic

ity

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8 Sp 2 biotic + abioticSp 1, Sp 2 bioticSp 1, Sp 2 biotic + abiotic

0 10 20 30 40 50 60 70 800.0

0.2

0.4

0.6

0.8

1.0

Dispersal rate (%)0 10 20 30 40 50 60 70 80

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

Dispersal rate (%)0 10 20 30 40 50 60 70 80

Rel

ativ

e pl

astic

ity

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

C D

Single species

gure 8: Effect of dispersal rate and pattern of phenotypic determination on trait plasticity (mean Pij) when the abiotic optima haveposite slopes. Plasticity is relative to the optimal reaction norm for the abiotic gradient. A, B, Species 2p predator. C, D, Species 2p prey.C, Plasticity of the predator. B, D, Plasticity of the prey. Open circles show the effect of dispersal rate on selection for phenotypic plasticityen there is only one species and plasticity is in response to the abiotic environment only.



result holds up in future studies, it may explain why plas-ticity is less ubiquitous than one might expect. Even whentrrobyexby

lufavetoFotyminthmwphbereprinbycoprhothadphco

ThetreandelooptoseinteacWhete

coco

correlations in our simulations (figs. 3, 7; see fig. 5 ofNuismer et al. 2010). The detailed differences with thatmmdeTfoteaFowFowtorethofofNcoofwen

spdinospormwthcow


ait development was responding to just the abiotic envi-nment, selection on plasticity, however, could be affectedbiotic interactions (figs. 5, 8). In contrast, coevolution aspressed by the among-deme correlation was little affectedwhether traits were plastic (figs. 3, 7).Our most striking result may be the asymmetry in evo-tionary responses of predators and prey (fig. 5), despite thect that the fitness functions were identical except for a re-rsal of the sign of the interaction. This difference is duean asymmetry in how fitness changes with phenotype.r the predator, fitness is maximized when its pheno-pe jointly matches the optimum for the abiotic environ-ent and the phenotype of its prey. Selection is stabiliz-g for both optima, although those optima may differ. Fore prey, again fitness is maximized when its phenotypeatches the optimum for the abiotic environment but alsohen that phenotype is as different as possible from theenotype of the predator. The latter selection is directionalcause fitness keeps increasing as the difference grows. Thesult is strong countergradient selection on plasticity of theey species, that is, selection for a reaction norm that slopesthe opposite direction expected from the optima favoredthe abiotic environment (fig. 5C, 5D). Because of thisuntergradient selection, if the plasticity of awild-collectedey species were measured in a growth chamber, green-use, or transplant garden in the absence of the predator,e resulting reaction norm might be interpreted as mal-aptive. We reinforce the message of Scheiner (2013) thatenotypic plasticity must always be interpreted in the fullntext of a species’ ecology and selective milieu.

Related Models
ovoudyteetseofsptephNplfuplplwHhate
e model in this article derives from those of Nuismeral. (2010) and Scheiner and Holt (2012) but differs in keyspects from both. The obvious difference with Scheinerd Holt (2012) is the addition of a biotic interaction fortermining plasticity. In addition, in this article, the eco-gical gradient had smaller differences in the phenotypictima between adjacent demes, weakening selection duethe abiotic environment. We made this change so thatlection by biotic and abiotic factors would be comparablemagnitude. Despite these differences, the general pat-rns of plasticity evolution in the absence of biotic inter-tions (fig. 2B) were similar to those previously reported.e conclude that the effects on plasticity evolution seenre are specifically the result of our inclusion of biotic in-ractions.Nuismer et al. (2010) focused on the question of whetherevolution would produce a signal of an among-demerrelation across species. We found similar among-deme


odel are twofold. First, in our model, fitness was deter-ined by themean phenotype of the other species in a givenme rather than by pairwise interactions of individuals.his difference alters which empirical examples are relevantr our results but does not seem to affect the overall pat-rn of coevolution. Second, in our model, dispersal was byGaussian dispersal kernel rather than by an island pattern.r such kernels, movement occurs among nearby demes,ith the probability of movement decreasing with distance.r island migration, movement occurs among all demesith equal probability, a pattern of movement that will tenderode among-deme correlations. The similarity of oursults with those of Nuismer et al. (2010) is likely becausee coevolutionary dynamic is driven by the phenotypeseach species rather than by the specific mechanistic basesthose phenotypes, that is, plasticity. We conclude, as diduismer et al. (2010), that the existence of an among-demerrelation is on its own weak evidence for the presencea coevolutionary interaction. Our results suggest that

hether such correlations occur may not be strongly influ-ced by the presence of plasticity in trait development.Our simulations considered only the simplest case ofatial variation in the abiotic environment—a linear gra-ent—coupled with a Gaussian dispersal kernel. We didt consider the effects of temporal variation combined withatial variation, alternative patterns of dispersal (e.g., islandnon-Gaussian), or different life-history patterns (move-ent before selection). On the basis of Scheiner (2013),hich considered both spatial and temporal variation fore single-species case, the resulting evolutionary outcomesuld be quite complex. Although our equilibrium resultsere similar to those of Nuismer et al. (2010), we more-er did not examine the dynamics of coevolution withr model. On the basis of Nuismer et al. (1999, 2000), thatnamic can also be quite complex, especially if fitness in-ractions vary both spatially and temporally (Gomulkiewiczal. 2000, 2003). This is likely if, for instance, instead of softlection there is hard selection, with the population sizeeach species responding to the interaction with the otherecies. Exploration of these additional factors as well asmporal variation and alternative fitness functions (e.g., aenotypic differences mechanism of species interactions;uismer et al. 2010) may reveal additional insights aboutasticity evolution in a coevolutionary context and permitrther scrutiny of our conclusion that plasticity may notay that large a role in coevolution. In this article, we ex-ored only the effects of soft selection; local population sizeas determined independently of the outcome of selection.ard selection—population size is density dependent—canve important effects on the transient dynamics of sys-ms (e.g., fig. 6 of Gomulkiewicz and Kirkpatrick 1992), es-



pecially on the propensity for extinction. However, it hasweaker effects on long-term evolutionary equilibria, unlessthKheas

Tecucaovorercogaredisptyvospprcamthhobinoyscoticabgljebecointrex

ticresefegrofprsism

th

cific interactions and plasticity and are not definitive. Onechallenge in developing tests of our theoretical predictionsisofcoNinan


ere are genetic trade-offs (e.g., fig. 5 of Gomulkiewicz andirkpatrick 1992). We expect that the general patterns foundre would be unchanged with hard selection; proving thatsertion awaits future explorations.

Testing Our Model
onfosp[3sudi
BemtothOsugegeacevsispsuamotifFoasabepotioplcoshouthar

Wmofpa05

sting models of plasticity evolution empirically is diffi-lt because measuring plasticity involves raising geneti-lly related individuals in multiple environments; more-er, comparisons often require finding metapopulationsspecies with different patterns of environmental het-

ogeneity. One way around the second difficulty is tompare the plasticities of different traits of a single or-nism. Our model predicts overall greater plasticity insponse to abiotic cues than to biotic cues. It also predictsfferent amounts of plasticity, depending on the type ofecies interaction. Nearly all species experience a variety ofpes of interspecific interactions. For example, an herbi-re or predator can also be a prey species, or competingecies can be prey for another species (as in keystoneedation), or a mutualist can also be a predator. In the lastse, for instance, we predict that traits that enhance theutualistic interaction should be more plastic than traitsat enhance the predatory interaction. This predictionlds whether the plasticity is in response to either theotic or the abiotic environment, as long as the traits aret genetically or developmentally correlated. A full anal-is of this prediction requires a model that explicitly in-rporates such correlations. A comparison of trait plas-ities within a single species based on responses to aniotic environmental signal has the virtue of using a sin-e environmental effector and traits that have been sub-ct to a single evolutionary history. Comparisons can alsomade using different biotic effectors, although it be-mes necessary to standardize environmental differencessome fashion. As far as we know, no one has comparedait plasticities in this way; such data, however, may wellist.Murren et al. (2014) examined the extent to which plas-ity of a given trait varied among populations or closelylated species, which provides a possible indication oflection on plasticity. Among types of environmental ef-ctors, they found that biotic factors tended to result ineater plasticity differences than did abiotic factors. Mostthose measures of biotic interactions were of traits ofey species (C. Murren, personal communication), con-tent with our predictions that prey species will vary theost in their amount of plasticity (fig. 2).We caution that our results provide only preliminaryeoretical expectations about the interplay of interspe-


that in natural communities, pairwise interactions areten embedded in richer communities, so a multispeciesevolutionary perspective may be needed (Ridenhour anduismer 2012). A given trait may influence how a speciesteracts with not just one but a whole suite of other species,d developmental cues might come from more than juste other species as well. Moreover, different functionalrms for expressing the relationships among traits, inter-ecific interactions, and fitness than assumed above (eq.b]) might lead to different insights, as would our as-mption about soft selection. These all would be importantrections for future theoretical studies.

Conclusions

oth coevolution and plasticity evolution have extensivepirical and theoretical literatures. Despite this rich his-ry, there been few attempts to directly test any of thoseeories and to assess how the two arenas affect each other.ur results provide encouragement for those attemptingch tests. For tests of coevolution theory, our results sug-st that researchers need not be unduly worried about thenetic bases of the traits involved in the species’ inter-tion. Whether those traits are plastic or fixed, a similarolutionary outcome is predicted. In addition, we foundmilar results to those of Nuismer and collaborators, de-ite major differences in model assumptions and structure,ggesting that these theoretical predictions are robust. Ex-ining the evolution of trait plasticity may provide an-her—albeit indirect—indicator of coevolution, especiallysuch traits can be examined for both interacting species.r tests of plasticity evolution theory, our results point tomode of testing by comparing different traits within theme individual. This across-trait mode of analysis maymore approachable than are comparisons among meta-pulations or species. However, our results also lend cau-n to interpretations of plasticity patterns when thoseasticity measurements are done outside the ecologicalntext of selection. For both sets of theories, our resultsow how the merging of two theory domains can deepenr understanding of both. Empirical research is neededat will engage with plasticity evolution in a coevolution-y context.

Acknowledgments

e thank M. Barfield for advice and comments on theodel and the manuscript. R.D.H. thanks the UniversityFlorida Foundation for support. R.G. was supported inrt by National Science Foundation (NSF) grants DMS40524 and DEB 1354264. This manuscript is based on



work done by S.M.S. while serving at the NSF. The viewsexpressed in this article do not necessarily reflect those ofth

Ap

As

Ba

De

Du

Ga

Go

Go

Go

Go

Ki

La

M

N

Nuismer, S. L., J. N. Thompson, and R. Gomulkiewicz. 1999. Geneflow and geographically structured coevolution. Proceedings of

—

O


e NSF or the U.S. Government.

Literature Cited

Pf

Po

Ri

Ro

Sc

Sc

Sc

Si

Sm

Th

Th

W

Editor: Susan Kalisz

pleton, R. D., and A. R. Palmer. 1988. Water-borne stimuli re-leased by predatory crabs and damaged prey inducemore predator-resistant shells in a marine gastropod. Proceedings of the NationalAcademy of Sciences of the USA 85:4387–4391.chehoug, E. T., and R. M. Callaway. 2012. Plasticity and trait-mediated indirect interactions among plants. Pages 489–507 inT. Ohgushi, O. Schmitz, and R. D. Holt, eds. Trait-mediated in-direct interactions: ecological and evolutionary perspectives. Cam-bridge University Press, Cambridge.rton, N. H. 1999. Clines in polygenic traits. Genetical Research74:223–236.Witt, T. J., and S. M. Scheiner. 2004. Phenotypic plasticity: func-tional and conceptual approaches. Oxford University Press, NewYork.dley, S. A., and J. Schmitt. 1996. Testing the adaptive plasticity hy-pothesis: density-dependent selection on manipulated stem lengthin Impatiens capensis. American Naturalist 147:445–465.ndon, S., and S. L. Nuismer. 2009. Interactions between geneticdrift, gene flow, and selection mosaics drive parasite local adap-tation. American Naturalist 173:212–224.mulkiewicz, R., R. D. Holt, and M. Barfield. 1999. The effects ofdensity dependence and immigration on local adaptation and nicheevolution in a black-hole sink environment. Theoretical PopulationBiology 55:283–296.mulkiewicz, R., and M. Kirkpatrick. 1992. Quantitative geneticsand the evolution of reaction norms. Evolution 46:390–411.mulkiewicz, R., S. L. Nuismer, and J. N. Thompson. 2003. Co-evolution in variable mutualisms. American Naturalist 162(suppl.):S80–S93.mulkiewicz, R., J. N. Thompson, R. D. Holt, S. L. Nuismer, andM. E. Hochberg. 2000. Hot spots, cold spots, and the geographicmosaic theory of coevolution. American Naturalist 156:156–174.mura, M. 1965. A stochastic model concerning the maintenance ofgenetic variability in quantitative characters. Proceedings of theNational Academy of Sciences of the USA 54:731–736.nde, R. 2014. Evolution of phenotypic plasticity and environ-mental tolerance of a labile quantitative character in a fluctuatingenvironment. Journal of Evolutionary Biology 27:866–875.urren, C. J., H. J. Maclean, S. E. Diamond, U. K. Steiner, M. A.Heskel, C. A. Handelsman, K. G. Cameron, et al. 2014. Evolutionarychange in continuous reaction norms. American Naturalist 183:453–467.uismer, S. L., R. Gomulkiewicz, and B. J. Ridenhour. 2010. When iscorrelation coevolution? American Naturalist 175:525–537.


the Royal Society B: Biological Sciences 266:605–609.——. 2000. Coevolutionary clines across selection mosaics. Evo-lution 54:1102–1115.hgushi, T. 2012. Community consequences of phenotypic plasticityof terrestrial plants: herbivore-initiated bottom-up trophic cas-cades. Pages 161–185 in T. Ohgushi, O. Schmitz, and R. D. Holt,eds. Trait-mediated indirect interactions: ecological and evolu-tionary perspectives. Cambridge University Press, Cambridge.ennig, D. W. 1992. Polyphenism in spadefoot toad tadpoles as alocally adjusted evolutionarily stable strategy. Evolution 46:1408–1420.lechová, J., N. Barton, and G. Marion. 2009. Species’ range: ad-aptation in space and time. American Naturalist 174:E186–E204.denhour, B. J., and S. L. Nuismer. 2012. Perspective: trait-mediatedindirect interactions and the coevolutionary process. Pages 207–220 in T. Ohgushi, O. Schmitz, and R. D. Holt, eds. Trait-mediatedindirect interactions: ecological and evolutionary perspectives.Cambridge University Press, Cambridge.binson, B. W., and D. W. Pfennig. 2013. Inducible competitorsand adaptive diversification. Current Zoology 59:537–552.heiner, S. M. 2013. The genetics of phenotypic plasticity. XII. Tem-poral and spatial heterogeneity. Ecology and Evolution 3:4596–4609.heiner, S. M., and R. D. Holt. 2012. The genetics of phenotypicplasticity. X. Variation versus uncertainty. Ecology and Evolution2:751–767.hoener, T. W., and D. A. Spiller. 2012. Perspective: kinds of trait-mediated indirect effects in ecological communities. a synthesis.Pages 9–27 in T. Ohgushi, O. Schmitz, and R. D. Holt, eds. Trait-mediated indirect interactions: ecological and evolutionary per-spectives. Cambridge University Press, Cambridge.mms, E. L., D. L. Taylor, J. Povich, R. P. Shefferson, J. L. Sachs, M.Urbina, and Y. Tausczik. 2006. An empirical test of partner choicemechanisms in a wild legume-rhizobium interaction. Proceedingsof the Royal Society B: Biological Sciences 273:77–81.ith, L. D., and A. R. Palmer. 1994. Effects of manipulated diet onsize and performance of brachyuran crab claws. Science 264:710–712.ompson, D. B. 1992. Consumption rates and the evolution of diet-induced plasticity in the head morphology of Melanopus femur-rubrum (Orthoptera: Acrididae). Oecologia 89:204–213.ompson, J. N. 2005. The geographic mosaic of coevolution. Uni-versity of Chicago Press, Chicago.alker, B., A. Kinzig, and J. Langridge. 1999. Plant attribute diversity,resilience, and ecosystem function: the nature and significance ofdominant and minor species. Ecosystems 2:95–113.

Associate Editor: Christina M. Caruso



the genetics of phenotypic plasticity. xiv. coevolution › rdholt › files › 297.pdfof natural...

Documents