INTERSPECIFIC COMPETITION ALTERS NONLINEAR SELECTION … · communities, interspecific competition...

ORIGINAL ARTICLE

doi:10.1111/j.1558-5646.2012.01749.x

INTERSPECIFIC COMPETITION ALTERSNONLINEAR SELECTION ON OFFSPRING SIZEIN THE FIELDDustin J. Marshall1,2,3 and Keyne Monro1,2

1School of Biological Sciences, The University of Queensland, Brisbane 4072, Australia2School of Biological Sciences, Monash University, Clayton 3800, Australia

3E-mail: [email protected]

Received August 18, 2011

Accepted July 16, 2012

Data Archived: Dryad doi:10.5061/dryad.d6v18

Offspring size is one of the most important life-history traits with consequences for both the ecology and evolution of most

organisms. Surprisingly, formal estimates of selection on offspring size are rare, and the degree to which selection (particularly

nonlinear selection) varies among environments remains poorly explored. We estimate linear and nonlinear selection on offspring

size, module size, and senescence rate for a sessile marine invertebrate in the field under three different intensities of interspecific

competition. The intensity of competition strongly modified the strength and form of selection acting on offspring size. We found

evidence for differences in nonlinear selection across the three environments. Our results suggest that the fitness returns of a

given offspring size depend simultaneously on their environmental context, and on the context of other offspring traits. Offspring

size effects can be more pervasive with regards to their influence on the fitness returns of other traits than previously recognized,

and we suggest that the evolution of offspring size cannot be understood in isolation from other traits. Overall, variability in the

form and strength of selection on offspring size in nature may reduce the efficacy of selection on offspring size and maintain

variation in this trait.

KEY WORDS: Egg size, life-history evolution, maternal effect.

Offspring size is a life-history trait of fundamental importance thataffects the ecology and evolution of many organisms. Estimatingthe relationship between offspring size and fitness has been amajor goal of evolutionary ecology, and over the past 70 years,we have accumulated a wealth of estimates of the offspring size–fitness relationship (Lack 1947; Bagenal 1969; Stanton 1984;Williams 1994; Bernardo 1996; Marshall and Keough 2008a).Despite this intense research effort, our understanding of the se-lective forces acting on offspring size remains surprisingly incom-plete. For example, while early theory predicted a single, optimaloffspring size for any one species (Smith and Fretwell 1974), innature, offspring size typically varies among populations, moth-ers, broods, and offspring (Fox and Czesak 2000; Marshall andKeough 2008b; Krist 2011). The maintenance of offspring size

variation at all levels of biological organization is thought to bedue to environmentally driven variation in selection on offspringsize (Einum and Fleming 2002), but this variation in selection(formally defined) has received relatively little attention.

The effect of offspring size on offspring performance appearsto be highly context dependent. A range of abiotic factors (e.g.,temperature) and biotic factors (e.g., competition, predation) canalter the fitness benefits of a particular offspring size (Bernardo1996; Marshall and Keough 2008a). Yet, despite the increasingnumber of studies that examine the influence of the local environ-ment on offspring size effects, there are very few formal estimatesof selection on offspring size (Einum and Fleming 2000; Dibattistaet al. 2007; Johnson et al. 2010), and even fewer that estimate howselection changes across environments. Meanwhile, much of the

3 2 8C! 2012 The Author(s). Evolution C! 2012 The Society for the Study of Evolution.Evolution 67-2: 328–337

NONLINEAR SELECTION ON OFFSPRING SIZE

observed variation in offspring size is assumed to be adaptive.For example, population-level variation in offspring size is of-ten described as an adaptive response to large-scale environmen-tal variation (Johnston and Leggett 2002). Variation in offspringsize among clutches is viewed as an adaptive response driven bymothers encountering different environments (Fox et al. 1997;Allen et al. 2008), and variation in offspring size within clutchesis assumed to be an adaptive strategy to cope with environ-mental undertainty (i.e., bet-hedging Einum and Fleming 2002;Crean and Marshall 2009). For patterns in offspring size varia-tion at any scale to reflect adaptation, variation in selection mustexist.

Importantly, environmental variation cannot only causechanges in directional (linear) selection in offspring size, it canalso cause changes in nonlinear selection. The impact of environ-mental variation on nonlinear selection on offspring size remainscompletely unknown. Indeed, while the fitness returns of differ-ent offspring sizes are often modeled theoretically as a nonlinearfunction (Smith and Fretwell 1974), formal estimates of nonlin-ear selection are rare (Einum and Fleming 2000; Dibattista et al.2007). Nonlinear selection on offspring size is important becauseit includes quadratic selection and correlational selection, both ofwhich are likely to be of key importance for an understanding ofoffspring size evolution.

Quadratic selection occurs when either extreme trait val-ues are favored (known as disruptive selection) or intermediatetrait values are favored (known as stabilizing selection). In hisinfluential review, Bernardo (1996) predicted that selection onoffspring size was likely to be stabilizing, but lamented that testswere lacking. Recent theoretical considerations of maternal bet-hedging strategies have re-emphasized the importance of esti-mating quadratic selection on offspring size (Crean and Marshall2009). Under some circumstances, diversified bet-hedging withregards to offspring size is predicted to be favored, but only whenoffspring size is subject to stabilizing selection, otherwise, con-servative bet-hedging strategies should be favored (Einum andFleming 2004; Marshall et al. 2008). Those studies that havelooked for evidence of stabilizing selection on offspring size havefound it (Einum and Fleming 2000; Dibattista et al. 2007), butfew studies have examined how environmental variation affectsquadratic selection on offspring size.

Correlational selection occurs when selection acts on the co-variance between traits in combination rather than on individualtraits alone (Blows and Brooks 2003). Estimating correlational se-lection on offspring size and other traits is particularly importantbecause the benefits of offspring size are unlikely to be inde-pendent of the context of other traits. In other words, the fitnessconsequences of increased offspring size are likely to depend onother traits that the offspring expresses. Ignoring the potentialfor correlational selection on combinations of offspring size and

other offspring traits could lead to erroneous conclusions aboutthe benefits of increased offspring size, either overestimating orunderestimating these benefits (depending on the distribution ofvalues of other traits in the population).

Overall, there is a clear need to examine whether selection(both linear and nonlinear) on offspring size varies among en-vironments. Here, we address this important knowledge gap byapplying a well-known multiple regression approach to estimateboth directional and nonlinear selection on offspring size, and twoother traits: a morphological trait (zooid size) and a life-historytrait (rate of colony senescence) in a colonial marine inverte-brate (see Materials and Methods for a description of these traits).These two traits were selected because they are important deter-minants of performance in colonial organisms. In marine benthiccommunities, interspecific competition can be intense—space isa limiting resource and overgrowth by other species can result inmortality and reduced fecundity (Hart and Marshall 2009). Wetherefore manipulated the level of interspecific competition thatour focal colonies experienced with three treatments: the absenceof interspecific competition, the presence of moderate interspe-cific competition, and in the presence of intense interspecificcompetition.

Materials and MethodsSTUDY SPECIES AND SITE

Watersipora subtorquata is an encrusting bryozoan and an abun-dant member of the fouling community on man-made structuresalong the coast of Australia. Watersipora filters planktonic foodfrom the water column, and as a colony, it grows outwards withthe leading edge of zooids (the feeding units that make up thecolony) growing, with a middle band of reproductive zooids, andan inner region of senescing and senesced zooids. It broods itslarvae for approximately 1 week, whereupon the nonfeeding lar-vae are released and spend only minutes to hours in the plank-ton before settling and metamorphosing (Marshall and Keough2004). Previous studies on this species have shown that offspringsize can strongly affect postmetamorphic performance, but the ef-fects of offspring size are highly variable over large spatial scales(Marshall and Keough 2004, 2008b, 2009). Our experiment wasdone at Redcliffe Marina, Brisbane, Australia in Austral Spring–Summer, 2008.

Reproductively mature Watersipora colonies were collectedfrom Redcliffe, transported to the laboratory, and induced tospawn after a being kept in constant darkness in lightproof aquariafor 72 h. We collected larvae from at least 45 colonies andtook care to collect larvae of a range of sizes from each colonythough we did not track which larva came from which colony. Wecollected the spawned larvae, settled them on pieces of acetatesheeting in a drop of seawater, and then allowed the settlers to

EVOLUTION FEBRUARY 2013 3 2 9

D. J. MARSHALL AND K. MONRO

metamorphose for 48 h in a constant temperature cabinet at 23"C.The metamorphosed settlers, known as ancestrulae, were thenmeasured using standard techniques under a dissecting micro-scope (Marshall and Keough 2008b), and the acetate sheets (withthe settlers attached) were then glued to PVC settlement plates(100 # 100 # 8 mm). The range of offspring sizes used in thisstudy reflected the natural variation in offspring size displayed bythis species in the field (Marshall and Keough 2008b).

TRAIT AND FITNESS MEASUREMENTS

We selected two traits along with offspring size that arethought to be important determinants of performance in modu-lar organisms—senescence rate and module size. We chose thesetraits in particular as we hypothesized that each could influenceselection on offspring size.

Offspring sizeWe used size of settlers approximately 12 h after settlement as ourmeasure of offspring size. Settler size is highly correlated withlarval size in Watersipora, and the correlation is independent ofthe length of the larval period (Marshall and Keough 2004), suchthat offspring size can be reliably inferred from measurements ofnewly settled juveniles.

Zooid sizeZooid size is an important morphological trait in bryozoans as itcan affect a range of life-history parameters in modular organisms(McKinney and Jackson 1991). For example, larger modules mayacquire more resources but take more resources to produce in thefirst place. After 4 weeks in the field, we returned the colonies tothe laboratory and measured the size of zooids within the colonyby photographing each colony under a dissecting microscope. Wemeasured zooid size at one time only because collecting thesedata is extremely time consuming (every zooid was measured togenerate a mean zooid size for the colony), and we wanted tominimize the time that the colonies spent in the laboratory. Pilotstudies suggested that zooid size remains relatively constant overtime in Watersipora.

SenescenceSenescence, where portions of the colony become inactive or un-dergo apoptosis, is common in many bryozoans, but the adaptivesignificance of senescence in this group generally remains unclear(McKinney and Jackson 1991; Orive 1995). In Watersipora, asthe colony ages, the central (i.e., oldest) portions of the colonysenesce, and the senescing region slowly spreads from the center(Hart and Keough 2009). The rate at which senescence occursvaries among colonies and may be related to how quickly re-sources are transferred from senescing zooids to reproduction,given that the zooids immediately adjacent to the senescing re-

gion tend to be those that are reproductive at that time (Hartand Keough 2009). Thus in Watersipora, senescence has beendescribed as an adaptive strategy by which the least productivezooids in the colony (i.e., those in the center of the colony thataccess water that has been depleted of planktonic food) senesceand free up resources that are used for reproduction elsewhere(Hart and Keough 2009). Zooids that have senesced are easilydistinguished from zooids that remain alive, with senesced re-gions clearly visible as black regions on the colony. We calculatedthe proportion of the colony that had senesced at the end of theexperimental period as our measure of senescence.

FitnessWe estimated reproduction weekly and defined our measure ofindividual fitness as the total number of offspring produced byeach colony over the entire experimental period. We recognizethat this estimate of fitness probably does not encompass theentire reproductive output of the colonies but we are confidentthat this estimate is a good and most practical predictor of truefitness. Importantly, estimates of fecundity across the study periodwere correlated positively (correlation coefficients among weeksranged between 0.484 and 0.74), suggesting that there was notrade-off between early and late reproduction. Zooids that arebrooding larvae are also clearly distinguishable in Watersiporaand reproduction can be estimated based on digital photographsof the colony. Watersipora colonies brood their offspring for lessthan a week in Brisbane, and brooded larvae that are identifiablein 1 week have been released from the colony by the subsequentweek.

COMPETITION TREATMENTS AND FIELD

DEPLOYMENT

New settlers were numbered and randomly assigned to one ofthree treatments: competition-free, intermediate competition, andhigh competition. Settlers assigned to the competition-free en-vironment were attached to blank settlement plates that werecleaned weekly throughout the experiment to remove any organ-isms that had naturally colonized the plates in the field. Settlersassigned to the intermediate competition environment were at-tached to blank settlement plates that were allowed to accumulatesettlement from other organisms (including tubeworms, ascidians,bryozoans, and sponges) throughout the experimental period. Set-tlers assigned to the high competition environment were attachedto settlement plates on which a fouling community had been al-lowed to develop by placing the plates into the field for 3 weeksprior to the initiation of the experiment. Previous studies show thatthese treatment conditions reflect the range of natural conditionsand are known to affect performance in this species (Marshall andKeough 2009). In this study, our treatments resulted in very differ-ent growing conditions for our focal colonies: individuals in the

3 3 0 EVOLUTION FEBRUARY 2013


competition-free environment were the largest and most fecund atthe end of the experiment, and individuals in the high competitionenvironment were the smallest and the least fecund (see Resultsand Figures S1 and S2). Note that because the treatments wereapplied at the scale of settlement plates, we had 60 independentreplicates of each treatment.

Once the settlers were attached onto the settlement plates,we assigned settlement plates haphazardly to PVC backing plates(400 # 400 # 12 mm), taking care to ensure that there was anequal number of plates from each treatment per backing plate.We then deployed the backing plates into the field by suspendingthe settlement plates facedown 1 m below the water surface (fora detailed description of the field deployment, see Marshall andKeough 2009). Each week, plates within each backing panel wererearranged haphazardly so as to preclude artefactual covariancebetween the values of our focal traits, local environmental condi-tions, and fitness (Rausher 1992). As our focal traits varied at thescale of plate, this was the appropriate scale of rearrangement.

We deployed a total of 180 settlers (60 in each treatment)across 30 backing plates. We monitored the survival, growth,senescence rate, and reproduction of our focal colonies weeklyfor 6 weeks in the field by photographing each colony with adigital camera in the field. On a few occasions (n = 9), we wereunable to adequately discern colony size, reproductive status, orsenescence rate from our photographs and as such, some measuresfor some colonies were omitted from the final analyses. None ofthe traits was correlated with another (correlation coefficients var-ied between 0.02 and 0.09), precluding the possibility of indirectselection being generated by one trait on another.

DATA ANALYSIS

We first determined whether there was an effect of the competitiveenvironment that colonies experienced on colony size and colonyreproductive output with repeated measures ANOVA. We thenexplored the effect of including the random factor "Backing plate"in our analyses, but there were no significant interactions betweenthe random factor and the fixed factor of interest and backing plateexplained little of the variation so Backing plate was excludedfrom the final model. There was very little mortality (<10%)across the experimental period, and the little mortality that didoccur was unrelated to our traits of interest, so we did not analyzemortality formally. We found strong effects of our treatments onthe performance of our focal colonies; we also found that somemean trait values were affected by our treatments (see Results andFigures S1 and S2).

We estimated selection on our traits of interest with a multi-ple regression approach (Lande and Arnold 1983), along with anextension of this approach as advocated by Phillips and Arnold(1989). The multiple regression approach is probably more fa-miliar and intuitively appealing to many readers, and it has the

advantage of generating standardized estimates of selection thatcan be later used in meta-analyses of selection to great effect(Kingsolver et al. 2001). Canonical analyses essentially extendthe traditional multiple regression approach and they have the ad-vantage of providing a more complete picture of selection on traitcombinations, and can be more sensitive than the standard mul-tiple regression approach (Blows and Brooks 2003). The outputsof canonical analyses are more difficult to interpret and are lessamenable to making comparisons across studies, however, and sowe present results from both the multiple regression analyses andthe canonical analyses.

We used partial F tests to test for differences in the strengthand form of selection experienced by individuals in the differentenvironments, following the model reduction approach describedin Chenoweth and Blows (2005).

ESTIMATING SELECTION GRADIENTS

Standardized linear (!) and nonlinear (") selection gradients wereestimated in each environment by regressing relative fitness (rel-ative to the population mean in each environment) on standard-ized phenotypic traits. Importantly, we used separate models toestimate linear and nonlinear gradients as appropriate and wecalculated the appropriate coefficient values for each type of se-lection (Stinchcombe et al. 2008). We then used randomizationtests of significance of all gradients as such tests avoid the prob-lems of parametric tests on non-normal distributions (Reynoldset al. 2010) when transformation of fitness is inadvisable. Theresults of these analyses are presented in Figure S3.

NONLINEAR SELECTION

It can be difficult to interpret the strength and significance ofnonlinear selection from examination of the " matrix alone, asnonlinear selection is frequently strongest on trait combinations(Phillips and Arnold 1989; Blows and Brooks 2003). Therefore,to assess multivariate selection on traits, we performed a canon-ical rotation of the " matrix. Canonical analysis determines thenormalized eigenvectors (mi), which contain the loadings of theoriginal traits on canonical axes, and their associated eigenval-ues (#i), which describe the form of nonlinear selection (Phillipsand Arnold 1989; Blows and Brooks 2003). Positive eigenval-ues indicate concave selection along eigenvectors (which mayindicate disruptive selection), and negative eigenvalues indicateconvex selection (which may indicate stabilizing selection). Thesignificance of eigenvalues was tested using the double linearregression method of Bisgaard and Ankenman (1996), with P-values calculated from permutation tests (Reynolds et al. 2010).We used thin-plate splines (with a smoothing parameter that min-imized the generalized cross validation score) to visualize themajor axes of the fitness surface identified by canonical analy-ses (Blows and Brooks 2003). As only significant eigenvectors



were visualized, the fitness surface was plotted on a single axisfor the competition-free environment, and on two axes for theintermediate- and high-competition environments.

PATH ANALYSES

It is important to note that the classic regression approach tomeasuring selection assumes that all focal traits affect fitness di-rectly. Thus, we also used path analyses of selection (see Scheineret al. 2000) to explore a different causal structure among our focaltraits—namely, that offspring size and senescence affect fitnessdirectly, but that offspring size and zooid size also affect fitness in-directly via their impacts on senescence. We restricted these anal-yses to linear selection only, given the complexity of modelingindirect nonlinear effects within this framework. Because com-parisons of model fit in each competitive environment found littlesupport for path models over regression models (based on differ-ences in model AIC scores, converted to P-values to aid interpre-tation; Anderson 2008), we present these results in Figure S3 only.

ResultsEFFECTS OF THE ENVIRONMENT ON PERFORMANCE

The environment that colonies experienced had a strong effecton both the growth rates and the total reproductive output of thecolonies (Effects on growth: Treatment: F2,160 = 90.2, P < 0.001;Treatment # Time: F12,960 = 78.15, P < 0.001; Figure S1; Ef-fect on reproduction: F2,160 = 59.67, P < 0.001; Figure S2A).Colonies that experienced no competition grew and reproducedthe most while colonies that experienced competition from an es-tablished community grew and reproduced the least. There was asignificant difference in the amount of colony senescence amongthe different treatments: colonies experiencing high competitionhad a higher proportion of senesced zooids than colonies experi-encing intermediate or no competition (F2,131 = 3.73, P = 0.0265;Figure S2B). There was no significance difference in either thesize of zooids or initial offspring size among the treatments (Zooidsize: F2,131 = 0.3024, P = 0.739; Offspring size: F2,131 = 1.02,P = 0.361).

DIFFERENCES IN SELECTION AMONG

ENVIRONMENTS

There was marginal evidence for linear selection varying amongenvironments (F3,157 = 2.25, P = 0.084). Explorations with AN-COVA suggested that linear selection on offspring size differedamong environments (F2,155 = 3.23, P = 0.042), but no differ-ences in selection on the other two traits across environments.We found strong evidence that nonlinear selection varied amongenvironments (F6,139 = 2.52, P = 0.023). Further analyses sug-gested that quadratic selection differed among environments and

More senescence Smaller zooid size

Less senescence Larger zooid size

-2 -1 0 1 2 3 4

M3

0.0

0.5

1.0

1.5

Rep

rodu

ctiv

e ou

tput

Figure 1. Fitness of Watersipora subtorquata colonies in acompetition-free environment showing multivariate selection onphenotypic trait combinations. Individuals with intermediate val-ues of offspring size, zooid size, and senescence rate had the high-est fecundity. Line represents best quadratic approximations ofselection based on multivariate canonical analyses.

marginal evidence for differences in correlational selection amongenvironments (quadratic: F3,145 = 3.06, P = 0.030; correlational:F3,145 = 2.35, P = 0.074). Exploration with ANCOVA suggestedthat this difference in quadratic selection among environmentswas driven entirely by differences in quadratic selection on off-spring size among environments (F2,145 = 3.93, P = 0.022).Inspection of the data and the ! and " matrices suggested thatthere was very little selection on our traits of interest in themost benign environment, but there was strong selection in theother two environments (Table 1). In the intermediate environ-ment, there was strong stabilizing selection while in the harsh-est environment, positive directional selection acted on offspringsize.

Our canonical analyses revealed a complex picture of se-lection on our traits of interest. We detected significant convexselection acting on one eigenvector (m3) of trait combinations inthe competition-free environment (Table 2). In this environment,senescence loads positively, offspring size loads very little, andzooid size loads negatively, and the visualization of the fitnesssurface along this canonical axis suggests intermediate values ofzooid size and senescence yield the highest fitness (Figure 1).In the intermediate-competition environment, strong convexselection acted along two eigenvectors (Figure 2): on one



Table 1. Standardized gradients of directional selection (!) and nonlinear selection (") on offspring size, senescence rate, and zooidsize across competition environments in Watersipora subtorquata. Quadratic and correlational selection gradients are the diagonal andoff-diagonal elements of " respectively, errors shown in parentheses below each estimate (P < 0.05 shown in bold).

"

! Offspring size Senescence rate Zooid size

Competition freeOffspring size 0.093 0.001

(0.074) (0.058)Senescence rate $0.021 $0.050 $0.078

(0.079) (0.093) (0.065)Zooid size $0.036 $0.090 0.086 $0.053

(0.080) (0.016) (0.0880) (0.069)Intermediate competitionOffspring size 0.038 !0.205

(0.133) (0.093)Senescence rate $0.003 0.015 $0.118

(0.131) (0.148) (0.089)Zooid size 0.061 0.016 $0.079 $0.129

(0.134) (0.177) (0.142) (0.091)High competitionOffspring size 0.422 0.140

(0.157) (0.129)Senescence rate $0.162 $0.388 0.0107

(0.151) (0.209) (0.101)Zooid size $0.150 $0.111 $0.180 $0.064

(0.156) (0.167) (0.197) (0.113)

Table 2. Matrix of eigenvectors (mi) and their associated eigenvalues (#) from the canonical analysis of " for total reproductive output,under different levels of competition (P < 0.05 shown in bold).

Original traitsSelection# (Std. dev.) Offspring size Senescence rate Zooid size

Competition freem1 0.042 (0.097) 0.782 $0.345 $0.518m2 $0.061 (0.123) 0.617 0.542 0.569m3 !0.111 (0.098) 0.084 !0.765 0.637

Intermediate competitionm1 $0.084 (0.083) 0.002 0.753 $0.658m2 !0.161 (0.085) 0.234 0.639 0.732m3 !0.208 (0.066) 0.972 !0.156 !0.175

High competitionm1 0.280 (0.147) 0.806 !0.591 0.031m2 0.017 (0.146) $0.372 $0.466 0.802m3 !0.211 (0.111) 0.460 0.659 0.596

eigenvector (m2), offspring size loads lightly while both senes-cence and zooid size load heavily; on the other eigenvector, off-spring size loads very heavily while other traits load very little(Table 2; Figure 2).

In the high-competition environment, canonical analysisshowed both significant concave and convex selection acting

along two eigenvectors of trait combinations (m1 and m3 re-spectively, Table 2). On the first eigenvector, offspring size loadspositively and strongly while senescence rate loads negativelyand zooid size loads very little. The fitness surface suggests thatlarger offspring sizes and low senescence rates yield the high-est fitness returns (Figure 3). On the third eigenvector, all three



Larger zooids More Senescence

Smaller zooids Less senescence

Larger o!spring size Less senescence Smaller zooids

Smaller o!spring sizeMore senescenceLarger zooids

Figure 2. Fitness of Watersipora subtorquata colonies in anintermediate-competition environments showing multivariate se-lection on phenotypic trait combinations. Individuals with in-termediate values of offspring size, zooid size, and senescencerate had the highest fecundity. Surface represents best quadraticapproximations of selection based on multivariate canonicalanalyses.

Larger o!spring size Less senescence

Larger o!spring size More senescence Larger zooids

Smaller o!spring size More senescence

Smaller o!spring size Less senescence Smaller zooids

Figure 3. Fitness of Watersipora subtorquata colonies in a high-competition environments showing multivariate selection on phe-notypic trait combinations. Individuals with the greatest valuesof offspring size, zooid size, and lowest values for senescencerate had the highest fecundity. Surface represents best quadraticapproximations of selection based on multivariate canonicalanalyses.

traits load positively and approximately equally and the fitnesssurface suggests that an intermediate phenotype (with regards tothe linear combination of all three traits) is favored (Figure 3).Overall, in the high-competition environment, the phenotypewith the highest fitness returns combined the highest values form1 and intermediate values of m2 (but with the latter of lesserimportance).

DiscussionThe strength and form of selection on offspring size varied amongthe different environments in the field: nonlinear selection onoffspring size was strongest in the harsher environments and weakto nonexistent in the most benign environment. Our results alsosuggest that failing to consider nonlinear selection could lead to anincomplete understanding of the relationship between offspringsize and fitness.

CAUSES AND CONSEQUENCES OF ENVIRONMENTAL

VARIABILITY IN SELECTION ON OFFSPRING SIZE

We found no nonlinear selection on offspring size in the mostbenign environment. Our study joins a growing list showing thatunder benign conditions, selection on offspring size is weakerthan under harsher conditions (Mousseau and Fox 1998), but itis important to note that this pattern is not universal (Allen et al.2008). In nature, it is probably rare for Watersipora to encounterentirely competition-free environments for sustained periods oftime, so we suspect that selection in this environment only playsa minor part in shaping offspring size variation. Nevertheless, ourfinding does illustrate that excluding the effects of interspecificcompetition (as is typical of most laboratory-based studies of off-spring size) can result in the dramatic underestimation of selectionon offspring size.

In the two harsher environments, nonlinear selection wasstrong but context dependent. We found evidence for stabilizingselection on offspring size in the intermediate environment andevidence for nonlinear selection for increased offspring size inthe harshest environment. In the harshest environment used here,competition for food is intense and so increased size (and pre-sumably, energy reserves) may assist offspring in coping withthis competition. The detection of stabilizing selection on off-spring size in the intermediate environment is harder to explain;we expected similar benefits of increased offspring size in theintermediate environment if food competition alone drove oureffects. It could be that the two surrounding communities (theintermediate and the harshest environment) were composed ofdifferent species and this difference led to a difference in se-lection on offspring size. Regardless, this effect warrants furtherexploration as it was not anticipated.

The detection of stabilizing selection on offspring size in twoenvironments has some interesting implications for the way weview offspring size effects. Traditionally, it has been temptingto view the performance of larger offspring as invariably eithergreater than, or at least equal to, the performance of smaller off-spring (Marshall et al. 2006; Jorgensen et al. 2011). However,our detection of stabilizing selection suggests that, in this species,very large offspring will have lower performance than offspringof intermediate size under some conditions. This finding is in



keeping with some studies that have similarly found that largeroffspring do not always perform better and sometimes even doworse than smaller offspring (Kaplan 1992; Einum and Fleming2000; Dibattista et al. 2007; Jacobs and Sherrard 2010). As notedin the Introduction, the form of selection on offspring size willhave a critical impact on the type of bet-hedging that mothersmay employ to cope with unpredictable environments (Crean andMarshall 2009). Our finding suggests that, in Watersipora, se-lection may favor mothers that employ a diversified bet-hedgingstrategy when the offspring environment is unpredictable. Im-portantly, stabilizing selection on offspring size does not meanthat mothers should produce the particular offspring size thatselection most favors (Marshall and Uller 2007). Instead, the op-timal offspring size that mothers should produce will probablybe product of both the offspring size–performance relationshipand the trade-off between offspring size and number (Smith andFretwell 1974; Wolf and Wade 2001; Johnson et al. 2010). To es-timate selection on mothers, rather than offspring, our estimatesof selection on offspring size would have to be combined withestimates of a size-number trade-off that mothers face as recentlydone by Johnson et al. (2010), but such data are lacking for ourspecies.

Our finding that selection strongly favored larger offspringin the harshest environment is in keeping with offspring pro-visioning patterns in this species. Previous work has shown thatWatersipora colonies grown in the presence of a competitive com-munity produce larger offspring than colonies grown in the ab-sence of competitors (Marshall and Keough 2009). Our resultshere would suggest that this plasticity in offspring provisioningrepresents an "anticipatory maternal effect" (sensu Marshall andUller 2007), whereby colonies respond to environmental cuesin their own habitat and provision their offspring accordingly.It should be noted however that offspring from a single colonyprobably encounter both habitat types in nature because of thescale of dispersal; hence mothers face the daunting challenge ofprovisioning their offspring on the chance that they will encountereither environment.

A number of recent reviews have highlighted the importanceof variability in selection for determining the tempo and trajectoryof evolution (Orr 2009; Bell 2010; Calsbeek 2012). Our resultsshow that selection is highly variable over small spatial scales.Furthermore, our environments span the range of successionalstages in the surrounding community that Watersipora will ex-perience, sometimes within a single lifetime (recall that to createour harshest environment, we allowed a community to develop foronly 3 weeks prior to starting the experiment and W. subtorquatacolonies can live for twice this time). As such, our results suggestthat environmentally driven variation in selection will reduce itsefficacy and may maintain variance in offspring size.

CORRELATION SELECTION ON OFFSPRING SIZE

Offspring size effects have long been recognized as highly contextdependent: the local environmental conditions strongly affect theoffspring size–performance relationship (Mousseau and Dingle1991; Bernardo 1996). Importantly, our findings suggest that off-spring size is also context dependent with regards to other traitvalues—the fitness return of a particular offspring size depends onvalues of other traits. For example, canonical analysis suggestedthat the fitness benefits of increased offspring size are increased inindividuals with lower senescence rates when they experience in-tense competition. Such correlational selection should, over time,result in increased negative covariance between offspring size andsenescence rates (Blows and Brooks 2003). More generally, thereis some evidence for covariance between offspring size and othertraits at both the population level (Johnston and Leggett 2002;Bashey 2008; Walsh and Reznick 2010), and within populations(Martin and Pfennig 2010). Coevolution between offspring sizeand other traits has also been implicated in a range of other systems(Krug and Zimmer 2000; Walsh et al. 2006; Benton et al. 2008;Miles and Wayne 2009) suggesting offspring size often interactswith other traits to affect fitness. It would therefore be interestingto examine whether correlational selection operates on offspringsize and other traits in organisms where covariance between thesetraits has already been detected. For example, Pfennig and Martin(2009) did not formally analyze correlational selection on off-spring size and "carnivory traits" in spadefooted toads (Pfennigand Martin 2009), but we suspect that correlational selection isacting in this species, given increases in offspring size alter selec-tion on other carnivory traits. We suggest that the prevalence ofcorrelational selection on offspring size has been underestimatedmore generally.

Biologists have long recognized that offspring size has per-vasive longitudinal influences throughout the life history (Lack1947). For example, some studies show that initial offspring sizecan affect subsequent size throughout the life history (Dias andMarshall 2010). The presence of correlational selection on off-spring size in combination with other traits, however, suggeststhat variation in maternal investment will also generate selectionon traits other than offspring size where there would otherwisebe none (Blows and Brooks 2003). Thus, the influence of initialmaternal investment on the phenotype of offspring may be morepervasive than has been recognized previously—affecting selec-tion on traits that are not directly correlated with offspring size.For example, variation in initial offspring provisioning could con-strain the range of values of other traits that yield a fitness benefitfor offspring. This maternal effect is more insinuating than thosetypically discussed, and warrants further exploration. On the otherhand, the presence of correlational selection also raises the pos-sibility that the range of offspring sizes that yield high fitness



returns is more constrained than has been recognized. If the fit-ness benefits of increased offspring size are highly dependent onthe values of other traits (as they are when correlational selectionoccurs), mothers may evolve to provision offspring according tothe values of other offspring traits.

We found evidence for correlational selection on senescencein conjunction with other traits, but this selection depended onthe local environment. In environments with less competition,colonies that senesced more had higher reproductive output thancolonies that senesced very little, whereas in the harshest envi-ronment, this pattern was reversed. While there has been muchdebate regarding the evolution of senescence in clonal organismsand marine invertebrates in particular (McKinney and Jackson1991; Chadwickfurman and Weissman 1995; Orive 1995; Bayerand Todd 1997; Gardner and Mangel 1997), there are few empir-ical studies of senescence, and as far as we are aware, no formalestimates of the selective value of senescence in a marine inverte-brate. It appears that some senescence can benefit colony fitness,but that the benefits of senescence are highly context dependent—both the context of other life-history traits (i.e., offspring size) andlocal environmental conditions (intensity of competition) can al-ter the fitness returns of senescence. An interesting next step willbe to determine whether genetic trade-offs between senescenceand other components of fitness exist in this species.

Overall, we found a complex web of selection acting on off-spring size that varied among environments. Our results suggestthat further examinations of variability in selection in the fieldare required as our results suggest that selection can vary overremarkably small spatial and temporal scales. We suspect thatnonlinear selection on offspring size and other offspring traits iscommon but without more studies, it remains impossible to knowits prevalence.

ACKNOWLEDGMENTSThe authors wish to thank the K. Donohue, C. Fox, and two anonymousreviewers for comments that greatly improved the manuscript. We thankS. Chenoweth for statistical advice. We thank Tane Sinclair-Taylor forassistance in the field and laboratory. DJM and KM were supported bygrants and fellowships from the Australian Research Council.

LITERATURE CITEDAllen, R. M., Y. M. Buckley, and D. J. Marshall. 2008. Offspring size plasticity

in response to intraspecific competition: an adaptive maternal effectacross life-history stages. Am. Nat. 171:225–237.

Anderson, D. R. 2008. Model based inference in the life sciences: a primer onevidence. Springer, New York.

Bagenal, T. B. 1969. Relationship between egg size and fry survival in browntrout Salmo trutta L. J. Fish. Biol. 1:349–353.

Bashey, F. 2008. Competition as a selective mechanism for larger offspringsize in guppies. Oikos 117:104–113.

Bayer, M. M. and C. D. Todd. 1997. Evidence for zooid senescence in themarine bryozoan Electra pilosa. Invertebr. Biol. 116:331–340.

Bell, G. 2010. Fluctuating selection: the perpetual renewal of adaptation invariable environments. Philos. Trans. R. Soc. Lond. B 365:87–97.

Benton, T. G., J. J. H. St Clair, and S. J. Plaistow. 2008. Maternal effectsmediated by maternal age: from life histories to population dynamics. J.Anim. Ecol. 77:1038–1046.

Bernardo, J. 1996. The particular maternal effect of propagule size, especiallyegg size: patterns models, quality of evidence and interpretations. Amer.Zool. 36:216–236.

Bisgaard, S. and B. Ankenman. 1996. Standard errors for the eigenvalues insecond-order response surface models. Technometrics 38:238–246.

Blows, M. W. and R. Brooks. 2003. Measuring nonlinear selection. Am. Nat.162:815–820.

Calsbeek, B. 2012. Exploring variation in fitness surfaces over time or space.Evolution 66:1126–1137.

Chadwickfurman, N. E. and I. L. Weissman. 1995. Life histories and senes-cence of Botryllus schlosseri (Chordata, Ascidiacea) in Monterey Bay.Biol. Bull. 189:36–41.

Chenoweth, S. F. and M. W. Blows. 2005. Contrasting mutual sexual selec-tion on homologous signal traits in Drosophila serrata. Am. Nat. 165:281–289.

Crean, A. J. and D. J. Marshall. 2009. Coping with environmental uncertainty:dynamic bet-hedging as a maternal effect. Philos. Trans. R. Soc. Lond.B 364:1087–1096.

Dias, G. M. and D. J. Marshall. 2010. Does the relationship between offspringsize and performance change across the life-history? Oikos 119:154–162.

Dibattista, J. D., K. A. Feldheim, S. H. Gruber, and A. P. Hendry. 2007. Whenbigger is not better: selection against large size, high condition and fastgrowth in juvenile lemon sharks. J. Evol. Biol. 20:201–212.

Einum, S. and I. A. Fleming. 2000. Highly fecund mothers sacrifice offspringsurvival to maximize fitness. Nature 405:565–567.

———. 2002. Does within-population variation in fish egg size reflect ma-ternal influences on optimal values. Am. Nat. 160:756–765.

———. 2004. Environmental unpredictability and offspring size: conservativeversus diversified bet-hedging. Evol. Ecol. Res. 6:443–455.

Fox, C. W. and M. E. Czesak. 2000. Evolutionary ecology of progeny size inarthropods. Annu. Rev. Entomol. 45:341–369.

Fox, C. W., M. S. Thakar, and T. A. Mosseau. 1997. Egg size plasticity in aseed beetle: an adaptive maternal effect. Am. Nat. 149:149–163.

Gardner, S. N. and M. Mangel. 1997. Can a clonal organism escape senes-cence? Am. Nat. 150:462–490.

Hart, S. P. and M. J. Keough. 2009. Does size predict demographic fate?Modular demography and constraints on growth determine response todecreases in size. Ecology 90:1670–1678.

Hart, S. P. and D. J. Marshall. 2009. Spatial arrangement affects popula-tion dynamics and competition independent of community composition.Ecology 90:1485–1491.

Jacobs, M. W. and K. M. Sherrard. 2010. Bigger is not always better: offspringsize does not predict growth or survival for seven ascidian species.Ecology 91:3598–3608.

Johnson, D. W., M. R. Christie, and J. Moye. 2010. Quantifying evolutionarypotential of marine fish larvae: heritability, selection, and evolutionaryconstraints. Evolution 64:2614–2628.

Johnston, T. A. and W. C. Leggett. 2002. Maternal and environmentalgradients in the egg size of an iteroparous fish. Ecology 83:1777–1791.

Jorgensen, C., S. K. Auer, and D. N. Reznick. 2011. A model for optimaloffspring size in fish, including live-bearing and parental effects. Am.Nat. 177:E119–E135.

Kaplan, R. H. 1992. Greater maternal investment can decrease offspring suri-vival in the frog Bombina orientalis. Ecology 73:280–288.



Kingsolver, J. G., H. E. Hoekstra, J. M. Hoekstra, D. Berrigan, S. N. Vignieri,C. E. Hill, A. Hoang, P. Gibert, and P. Beerli. 2001. The strength ofphenotypic selection in natural populations. Am. Nat. 157:245–261.

Krist, M. 2011. Egg size and offspring quality: a meta-analysis in birds. Biol.Rev. 86:692–716.

Krug, P. J. and R. K. Zimmer. 2000. Developmental dimorphism and ex-pression of chemosensory-mediated behavior: habitat selection by aspecialist marine herbivore. J. Exp. Biol. 203:1741–1754.

Lack, D. 1947. The significance of clutch size. Ibis 89:302–352.Lande, R. and S. J. Arnold. 1983. The measurement of selection on correlated

characters. Evolution 37:1210–1226.Marshall, D. J., R. Bonduriansky, and L. F. Bussiere. 2008. Offspring size

variation within broods as a bet-hedging strategy in unpredictable envi-ronments. Ecology 89:2506–2517.

Marshall, D. J. and M. J. Keough. 2004. Variable effects of larval size on post-metamorphic performance in the field. Mar. Ecol. Prog. Ser. 279:73–80.

———. 2008a. The evolutionary ecology of offspring size in marine inverte-brates. Adv. Mar. Biol. 53:1–60.

———. 2008b. The relationship between offspring size and performance inthe sea. Am. Nat. 171:214–224.

———. 2009. Does interspecific competition affect offspring provisioning?Ecology 90:487–495.

Marshall, D. J. and T. Uller. 2007. When is a maternal effect adaptive? Oikos116:1957–1963.

Marshall, D. J., C. N. Cook, and R. B. Emlet. 2006. Offspring size effects me-diate competitive interactions in a colonial marine invertebrate. Ecology87:214–225.

Martin, R. A. and D. W. Pfennig. 2010. Maternal investment influences ex-pression of resource polymorphism in amphibians: implications for theevolution of novel resource-use phenotypes. PLoS One 5:7.

McKinney, F. K. and J. B. C. Jackson. 1991. Bryozoan evolution. Universityof Chicago Press, Chicago.

Miles, C. M. and M. L. Wayne. 2009. Life history trade-offs and responseto selection on egg size in the polychaete worm Hydroides elegans.Genetica 135:289–298.

Mousseau, T. A. and H. Dingle. 1991. Maternal effects in insect life histories.Annu. Rev. Entomol. 36:511–534.

Mousseau, T. A. and C. W. Fox. 1998. The adaptive significance of maternaleffects. Trends Ecol. Evol. 13:403–407.

Orive, M. E. 1995. Senescence in organisms with clonal reproduction andcomplex life histories. Am. Nat. 145:90–108.

Orr, H. A. 2009. Fitness and its role in evolutionary genetics. Nat. Rev. Genet.10:531–539.

Pfennig, D. W. and R. A. Martin. 2009. A maternal effect mediates rapidpopulation divergence and character displacement in spadefoot toads.Evolution 63:898–909.

Phillips, P. C. and S. J. Arnold. 1989. Visualizing multivariate selection.Evolution 43:1209–1222.

Rausher, M. D. 1992. The measurement of selection on quantitative traits:biases due to environmental covariances between traits and fitness. Evo-lution 46:616–626.

Reynolds, R. J., D. K. Childers, and N. M. Pajewski. 2010. The distributionand hypothesis testing of eigenvalues from the canonical analysis ofthe gamma matrix of quadratic and correlational selection gradients.Evolution 64:1076–1085.

Scheiner, S., R. J. Mitchell, and H. S. Callahan. 2000. Using path analysis tomeasure natural selection. J. Evol. Biol. 13:423–433.

Smith, C. C. and S. D. Fretwell. 1974. The optimal balance between size andnumber of offspring. Am. Nat. 108:499–506.

Stanton, M. L. 1984. Seed variation in wild radish: effect of seed size oncomponents of seedling and adult fitness. Ecology 65:1105–1112.

Stinchcombe, J. R., A. F. Agrawal, P. A. Hohenlohe, S. J. Arnold, and M. W.Blows. 2008. Estimating nonlinear selection gradients using quadraticregression coefficients: double or nothing? Evolution 62:2435–2440.

Walsh, M. R. and D. N. Reznick. 2010. Influence of the indirect effects ofguppies on life history evolution in Rivulus hartii. Evolution 64:1583–1593.

Walsh, M. R., S. B. Munch, S. Chiba, and D. O. Conover. 2006. Maladaptivechanges in multiple traits caused by fishing: impediments to populationrecovery. Ecol. Lett. 9:142–148.

Williams, T. D. 1994. Intraspecific variation in egg size and egg compositionin birds: effects on offspring fitness. Biol. Rev. 68:38–59.

Wolf, J. B. and M. J. Wade. 2001. On the assignment of fitness to parentsand offspring: whose fitness is it and when does it matter? J. Evol. Biol.14:347–356.

Associate Editor: K. Donohue

Supporting InformationThe following supporting information is available for this article:

APPENDIXFigure S1. Effect of competition on colony size in Watersipora subtorquata in the field.Figure S2. Effect of competition on (A) total reproductive output; and (B) proportion of colony that had senesced in Watersiporasubtorquata.Figure S3. Comparisons of regression models versus path models of selection in different competitive environments.

Supporting Information may be found in the online version of this article.

Please note: Wiley-Blackwell is not responsible for the content or functionality of any supporting information supplied by theauthors. Any queries (other than missing material) should be directed to the corresponding author for the article.


0"

500"

1000"

1500"

Colony"area"

(mm

2 )"

1" 2" 3" 4" 5" 6" 7"

Week"

High"

Intermediate"

Low"

Figure"2)"Effect"of"compeGGon"on"colony"size"in"Watersipora*subtorquata*in"the"field."

Total"Rep

rodu

cGon

"

100"

300"

500"

Prop

orGo

n"sene

sced

"

High"Low" Intermediate"

0.10"

0.15"

0.05"

Figure"2)"Effect"of"compeGGon"on"a)"total"reproducGve"output;"and""b)"proporGon"of"colony"that"had"senesced"in"Watersipora*subtorquata"

a)"

b)"

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!Appendix!2!Figure!1.!!Comparisons!of!regression!models!vs#path!models!of!selection!in!different!competitive!environments.!In!regression!models,!paths!(single>arrow!lines)!show!the!direct!effects!(black!lines!and!coefficients)!of!offspring!size,!colony!senescence!and!colony!zooid!size!on!fitness,!given!trait!variances!(V)!and!covariances!(CV,!double>arrow!lines).!In!path!models,!paths!show!the!direct!effects!of!offspring!size!and!senescence!on!fitness,!plus!the!indirect!effects!(grey!lines!and!coefficients)!of!offspring!size!and!zooid!size!on!fitness,!acting!via!their!direct!effects!on!senescence!(with!paths!now!replacing!covariances!between!these!traits).!!Significant!coefficients!are!in!bold.!For!all!environments,!differences!in!the!fit!of!alternative!models!(measured!by!changes!in!their!AIC!scores,!converted!to!P>values!to!aid!interpretation)!are!minimal,!indicating!similar!support!for!each!model!type.!

INTERSPECIFIC COMPETITION ALTERS NONLINEAR SELECTION … · communities, interspecific competition...

Documents

Transcript of INTERSPECIFIC COMPETITION ALTERS NONLINEAR SELECTION … · communities, interspecific competition...